Paper/patches/server/0018-Rewrite-chunk-system.patch

18176 lines
822 KiB
Diff
Raw Normal View History

Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Spottedleaf <Spottedleaf@users.noreply.github.com>
Date: Thu, 11 Mar 2021 02:32:30 -0800
Subject: [PATCH] Rewrite chunk system
== AT ==
public net.minecraft.server.level.ChunkMap setViewDistance(I)V
public net.minecraft.server.level.ChunkHolder pos
public net.minecraft.server.level.ChunkMap overworldDataStorage
public-f net.minecraft.world.level.chunk.storage.RegionFileStorage
public net.minecraft.server.level.ChunkMap getPoiManager()Lnet/minecraft/world/entity/ai/village/poi/PoiManager;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
diff --git a/src/main/java/ca/spottedleaf/starlight/common/light/StarLightInterface.java b/src/main/java/ca/spottedleaf/starlight/common/light/StarLightInterface.java
index 9a5fa60cb8156fe254a123e237d957ccb82f7195..0f7d36933e34e1d1b9dd27d8b0c35ff883818526 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/ca/spottedleaf/starlight/common/light/StarLightInterface.java
+++ b/src/main/java/ca/spottedleaf/starlight/common/light/StarLightInterface.java
@@ -41,14 +41,14 @@ public final class StarLightInterface {
protected final ArrayDeque<SkyStarLightEngine> cachedSkyPropagators;
protected final ArrayDeque<BlockStarLightEngine> cachedBlockPropagators;
- protected final LightQueue lightQueue = new LightQueue(this);
+ public final io.papermc.paper.chunk.system.light.LightQueue lightQueue; // Paper - replace light queue
protected final LayerLightEventListener skyReader;
protected final LayerLightEventListener blockReader;
protected final boolean isClientSide;
- protected final int minSection;
- protected final int maxSection;
+ public final int minSection; // Paper - public
+ public final int maxSection; // Paper - public
protected final int minLightSection;
protected final int maxLightSection;
@@ -182,6 +182,7 @@ public final class StarLightInterface {
StarLightInterface.this.sectionChange(pos, notReady);
}
};
+ this.lightQueue = new io.papermc.paper.chunk.system.light.LightQueue(this); // Paper - replace light queue
}
protected int getSkyLightValue(final BlockPos blockPos, final ChunkAccess chunk) {
@@ -325,7 +326,7 @@ public final class StarLightInterface {
return this.lightAccess;
}
- protected final SkyStarLightEngine getSkyLightEngine() {
+ public final SkyStarLightEngine getSkyLightEngine() { // Paper - public
if (this.cachedSkyPropagators == null) {
return null;
}
@@ -340,7 +341,7 @@ public final class StarLightInterface {
return ret;
}
- protected final void releaseSkyLightEngine(final SkyStarLightEngine engine) {
+ public final void releaseSkyLightEngine(final SkyStarLightEngine engine) { // Paper - public
if (this.cachedSkyPropagators == null) {
return;
}
@@ -349,7 +350,7 @@ public final class StarLightInterface {
}
}
- protected final BlockStarLightEngine getBlockLightEngine() {
+ public final BlockStarLightEngine getBlockLightEngine() { // Paper - public
if (this.cachedBlockPropagators == null) {
return null;
}
@@ -364,7 +365,7 @@ public final class StarLightInterface {
return ret;
}
- protected final void releaseBlockLightEngine(final BlockStarLightEngine engine) {
+ public final void releaseBlockLightEngine(final BlockStarLightEngine engine) { // Paper - public
if (this.cachedBlockPropagators == null) {
return;
}
@@ -511,57 +512,15 @@ public final class StarLightInterface {
}
public void scheduleChunkLight(final ChunkPos pos, final Runnable run) {
- this.lightQueue.queueChunkLighting(pos, run);
+ throw new UnsupportedOperationException("No longer implemented, use the new lightQueue field to queue tasks"); // Paper - replace light queue
}
public void removeChunkTasks(final ChunkPos pos) {
- this.lightQueue.removeChunk(pos);
+ throw new UnsupportedOperationException("No longer implemented, use the new lightQueue field to queue tasks"); // Paper - replace light queue
}
public void propagateChanges() {
- if (this.lightQueue.isEmpty()) {
- return;
- }
-
- final SkyStarLightEngine skyEngine = this.getSkyLightEngine();
- final BlockStarLightEngine blockEngine = this.getBlockLightEngine();
-
- try {
- LightQueue.ChunkTasks task;
- while ((task = this.lightQueue.removeFirstTask()) != null) {
- if (task.lightTasks != null) {
- for (final Runnable run : task.lightTasks) {
- run.run();
- }
- }
-
- final long coordinate = task.chunkCoordinate;
- final int chunkX = CoordinateUtils.getChunkX(coordinate);
- final int chunkZ = CoordinateUtils.getChunkZ(coordinate);
-
- final Set<BlockPos> positions = task.changedPositions;
- final Boolean[] sectionChanges = task.changedSectionSet;
-
- if (skyEngine != null && (!positions.isEmpty() || sectionChanges != null)) {
- skyEngine.blocksChangedInChunk(this.lightAccess, chunkX, chunkZ, positions, sectionChanges);
- }
- if (blockEngine != null && (!positions.isEmpty() || sectionChanges != null)) {
- blockEngine.blocksChangedInChunk(this.lightAccess, chunkX, chunkZ, positions, sectionChanges);
- }
-
- if (skyEngine != null && task.queuedEdgeChecksSky != null) {
- skyEngine.checkChunkEdges(this.lightAccess, chunkX, chunkZ, task.queuedEdgeChecksSky);
- }
- if (blockEngine != null && task.queuedEdgeChecksBlock != null) {
- blockEngine.checkChunkEdges(this.lightAccess, chunkX, chunkZ, task.queuedEdgeChecksBlock);
- }
-
- task.onComplete.complete(null);
- }
- } finally {
- this.releaseSkyLightEngine(skyEngine);
- this.releaseBlockLightEngine(blockEngine);
- }
+ throw new UnsupportedOperationException("No longer implemented, task draining is now performed by the light thread"); // Paper - replace light queue
}
protected static final class LightQueue {
diff --git a/src/main/java/co/aikar/timings/TimingsExport.java b/src/main/java/co/aikar/timings/TimingsExport.java
index 38f01952153348d937e326da0ec102cd9b0f80af..43380d5e3a40b64bebdf3c0e7c48eca8998c8ac0 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/co/aikar/timings/TimingsExport.java
+++ b/src/main/java/co/aikar/timings/TimingsExport.java
@@ -163,7 +163,11 @@ public class TimingsExport extends Thread {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
pair("gamerules", toObjectMapper(world.getWorld().getGameRules(), rule -> {
return pair(rule, world.getWorld().getGameRuleValue(rule));
})),
- pair("ticking-distance", world.getChunkSource().chunkMap.getEffectiveViewDistance())
+ // Paper start - replace chunk loader system
+ pair("ticking-distance", world.getChunkSource().chunkMap.playerChunkManager.getTargetTickViewDistance()),
+ pair("no-ticking-distance", world.getChunkSource().chunkMap.playerChunkManager.getTargetNoTickViewDistance()),
+ pair("sending-distance", world.getChunkSource().chunkMap.playerChunkManager.getTargetSendDistance())
+ // Paper end - replace chunk loader system
));
}));
diff --git a/src/main/java/co/aikar/timings/WorldTimingsHandler.java b/src/main/java/co/aikar/timings/WorldTimingsHandler.java
index 2f0d9b953802dee821cfde82d22b0567cce8ee91..22687667ec69a954261e55e59261286ac1b8b8cd 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/co/aikar/timings/WorldTimingsHandler.java
+++ b/src/main/java/co/aikar/timings/WorldTimingsHandler.java
@@ -59,6 +59,16 @@ public class WorldTimingsHandler {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public final Timing miscMobSpawning;
+ public final Timing poiUnload;
+ public final Timing chunkUnload;
+ public final Timing poiSaveDataSerialization;
+ public final Timing chunkSave;
+ public final Timing chunkSaveDataSerialization;
+ public final Timing chunkSaveIOWait;
+ public final Timing chunkUnloadPrepareSave;
+ public final Timing chunkUnloadPOISerialization;
+ public final Timing chunkUnloadDataSave;
+
public WorldTimingsHandler(Level server) {
String name = ((PrimaryLevelData) server.getLevelData()).getLevelName() + " - ";
@@ -112,6 +122,16 @@ public class WorldTimingsHandler {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
miscMobSpawning = Timings.ofSafe(name + "Mob spawning - Misc");
+
+ poiUnload = Timings.ofSafe(name + "Chunk unload - POI");
+ chunkUnload = Timings.ofSafe(name + "Chunk unload - Chunk");
+ poiSaveDataSerialization = Timings.ofSafe(name + "Chunk save - POI Data serialization");
+ chunkSave = Timings.ofSafe(name + "Chunk save - Chunk");
+ chunkSaveDataSerialization = Timings.ofSafe(name + "Chunk save - Chunk Data serialization");
+ chunkSaveIOWait = Timings.ofSafe(name + "Chunk save - Chunk IO Wait");
+ chunkUnloadPrepareSave = Timings.ofSafe(name + "Chunk unload - Async Save Prepare");
+ chunkUnloadPOISerialization = Timings.ofSafe(name + "Chunk unload - POI Data Serialization");
+ chunkUnloadDataSave = Timings.ofSafe(name + "Chunk unload - Data Serialization");
}
public static Timing getTickList(ServerLevel worldserver, String timingsType) {
diff --git a/src/main/java/com/destroystokyo/paper/io/IOUtil.java b/src/main/java/com/destroystokyo/paper/io/IOUtil.java
new file mode 100644
index 0000000000000000000000000000000000000000..e064f96c90afd1a4890060baa055cfd0469b6a6f
--- /dev/null
+++ b/src/main/java/com/destroystokyo/paper/io/IOUtil.java
@@ -0,0 +1,63 @@
+package com.destroystokyo.paper.io;
+
+import org.bukkit.Bukkit;
+
+@Deprecated(forRemoval = true)
+public final class IOUtil {
+
+ /* Copied from concrete or concurrentutil */
+
+ public static long getCoordinateKey(final int x, final int z) {
+ return ((long)z << 32) | (x & 0xFFFFFFFFL);
+ }
+
+ public static int getCoordinateX(final long key) {
+ return (int)key;
+ }
+
+ public static int getCoordinateZ(final long key) {
+ return (int)(key >>> 32);
+ }
+
+ public static int getRegionCoordinate(final int chunkCoordinate) {
+ return chunkCoordinate >> 5;
+ }
+
+ public static int getChunkInRegion(final int chunkCoordinate) {
+ return chunkCoordinate & 31;
+ }
+
+ public static String genericToString(final Object object) {
+ return object == null ? "null" : object.getClass().getName() + ":" + object.toString();
+ }
+
+ public static <T> T notNull(final T obj) {
+ if (obj == null) {
+ throw new NullPointerException();
+ }
+ return obj;
+ }
+
+ public static <T> T notNull(final T obj, final String msgIfNull) {
+ if (obj == null) {
+ throw new NullPointerException(msgIfNull);
+ }
+ return obj;
+ }
+
+ public static void arrayBounds(final int off, final int len, final int arrayLength, final String msgPrefix) {
+ if (off < 0 || len < 0 || (arrayLength - off) < len) {
+ throw new ArrayIndexOutOfBoundsException(msgPrefix + ": off: " + off + ", len: " + len + ", array length: " + arrayLength);
+ }
+ }
+
+ public static int getPriorityForCurrentThread() {
+ return Bukkit.isPrimaryThread() ? PrioritizedTaskQueue.HIGHEST_PRIORITY : PrioritizedTaskQueue.NORMAL_PRIORITY;
+ }
+
+ @SuppressWarnings("unchecked")
+ public static <T extends Throwable> void rethrow(final Throwable throwable) throws T {
+ throw (T)throwable;
+ }
+
+}
diff --git a/src/main/java/com/destroystokyo/paper/io/PaperFileIOThread.java b/src/main/java/com/destroystokyo/paper/io/PaperFileIOThread.java
new file mode 100644
index 0000000000000000000000000000000000000000..f2c27e0ac65be4b75c1d86ef6fd45fdb538d96ac
--- /dev/null
+++ b/src/main/java/com/destroystokyo/paper/io/PaperFileIOThread.java
@@ -0,0 +1,474 @@
+package com.destroystokyo.paper.io;
+
+import com.mojang.logging.LogUtils;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.storage.RegionFile;
+import org.slf4j.Logger;
+
+import java.io.IOException;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.atomic.AtomicLong;
+import java.util.function.Consumer;
+import java.util.function.Function;
+
+/**
+ * Prioritized singleton thread responsible for all chunk IO that occurs in a minecraft server.
+ *
+ * <p>
+ * Singleton access: {@link Holder#INSTANCE}
+ * </p>
+ *
+ * <p>
+ * All functions provided are MT-Safe, however certain ordering constraints are (but not enforced):
+ * <li>
+ * Chunk saves may not occur for unloaded chunks.
+ * </li>
+ * <li>
+ * Tasks must be scheduled on the main thread.
+ * </li>
+ * </p>
+ *
+ * @see Holder#INSTANCE
+ * @see #scheduleSave(ServerLevel, int, int, CompoundTag, CompoundTag, int)
+ * @see #loadChunkDataAsync(ServerLevel, int, int, int, Consumer, boolean, boolean, boolean)
+ * @deprecated
+ */
+@Deprecated(forRemoval = true)
+public final class PaperFileIOThread extends QueueExecutorThread {
+
+ public static final Logger LOGGER = LogUtils.getLogger();
+ public static final CompoundTag FAILURE_VALUE = new CompoundTag();
+
+ public static final class Holder {
+
+ public static final PaperFileIOThread INSTANCE = new PaperFileIOThread();
+
+ static {
+ // Paper - fail hard on usage
+ }
+ }
+
+ private final AtomicLong writeCounter = new AtomicLong();
+
+ private PaperFileIOThread() {
+ super(new PrioritizedTaskQueue<>(), (int)(1.0e6)); // 1.0ms spinwait time
+ this.setName("Paper RegionFile IO Thread");
+ this.setPriority(Thread.NORM_PRIORITY - 1); // we keep priority close to normal because threads can wait on us
+ this.setUncaughtExceptionHandler((final Thread unused, final Throwable thr) -> {
+ LOGGER.error("Uncaught exception thrown from IO thread, report this!", thr);
+ });
+ }
+
+ /* run() is implemented by superclass */
+
+ /*
+ *
+ * IO thread will perform reads before writes
+ *
+ * How reads/writes are scheduled:
+ *
+ * If read in progress while scheduling write, ignore read and schedule write
+ * If read in progress while scheduling read (no write in progress), chain the read task
+ *
+ *
+ * If write in progress while scheduling read, use the pending write data and ret immediately
+ * If write in progress while scheduling write (ignore read in progress), overwrite the write in progress data
+ *
+ * This allows the reads and writes to act as if they occur synchronously to the thread scheduling them, however
+ * it fails to properly propagate write failures. When writes fail the data is kept so future reads will actually
+ * read the failed write data. This should hopefully act as a way to prevent data loss for spurious fails for writing data.
+ *
+ */
+
+ /**
+ * Attempts to bump the priority of all IO tasks for the given chunk coordinates. This has no effect if no tasks are queued.
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param priority Priority level to try to bump to
+ */
+ public void bumpPriority(final ServerLevel world, final int chunkX, final int chunkZ, final int priority) {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ public CompoundTag getPendingWrite(final ServerLevel world, final int chunkX, final int chunkZ, final boolean poiData) {
+ // Paper start - rewrite chunk system
+ return io.papermc.paper.chunk.system.io.RegionFileIOThread.getPendingWrite(
+ world, chunkX, chunkZ, poiData ? io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA :
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA
+ );
+ // Paper end - rewrite chunk system
+ }
+
+ /**
+ * Sets the priority of all IO tasks for the given chunk coordinates. This has no effect if no tasks are queued.
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param priority Priority level to set to
+ */
+ public void setPriority(final ServerLevel world, final int chunkX, final int chunkZ, final int priority) {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ /**
+ * Schedules the chunk data to be written asynchronously.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * This function presumes a chunk load for the coordinates is not called during this function (anytime after is OK). This means
+ * saves must be scheduled before a chunk is unloaded.
+ * </li>
+ * <li>
+ * Writes may be called concurrently, although only the "later" write will go through.
+ * </li>
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param poiData Chunk point of interest data. If {@code null}, then no poi data is saved.
+ * @param chunkData Chunk data. If {@code null}, then no chunk data is saved.
+ * @param priority Priority level for this task. See {@link PrioritizedTaskQueue}
+ * @throws IllegalArgumentException If both {@code poiData} and {@code chunkData} are {@code null}.
+ * @throws IllegalStateException If the file io thread has shutdown.
+ */
+ public void scheduleSave(final ServerLevel world, final int chunkX, final int chunkZ,
+ final CompoundTag poiData, final CompoundTag chunkData,
+ final int priority) throws IllegalArgumentException {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ private void scheduleWrite(final ChunkDataController dataController, final ServerLevel world,
+ final int chunkX, final int chunkZ, final CompoundTag data, final int priority, final long writeCounter) {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ /**
+ * Same as {@link #loadChunkDataAsync(ServerLevel, int, int, int, Consumer, boolean, boolean, boolean)}, except this function returns
+ * a {@link CompletableFuture} which is potentially completed <b>ASYNCHRONOUSLY ON THE FILE IO THREAD</b> when the load task
+ * has completed.
+ * <p>
+ * Note that if the chunk fails to load the returned future is completed with {@code null}.
+ * </p>
+ */
+ public CompletableFuture<ChunkData> loadChunkDataAsyncFuture(final ServerLevel world, final int chunkX, final int chunkZ,
+ final int priority, final boolean readPoiData, final boolean readChunkData,
+ final boolean intendingToBlock) {
+ final CompletableFuture<ChunkData> future = new CompletableFuture<>();
+ this.loadChunkDataAsync(world, chunkX, chunkZ, priority, future::complete, readPoiData, readChunkData, intendingToBlock);
+ return future;
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * If a chunk fails to load, the {@code onComplete} parameter is completed with {@code null}.
+ * </li>
+ * <li>
+ * It is possible for the {@code onComplete} parameter to be given {@link ChunkData} containing data
+ * this call did not request.
+ * </li>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param priority Priority level for this task. See {@link PrioritizedTaskQueue}
+ * @param onComplete Consumer to execute once this task has completed
+ * @param readPoiData Whether to read point of interest data. If {@code false}, the {@code NBTTagCompound} will be {@code null}.
+ * @param readChunkData Whether to read chunk data. If {@code false}, the {@code NBTTagCompound} will be {@code null}.
+ * @return The {@link PrioritizedTaskQueue.PrioritizedTask} associated with this task. Note that this task does not support
+ * cancellation.
+ */
+ public void loadChunkDataAsync(final ServerLevel world, final int chunkX, final int chunkZ,
+ final int priority, final Consumer<ChunkData> onComplete,
+ final boolean readPoiData, final boolean readChunkData,
+ final boolean intendingToBlock) {
+ if (!PrioritizedTaskQueue.validPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority: " + priority);
+ }
+
+ if (!(readPoiData | readChunkData)) {
+ throw new IllegalArgumentException("Must read chunk data or poi data");
+ }
+
+ final ChunkData complete = new ChunkData();
+ // Paper start - rewrite chunk system
+ final java.util.List<io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType> types = new java.util.ArrayList<>();
+ if (readPoiData) {
+ types.add(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA);
+ }
+ if (readChunkData) {
+ types.add(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA);
+ }
+ final ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority newPriority;
+ switch (priority) {
+ case PrioritizedTaskQueue.HIGHEST_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.BLOCKING;
+ case PrioritizedTaskQueue.HIGHER_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.HIGHEST;
+ case PrioritizedTaskQueue.HIGH_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.HIGH;
+ case PrioritizedTaskQueue.NORMAL_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.NORMAL;
+ case PrioritizedTaskQueue.LOW_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.LOW;
+ case PrioritizedTaskQueue.LOWEST_PRIORITY -> newPriority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.IDLE;
+ default -> throw new IllegalStateException("Legacy priority " + priority + " should be valid");
+ }
+ final Consumer<io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileData> transformComplete = (io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileData data) -> {
+ if (readPoiData) {
+ if (data.getThrowable(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA) != null) {
+ complete.poiData = FAILURE_VALUE;
+ } else {
+ complete.poiData = data.getData(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA);
+ }
+ }
+
+ if (readChunkData) {
+ if (data.getThrowable(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA) != null) {
+ complete.chunkData = FAILURE_VALUE;
+ } else {
+ complete.chunkData = data.getData(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA);
+ }
+ }
+
+ onComplete.accept(complete);
+ };
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.loadChunkData(world, chunkX, chunkZ, transformComplete, intendingToBlock, newPriority, types.toArray(new io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType[0]));
+ // Paper end - rewrite chunk system
+
+ }
+
+ // Note: the onComplete may be called asynchronously or synchronously here.
+ private void scheduleRead(final ChunkDataController dataController, final ServerLevel world,
+ final int chunkX, final int chunkZ, final Consumer<CompoundTag> onComplete, final int priority,
+ final boolean intendingToBlock) {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ /**
+ * Same as {@link #loadChunkDataAsync(ServerLevel, int, int, int, Consumer, boolean, boolean, boolean)}, except this function returns
+ * the {@link ChunkData} associated with the specified chunk when the task is complete.
+ * @return The chunk data, or {@code null} if the chunk failed to load.
+ */
+ public ChunkData loadChunkData(final ServerLevel world, final int chunkX, final int chunkZ, final int priority,
+ final boolean readPoiData, final boolean readChunkData) {
+ return this.loadChunkDataAsyncFuture(world, chunkX, chunkZ, priority, readPoiData, readChunkData, true).join();
+ }
+
+ /**
+ * Schedules the given task at the specified priority to be executed on the IO thread.
+ * <p>
+ * Internal api. Do not use.
+ * </p>
+ */
+ public void runTask(final int priority, final Runnable runnable) {
+ throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ }
+
+ static final class GeneralTask extends PrioritizedTaskQueue.PrioritizedTask implements Runnable {
+
+ private final Runnable run;
+
+ public GeneralTask(final int priority, final Runnable run) {
+ super(priority);
+ this.run = IOUtil.notNull(run, "Task may not be null");
+ }
+
+ @Override
+ public void run() {
+ try {
+ this.run.run();
+ } catch (final Throwable throwable) {
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ LOGGER.error("Failed to execute general task on IO thread " + IOUtil.genericToString(this.run), throwable);
+ }
+ }
+ }
+
+ public static final class ChunkData {
+
+ public CompoundTag poiData;
+ public CompoundTag chunkData;
+
+ public ChunkData() {}
+
+ public ChunkData(final CompoundTag poiData, final CompoundTag chunkData) {
+ this.poiData = poiData;
+ this.chunkData = chunkData;
+ }
+ }
+
+ public static abstract class ChunkDataController {
+
+ // ConcurrentHashMap synchronizes per chain, so reduce the chance of task's hashes colliding.
+ public final ConcurrentHashMap<Long, ChunkDataTask> tasks = new ConcurrentHashMap<>(64, 0.5f);
+
+ public abstract void writeData(final int x, final int z, final CompoundTag compound) throws IOException;
+ public abstract CompoundTag readData(final int x, final int z) throws IOException;
+
+ public abstract <T> T computeForRegionFile(final int chunkX, final int chunkZ, final Function<RegionFile, T> function);
+ public abstract <T> T computeForRegionFileIfLoaded(final int chunkX, final int chunkZ, final Function<RegionFile, T> function);
+
+ public static final class InProgressWrite {
+ public long writeCounter;
+ public CompoundTag data;
+ }
+
+ public static final class InProgressRead {
+ public final CompletableFuture<CompoundTag> readFuture = new CompletableFuture<>();
+ }
+ }
+
+ public static final class ChunkDataTask extends PrioritizedTaskQueue.PrioritizedTask implements Runnable {
+
+ public ChunkDataController.InProgressWrite inProgressWrite;
+ public ChunkDataController.InProgressRead inProgressRead;
+
+ private final ServerLevel world;
+ private final int x;
+ private final int z;
+ private final ChunkDataController taskController;
+
+ public ChunkDataTask(final int priority, final ServerLevel world, final int x, final int z, final ChunkDataController taskController) {
+ super(priority);
+ this.world = world;
+ this.x = x;
+ this.z = z;
+ this.taskController = taskController;
+ }
+
+ @Override
+ public String toString() {
+ return "Task for world: '" + this.world.getWorld().getName() + "' at " + this.x + "," + this.z +
+ " poi: " + (this.taskController == null) + ", hash: " + this.hashCode(); // Paper - TODO rewrite chunk system
+ }
+
+ /*
+ *
+ * IO thread will perform reads before writes
+ *
+ * How reads/writes are scheduled:
+ *
+ * If read in progress while scheduling write, ignore read and schedule write
+ * If read in progress while scheduling read (no write in progress), chain the read task
+ *
+ *
+ * If write in progress while scheduling read, use the pending write data and ret immediately
+ * If write in progress while scheduling write (ignore read in progress), overwrite the write in progress data
+ *
+ * This allows the reads and writes to act as if they occur synchronously to the thread scheduling them, however
+ * it fails to properly propagate write failures
+ *
+ */
+
+ void reschedule(final int priority) {
+ // priority is checked before this stage // TODO what
+ this.queue.lazySet(null);
+ this.priority.lazySet(priority);
+ PaperFileIOThread.Holder.INSTANCE.queueTask(this);
+ }
+
+ @Override
+ public void run() {
+ if (true) throw new IllegalStateException("Shouldn't get here, use RegionFileIOThread"); // Paper - rewrite chunk system, fail hard on usage
+ ChunkDataController.InProgressRead read = this.inProgressRead;
+ if (read != null) {
+ CompoundTag compound = PaperFileIOThread.FAILURE_VALUE;
+ try {
+ compound = this.taskController.readData(this.x, this.z);
+ } catch (final Throwable thr) {
+ if (thr instanceof ThreadDeath) {
+ throw (ThreadDeath)thr;
+ }
+ LOGGER.error("Failed to read chunk data for task: " + this.toString(), thr);
+ // fall through to complete with null data
+ }
+ read.readFuture.complete(compound);
+ }
+
+ final Long chunkKey = Long.valueOf(IOUtil.getCoordinateKey(this.x, this.z));
+
+ ChunkDataController.InProgressWrite write = this.inProgressWrite;
+
+ if (write == null) {
+ // IntelliJ warns this is invalid, however it does not consider that writes to the task map & the inProgress field can occur concurrently.
+ ChunkDataTask inMap = this.taskController.tasks.compute(chunkKey, (final Long keyInMap, final ChunkDataTask valueInMap) -> {
+ if (valueInMap == null) {
+ throw new IllegalStateException("Write completed concurrently, expected this task: " + ChunkDataTask.this.toString() + ", report this!");
+ }
+ if (valueInMap != ChunkDataTask.this) {
+ throw new IllegalStateException("Chunk task mismatch, expected this task: " + ChunkDataTask.this.toString() + ", got: " + valueInMap.toString() + ", report this!");
+ }
+ return valueInMap.inProgressWrite == null ? null : valueInMap;
+ });
+
+ if (inMap == null) {
+ return; // set the task value to null, indicating we're done
+ }
+
+ // not null, which means there was a concurrent write
+ write = this.inProgressWrite;
+ }
+
+ for (;;) {
+ final long writeCounter;
+ final CompoundTag data;
+
+ //noinspection SynchronizationOnLocalVariableOrMethodParameter
+ synchronized (write) {
+ writeCounter = write.writeCounter;
+ data = write.data;
+ }
+
+ boolean failedWrite = false;
+
+ try {
+ this.taskController.writeData(this.x, this.z, data);
+ } catch (final Throwable thr) {
+ if (thr instanceof ThreadDeath) {
+ throw (ThreadDeath)thr;
+ }
+ LOGGER.error("Failed to write chunk data for task: " + this.toString(), thr);
+ failedWrite = true;
+ }
+
+ boolean finalFailWrite = failedWrite;
+
+ ChunkDataTask inMap = this.taskController.tasks.compute(chunkKey, (final Long keyInMap, final ChunkDataTask valueInMap) -> {
+ if (valueInMap == null) {
+ throw new IllegalStateException("Write completed concurrently, expected this task: " + ChunkDataTask.this.toString() + ", report this!");
+ }
+ if (valueInMap != ChunkDataTask.this) {
+ throw new IllegalStateException("Chunk task mismatch, expected this task: " + ChunkDataTask.this.toString() + ", got: " + valueInMap.toString() + ", report this!");
+ }
+ if (valueInMap.inProgressWrite.writeCounter == writeCounter) {
+ if (finalFailWrite) {
+ valueInMap.inProgressWrite.writeCounter = -1L;
+ }
+
+ return null;
+ }
+ return valueInMap;
+ // Hack end
+ });
+
+ if (inMap == null) {
+ // write counter matched, so we wrote the most up-to-date pending data, we're done here
+ // or we failed to write and successfully set the write counter to -1
+ return; // we're done here
+ }
+
+ // fetch & write new data
+ continue;
+ }
+ }
+ }
+}
diff --git a/src/main/java/com/destroystokyo/paper/io/PrioritizedTaskQueue.java b/src/main/java/com/destroystokyo/paper/io/PrioritizedTaskQueue.java
new file mode 100644
index 0000000000000000000000000000000000000000..7844a3515430472bd829ff246396bceb0797de1b
--- /dev/null
+++ b/src/main/java/com/destroystokyo/paper/io/PrioritizedTaskQueue.java
@@ -0,0 +1,299 @@
+package com.destroystokyo.paper.io;
+
+import java.util.concurrent.ConcurrentLinkedQueue;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.concurrent.atomic.AtomicReference;
+
+@Deprecated(forRemoval = true)
+public class PrioritizedTaskQueue<T extends PrioritizedTaskQueue.PrioritizedTask> {
+
+ // lower numbers are a higher priority (except < 0)
+ // higher priorities are always executed before lower priorities
+
+ /**
+ * Priority value indicating the task has completed or is being completed.
+ */
+ public static final int COMPLETING_PRIORITY = -1;
+
+ /**
+ * Highest priority, should only be used for main thread tasks or tasks that are blocking the main thread.
+ */
+ public static final int HIGHEST_PRIORITY = 0;
+
+ /**
+ * Should be only used in an IO task so that chunk loads do not wait on other IO tasks.
+ * This only exists because IO tasks are scheduled before chunk load tasks to decrease IO waiting times.
+ */
+ public static final int HIGHER_PRIORITY = 1;
+
+ /**
+ * Should be used for scheduling chunk loads/generation that would increase response times to users.
+ */
+ public static final int HIGH_PRIORITY = 2;
+
+ /**
+ * Default priority.
+ */
+ public static final int NORMAL_PRIORITY = 3;
+
+ /**
+ * Use for tasks not at all critical and can potentially be delayed.
+ */
+ public static final int LOW_PRIORITY = 4;
+
+ /**
+ * Use for tasks that should "eventually" execute.
+ */
+ public static final int LOWEST_PRIORITY = 5;
+
+ private static final int TOTAL_PRIORITIES = 6;
+
+ final ConcurrentLinkedQueue<T>[] queues = (ConcurrentLinkedQueue<T>[])new ConcurrentLinkedQueue[TOTAL_PRIORITIES];
+
+ private final AtomicBoolean shutdown = new AtomicBoolean();
+
+ {
+ for (int i = 0; i < TOTAL_PRIORITIES; ++i) {
+ this.queues[i] = new ConcurrentLinkedQueue<>();
+ }
+ }
+
+ /**
+ * Returns whether the specified priority is valid
+ */
+ public static boolean validPriority(final int priority) {
+ return priority >= 0 && priority < TOTAL_PRIORITIES;
+ }
+
+ /**
+ * Queues a task.
+ * @throws IllegalStateException If the task has already been queued. Use {@link PrioritizedTask#raisePriority(int)} to
+ * raise a task's priority.
+ * This can also be thrown if the queue has shutdown.
+ */
+ public void add(final T task) throws IllegalStateException {
+ int priority = task.getPriority();
+ if (priority != COMPLETING_PRIORITY) {
+ task.setQueue(this);
+ this.queues[priority].add(task);
+ }
+ if (this.shutdown.get()) {
+ // note: we're not actually sure at this point if our task will go through
+ throw new IllegalStateException("Queue has shutdown, refusing to execute task " + IOUtil.genericToString(task));
+ }
+ }
+
+ /**
+ * Polls the highest priority task currently available. {@code null} if none.
+ */
+ public T poll() {
+ T task;
+ for (int i = 0; i < TOTAL_PRIORITIES; ++i) {
+ final ConcurrentLinkedQueue<T> queue = this.queues[i];
+
+ while ((task = queue.poll()) != null) {
+ final int prevPriority = task.tryComplete(i);
+ if (prevPriority != COMPLETING_PRIORITY && prevPriority <= i) {
+ // if the prev priority was greater-than or equal to our current priority
+ return task;
+ }
+ }
+ }
+
+ return null;
+ }
+
+ /**
+ * Polls the highest priority task currently available. {@code null} if none.
+ */
+ public T poll(final int lowestPriority) {
+ T task;
+ final int max = Math.min(LOWEST_PRIORITY, lowestPriority);
+ for (int i = 0; i <= max; ++i) {
+ final ConcurrentLinkedQueue<T> queue = this.queues[i];
+
+ while ((task = queue.poll()) != null) {
+ final int prevPriority = task.tryComplete(i);
+ if (prevPriority != COMPLETING_PRIORITY && prevPriority <= i) {
+ // if the prev priority was greater-than or equal to our current priority
+ return task;
+ }
+ }
+ }
+
+ return null;
+ }
+
+ /**
+ * Returns whether this queue may have tasks queued.
+ * <p>
+ * This operation is not atomic, but is MT-Safe.
+ * </p>
+ * @return {@code true} if tasks may be queued, {@code false} otherwise
+ */
+ public boolean hasTasks() {
+ for (int i = 0; i < TOTAL_PRIORITIES; ++i) {
+ final ConcurrentLinkedQueue<T> queue = this.queues[i];
+
+ if (queue.peek() != null) {
+ return true;
+ }
+ }
+ return false;
+ }
+
+ /**
+ * Prevent further additions to this queue. Attempts to add after this call has completed (potentially during) will
+ * result in {@link IllegalStateException} being thrown.
+ * <p>
+ * This operation is atomic with respect to other shutdown calls
+ * </p>
+ * <p>
+ * After this call has completed, regardless of return value, this queue will be shutdown.
+ * </p>
+ * @return {@code true} if the queue was shutdown, {@code false} if it has shut down already
+ */
+ public boolean shutdown() {
+ return this.shutdown.getAndSet(false);
+ }
+
+ public abstract static class PrioritizedTask {
+
+ protected final AtomicReference<PrioritizedTaskQueue> queue = new AtomicReference<>();
+
+ protected final AtomicInteger priority;
+
+ protected PrioritizedTask() {
+ this(PrioritizedTaskQueue.NORMAL_PRIORITY);
+ }
+
+ protected PrioritizedTask(final int priority) {
+ if (!PrioritizedTaskQueue.validPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.priority = new AtomicInteger(priority);
+ }
+
+ /**
+ * Returns the current priority. Note that {@link PrioritizedTaskQueue#COMPLETING_PRIORITY} will be returned
+ * if this task is completing or has completed.
+ */
+ public final int getPriority() {
+ return this.priority.get();
+ }
+
+ /**
+ * Returns whether this task is scheduled to execute, or has been already executed.
+ */
+ public boolean isScheduled() {
+ return this.queue.get() != null;
+ }
+
+ final int tryComplete(final int minPriority) {
+ for (int curr = this.getPriorityVolatile();;) {
+ if (curr == COMPLETING_PRIORITY) {
+ return COMPLETING_PRIORITY;
+ }
+ if (curr > minPriority) {
+ // curr is lower priority
+ return curr;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, COMPLETING_PRIORITY))) {
+ return curr;
+ }
+ continue;
+ }
+ }
+
+ /**
+ * Forces this task to be completed.
+ * @return {@code true} if the task was cancelled, {@code false} if the task has already completed or is being completed.
+ */
+ public boolean cancel() {
+ return this.exchangePriorityVolatile(PrioritizedTaskQueue.COMPLETING_PRIORITY) != PrioritizedTaskQueue.COMPLETING_PRIORITY;
+ }
+
+ /**
+ * Attempts to raise the priority to the priority level specified.
+ * @param priority Priority specified
+ * @return {@code true} if successful, {@code false} otherwise.
+ */
+ public boolean raisePriority(final int priority) {
+ if (!PrioritizedTaskQueue.validPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority");
+ }
+
+ for (int curr = this.getPriorityVolatile();;) {
+ if (curr == COMPLETING_PRIORITY) {
+ return false;
+ }
+ if (priority >= curr) {
+ return true;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority))) {
+ PrioritizedTaskQueue queue = this.queue.get();
+ if (queue != null) {
+ //noinspection unchecked
+ queue.queues[priority].add(this); // silently fail on shutdown
+ }
+ return true;
+ }
+ continue;
+ }
+ }
+
+ /**
+ * Attempts to set this task's priority level to the level specified.
+ * @param priority Specified priority level.
+ * @return {@code true} if successful, {@code false} if this task is completing or has completed.
+ */
+ public boolean updatePriority(final int priority) {
+ if (!PrioritizedTaskQueue.validPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority");
+ }
+
+ for (int curr = this.getPriorityVolatile();;) {
+ if (curr == COMPLETING_PRIORITY) {
+ return false;
+ }
+ if (curr == priority) {
+ return true;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority))) {
+ PrioritizedTaskQueue queue = this.queue.get();
+ if (queue != null) {
+ //noinspection unchecked
+ queue.queues[priority].add(this); // silently fail on shutdown
+ }
+ return true;
+ }
+ continue;
+ }
+ }
+
+ void setQueue(final PrioritizedTaskQueue queue) {
+ this.queue.set(queue);
+ }
+
+ /* priority */
+
+ protected final int getPriorityVolatile() {
+ return this.priority.get();
+ }
+
+ protected final int compareAndExchangePriorityVolatile(final int expect, final int update) {
+ if (this.priority.compareAndSet(expect, update)) {
+ return expect;
+ }
+ return this.priority.get();
+ }
+
+ protected final int exchangePriorityVolatile(final int value) {
+ return this.priority.getAndSet(value);
+ }
+ }
+}
diff --git a/src/main/java/com/destroystokyo/paper/io/QueueExecutorThread.java b/src/main/java/com/destroystokyo/paper/io/QueueExecutorThread.java
new file mode 100644
index 0000000000000000000000000000000000000000..99f49b5625cf51d6c97640553cf5c420bb6fdd36
--- /dev/null
+++ b/src/main/java/com/destroystokyo/paper/io/QueueExecutorThread.java
@@ -0,0 +1,255 @@
+package com.destroystokyo.paper.io;
+
+import com.mojang.logging.LogUtils;
+import org.slf4j.Logger;
+
+import java.util.concurrent.ConcurrentLinkedQueue;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.locks.LockSupport;
+
+@Deprecated(forRemoval = true)
+public class QueueExecutorThread<T extends PrioritizedTaskQueue.PrioritizedTask & Runnable> extends Thread {
+
+ private static final Logger LOGGER = LogUtils.getLogger();
+
+ protected final PrioritizedTaskQueue<T> queue;
+ protected final long spinWaitTime;
+
+ protected volatile boolean closed;
+
+ protected final AtomicBoolean parked = new AtomicBoolean();
+
+ protected volatile ConcurrentLinkedQueue<Thread> flushQueue = new ConcurrentLinkedQueue<>();
+ protected volatile long flushCycles;
+
+ protected int lowestPriorityToPoll = PrioritizedTaskQueue.LOWEST_PRIORITY;
+
+ public int getLowestPriorityToPoll() {
+ return this.lowestPriorityToPoll;
+ }
+
+ public void setLowestPriorityToPoll(final int lowestPriorityToPoll) {
+ if (this.isAlive()) {
+ throw new IllegalStateException("Cannot set after starting");
+ }
+ this.lowestPriorityToPoll = lowestPriorityToPoll;
+ }
+
+ public QueueExecutorThread(final PrioritizedTaskQueue<T> queue) {
+ this(queue, (int)(1.e6)); // 1.0ms
+ }
+
+ public QueueExecutorThread(final PrioritizedTaskQueue<T> queue, final long spinWaitTime) { // in ms
+ this.queue = queue;
+ this.spinWaitTime = spinWaitTime;
+ }
+
+ @Override
+ public void run() {
+ final long spinWaitTime = this.spinWaitTime;
+ main_loop:
+ for (;;) {
+ this.pollTasks(true);
+
+ // spinwait
+
+ final long start = System.nanoTime();
+
+ for (;;) {
+ // If we are interrpted for any reason, park() will always return immediately. Clear so that we don't needlessly use cpu in such an event.
+ Thread.interrupted();
+ LockSupport.parkNanos("Spinwaiting on tasks", 1000L); // 1us
+
+ if (this.pollTasks(true)) {
+ // restart loop, found tasks
+ continue main_loop;
+ }
+
+ if (this.handleClose()) {
+ return; // we're done
+ }
+
+ if ((System.nanoTime() - start) >= spinWaitTime) {
+ break;
+ }
+ }
+
+ if (this.handleClose()) {
+ return;
+ }
+
+ this.parked.set(true);
+
+ // We need to parse here to avoid a race condition where a thread queues a task before we set parked to true
+ // (i.e it will not notify us)
+ if (this.pollTasks(true)) {
+ this.parked.set(false);
+ continue;
+ }
+
+ if (this.handleClose()) {
+ return;
+ }
+
+ // we don't need to check parked before sleeping, but we do need to check parked in a do-while loop
+ // LockSupport.park() can fail for any reason
+ do {
+ Thread.interrupted();
+ LockSupport.park("Waiting on tasks");
+ } while (this.parked.get());
+ }
+ }
+
+ protected boolean handleClose() {
+ if (this.closed) {
+ this.pollTasks(true); // this ensures we've emptied the queue
+ this.handleFlushThreads(true);
+ return true;
+ }
+ return false;
+ }
+
+ protected boolean pollTasks(boolean flushTasks) {
+ Runnable task;
+ boolean ret = false;
+
+ while ((task = this.queue.poll(this.lowestPriorityToPoll)) != null) {
+ ret = true;
+ try {
+ task.run();
+ } catch (final Throwable throwable) {
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ LOGGER.error("Exception thrown from prioritized runnable task in thread '" + this.getName() + "': " + IOUtil.genericToString(task), throwable);
+ }
+ }
+
+ if (flushTasks) {
+ this.handleFlushThreads(false);
+ }
+
+ return ret;
+ }
+
+ protected void handleFlushThreads(final boolean shutdown) {
+ Thread parking;
+ ConcurrentLinkedQueue<Thread> flushQueue = this.flushQueue;
+ do {
+ ++flushCycles; // may be plain read opaque write
+ while ((parking = flushQueue.poll()) != null) {
+ LockSupport.unpark(parking);
+ }
+ } while (this.pollTasks(false));
+
+ if (shutdown) {
+ this.flushQueue = null;
+
+ // defend against a race condition where a flush thread double-checks right before we set to null
+ while ((parking = flushQueue.poll()) != null) {
+ LockSupport.unpark(parking);
+ }
+ }
+ }
+
+ /**
+ * Notify's this thread that a task has been added to its queue
+ * @return {@code true} if this thread was waiting for tasks, {@code false} if it is executing tasks
+ */
+ public boolean notifyTasks() {
+ if (this.parked.get() && this.parked.getAndSet(false)) {
+ LockSupport.unpark(this);
+ return true;
+ }
+ return false;
+ }
+
+ protected void queueTask(final T task) {
+ this.queue.add(task);
+ this.notifyTasks();
+ }
+
+ /**
+ * Waits until this thread's queue is empty.
+ *
+ * @throws IllegalStateException If the current thread is {@code this} thread.
+ */
+ public void flush() {
+ final Thread currentThread = Thread.currentThread();
+
+ if (currentThread == this) {
+ // avoid deadlock
+ throw new IllegalStateException("Cannot flush the queue executor thread while on the queue executor thread");
+ }
+
+ // order is important
+
+ int successes = 0;
+ long lastCycle = -1L;
+
+ do {
+ final ConcurrentLinkedQueue<Thread> flushQueue = this.flushQueue;
+ if (flushQueue == null) {
+ return;
+ }
+
+ flushQueue.add(currentThread);
+
+ // double check flush queue
+ if (this.flushQueue == null) {
+ return;
+ }
+
+ final long currentCycle = this.flushCycles; // may be opaque read
+
+ if (currentCycle == lastCycle) {
+ Thread.yield();
+ continue;
+ }
+
+ // force response
+ this.parked.set(false);
+ LockSupport.unpark(this);
+
+ LockSupport.park("flushing queue executor thread");
+
+ // returns whether there are tasks queued, does not return whether there are tasks executing
+ // this is why we cycle twice twice through flush (we know a pollTask call is made after a flush cycle)
+ // we really only need to guarantee that the tasks this thread has queued has gone through, and can leave
+ // tasks queued concurrently that are unsychronized with this thread as undefined behavior
+ if (this.queue.hasTasks()) {
+ successes = 0;
+ } else {
+ ++successes;
+ }
+
+ } while (successes != 2);
+
+ }
+
+ /**
+ * Closes this queue executor's queue and optionally waits for it to empty.
+ * <p>
+ * If wait is {@code true}, then the queue will be empty by the time this call completes.
+ * </p>
+ * <p>
+ * This function is MT-Safe.
+ * </p>
+ * @param wait If this call is to wait until the queue is empty
+ * @param killQueue Whether to shutdown this thread's queue
+ * @return whether this thread shut down the queue
+ */
+ public boolean close(final boolean wait, final boolean killQueue) {
+ boolean ret = !killQueue ? false : this.queue.shutdown();
+ this.closed = true;
+
+ // force thread to respond to the shutdown
+ this.parked.set(false);
+ LockSupport.unpark(this);
+
+ if (wait) {
+ this.flush();
+ }
+ return ret;
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/PlayerChunkLoader.java b/src/main/java/io/papermc/paper/chunk/PlayerChunkLoader.java
new file mode 100644
index 0000000000000000000000000000000000000000..e77972c4c264100ffdd824bfa2dac58dbbc6d678
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/PlayerChunkLoader.java
@@ -0,0 +1,1128 @@
+package io.papermc.paper.chunk;
+
+import com.destroystokyo.paper.util.misc.PlayerAreaMap;
+import com.destroystokyo.paper.util.misc.PooledLinkedHashSets;
+import io.papermc.paper.configuration.GlobalConfiguration;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.IntervalledCounter;
+import io.papermc.paper.util.TickThread;
+import it.unimi.dsi.fastutil.longs.LongOpenHashSet;
+import it.unimi.dsi.fastutil.objects.Reference2IntOpenHashMap;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectLinkedOpenHashMap;
+import it.unimi.dsi.fastutil.objects.ReferenceLinkedOpenHashSet;
+import net.minecraft.network.protocol.game.ClientboundSetChunkCacheCenterPacket;
+import net.minecraft.network.protocol.game.ClientboundSetChunkCacheRadiusPacket;
+import net.minecraft.network.protocol.game.ClientboundSetSimulationDistancePacket;
+import io.papermc.paper.util.MCUtil;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.server.MinecraftServer;
+import net.minecraft.server.level.*;
+import net.minecraft.util.Mth;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.LevelChunk;
+import org.apache.commons.lang3.mutable.MutableObject;
+import org.bukkit.craftbukkit.entity.CraftPlayer;
+import org.bukkit.entity.Player;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.TreeSet;
+import java.util.concurrent.atomic.AtomicInteger;
+
+public final class PlayerChunkLoader {
+
+ public static final int MIN_VIEW_DISTANCE = 2;
+ public static final int MAX_VIEW_DISTANCE = 32;
+
+ public static final int TICK_TICKET_LEVEL = 31;
+ public static final int LOADED_TICKET_LEVEL = 33;
+
+ public static int getTickViewDistance(final Player player) {
+ return getTickViewDistance(((CraftPlayer)player).getHandle());
+ }
+
+ public static int getTickViewDistance(final ServerPlayer player) {
+ final ServerLevel level = (ServerLevel)player.level;
+ final PlayerLoaderData data = level.chunkSource.chunkMap.playerChunkManager.getData(player);
+ if (data == null) {
+ return level.chunkSource.chunkMap.playerChunkManager.getTargetTickViewDistance();
+ }
+ return data.getTargetTickViewDistance();
+ }
+
+ public static int getLoadViewDistance(final Player player) {
+ return getLoadViewDistance(((CraftPlayer)player).getHandle());
+ }
+
+ public static int getLoadViewDistance(final ServerPlayer player) {
+ final ServerLevel level = (ServerLevel)player.level;
+ final PlayerLoaderData data = level.chunkSource.chunkMap.playerChunkManager.getData(player);
+ if (data == null) {
+ return level.chunkSource.chunkMap.playerChunkManager.getLoadDistance();
+ }
+ return data.getLoadDistance();
+ }
+
+ public static int getSendViewDistance(final Player player) {
+ return getSendViewDistance(((CraftPlayer)player).getHandle());
+ }
+
+ public static int getSendViewDistance(final ServerPlayer player) {
+ final ServerLevel level = (ServerLevel)player.level;
+ final PlayerLoaderData data = level.chunkSource.chunkMap.playerChunkManager.getData(player);
+ if (data == null) {
+ return level.chunkSource.chunkMap.playerChunkManager.getTargetSendDistance();
+ }
+ return data.getTargetSendViewDistance();
+ }
+
+ protected final ChunkMap chunkMap;
+ protected final Reference2ObjectLinkedOpenHashMap<ServerPlayer, PlayerLoaderData> playerMap = new Reference2ObjectLinkedOpenHashMap<>(512, 0.7f);
+ protected final ReferenceLinkedOpenHashSet<PlayerLoaderData> chunkSendQueue = new ReferenceLinkedOpenHashSet<>(512, 0.7f);
+
+ protected final TreeSet<PlayerLoaderData> chunkLoadQueue = new TreeSet<>((final PlayerLoaderData p1, final PlayerLoaderData p2) -> {
+ if (p1 == p2) {
+ return 0;
+ }
+
+ final ChunkPriorityHolder holder1 = p1.loadQueue.peekFirst();
+ final ChunkPriorityHolder holder2 = p2.loadQueue.peekFirst();
+
+ final int priorityCompare = Double.compare(holder1 == null ? Double.MAX_VALUE : holder1.priority, holder2 == null ? Double.MAX_VALUE : holder2.priority);
+
+ final int lastLoadTimeCompare = Long.compare(p1.lastChunkLoad - p2.lastChunkLoad, 0);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ if ((holder1 == null || holder2 == null || lastLoadTimeCompare == 0 || holder1.priority < 0.0 || holder2.priority < 0.0) && priorityCompare != 0) {
+ return priorityCompare;
+ }
+
+ if (lastLoadTimeCompare != 0) {
+ return lastLoadTimeCompare;
+ }
+
+ final int idCompare = Integer.compare(p1.player.getId(), p2.player.getId());
+
+ if (idCompare != 0) {
+ return idCompare;
+ }
+
+ // last resort
+ return Integer.compare(System.identityHashCode(p1), System.identityHashCode(p2));
+ });
+
+ protected final TreeSet<PlayerLoaderData> chunkSendWaitQueue = new TreeSet<>((final PlayerLoaderData p1, final PlayerLoaderData p2) -> {
+ if (p1 == p2) {
+ return 0;
+ }
+
+ final int timeCompare = Long.compare(p1.nextChunkSendTarget - p2.nextChunkSendTarget, 0);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ if (timeCompare != 0) {
+ return timeCompare;
+ }
+
+ final int idCompare = Integer.compare(p1.player.getId(), p2.player.getId());
+
+ if (idCompare != 0) {
+ return idCompare;
+ }
+
+ // last resort
+ return Integer.compare(System.identityHashCode(p1), System.identityHashCode(p2));
+ });
+
+
+ // no throttling is applied below this VD for loading
+
+ /**
+ * The chunks to be sent to players, provided they're send-ready. Send-ready means the chunk and its 1 radius neighbours are loaded.
+ */
+ public final PlayerAreaMap broadcastMap;
+
+ /**
+ * The chunks to be brought up to send-ready status. Send-ready means the chunk and its 1 radius neighbours are loaded.
+ */
+ public final PlayerAreaMap loadMap;
+
+ /**
+ * Areamap used only to remove tickets for send-ready chunks. View distance is always + 1 of load view distance. Thus,
+ * this map is always representing the chunks we are actually going to load.
+ */
+ public final PlayerAreaMap loadTicketCleanup;
+
+ /**
+ * The chunks to brought to ticking level. Each chunk must have 2 radius neighbours loaded before this can happen.
+ */
+ public final PlayerAreaMap tickMap;
+
+ /**
+ * -1 if defaulting to [load distance], else always in [2, load distance]
+ */
+ protected int rawSendDistance = -1;
+
+ /**
+ * -1 if defaulting to [tick view distance + 1], else always in [tick view distance + 1, 32 + 1]
+ */
+ protected int rawLoadDistance = -1;
+
+ /**
+ * Never -1, always in [2, 32]
+ */
+ protected int rawTickDistance = -1;
+
+ // methods to bridge for API
+
+ public int getTargetTickViewDistance() {
+ return this.getTickDistance();
+ }
+
+ public void setTargetTickViewDistance(final int distance) {
+ this.setTickDistance(distance);
+ }
+
+ public int getTargetNoTickViewDistance() {
+ return this.getLoadDistance() - 1;
+ }
+
+ public void setTargetNoTickViewDistance(final int distance) {
+ this.setLoadDistance(distance == -1 ? -1 : distance + 1);
+ }
+
+ public int getTargetSendDistance() {
+ return this.rawSendDistance == -1 ? this.getLoadDistance() : this.rawSendDistance;
+ }
+
+ public void setTargetSendDistance(final int distance) {
+ this.setSendDistance(distance);
+ }
+
+ // internal methods
+
+ public int getSendDistance() {
+ final int loadDistance = this.getLoadDistance();
+ return this.rawSendDistance == -1 ? loadDistance : Math.min(this.rawSendDistance, loadDistance);
+ }
+
+ public void setSendDistance(final int distance) {
+ if (distance != -1 && (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE + 1)) {
+ throw new IllegalArgumentException("Send distance must be a number between " + MIN_VIEW_DISTANCE + " and " + (MAX_VIEW_DISTANCE + 1) + ", or -1, got: " + distance);
+ }
+ this.rawSendDistance = distance;
+ }
+
+ public int getLoadDistance() {
+ final int tickDistance = this.getTickDistance();
+ return this.rawLoadDistance == -1 ? tickDistance + 1 : Math.max(tickDistance + 1, this.rawLoadDistance);
+ }
+
+ public void setLoadDistance(final int distance) {
+ if (distance != -1 && (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE + 1)) {
+ throw new IllegalArgumentException("Load distance must be a number between " + MIN_VIEW_DISTANCE + " and " + (MAX_VIEW_DISTANCE + 1) + ", or -1, got: " + distance);
+ }
+ this.rawLoadDistance = distance;
+ }
+
+ public int getTickDistance() {
+ return this.rawTickDistance;
+ }
+
+ public void setTickDistance(final int distance) {
+ if (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE) {
+ throw new IllegalArgumentException("View distance must be a number between " + MIN_VIEW_DISTANCE + " and " + MAX_VIEW_DISTANCE + ", got: " + distance);
+ }
+ this.rawTickDistance = distance;
+ }
+
+ /*
+ Players have 3 different types of view distance:
+ 1. Sending view distance
+ 2. Loading view distance
+ 3. Ticking view distance
+
+ But for configuration purposes (and API) there are:
+ 1. No-tick view distance
+ 2. Tick view distance
+ 3. Broadcast view distance
+
+ These aren't always the same as the types we represent internally.
+
+ Loading view distance is always max(no-tick + 1, tick + 1)
+ - no-tick has 1 added because clients need an extra radius to render chunks
+ - tick has 1 added because it needs an extra radius of chunks to load before they can be marked ticking
+
+ Loading view distance is defined as the radius of chunks that will be brought to send-ready status, which means
+ it loads chunks in radius load-view-distance + 1.
+
+ The maximum value for send view distance is the load view distance. API can set it lower.
+ */
+
+ public PlayerChunkLoader(final ChunkMap chunkMap, final PooledLinkedHashSets<ServerPlayer> pooledHashSets) {
+ this.chunkMap = chunkMap;
+ this.broadcastMap = new PlayerAreaMap(pooledHashSets,
+ null,
+ (ServerPlayer player, int rangeX, int rangeZ, int currPosX, int currPosZ, int prevPosX, int prevPosZ,
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> newState) -> {
+ PlayerChunkLoader.this.onChunkLeave(player, rangeX, rangeZ);
+ });
+ this.loadMap = new PlayerAreaMap(pooledHashSets,
+ null,
+ (ServerPlayer player, int rangeX, int rangeZ, int currPosX, int currPosZ, int prevPosX, int prevPosZ,
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> newState) -> {
+ if (newState != null) {
+ return;
+ }
+ PlayerChunkLoader.this.isTargetedForPlayerLoad.remove(CoordinateUtils.getChunkKey(rangeX, rangeZ));
+ });
+ this.loadTicketCleanup = new PlayerAreaMap(pooledHashSets,
+ null,
+ (ServerPlayer player, int rangeX, int rangeZ, int currPosX, int currPosZ, int prevPosX, int prevPosZ,
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> newState) -> {
+ if (newState != null) {
+ return;
+ }
+ ChunkPos chunkPos = new ChunkPos(rangeX, rangeZ);
+ PlayerChunkLoader.this.chunkMap.level.getChunkSource().removeTicketAtLevel(TicketType.PLAYER, chunkPos, LOADED_TICKET_LEVEL, chunkPos);
+ if (PlayerChunkLoader.this.chunkTicketTracker.remove(chunkPos.toLong())) {
+ --PlayerChunkLoader.this.concurrentChunkLoads;
+ }
+ });
+ this.tickMap = new PlayerAreaMap(pooledHashSets,
+ (ServerPlayer player, int rangeX, int rangeZ, int currPosX, int currPosZ, int prevPosX, int prevPosZ,
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> newState) -> {
+ if (newState.size() != 1) {
+ return;
+ }
+ LevelChunk chunk = PlayerChunkLoader.this.chunkMap.level.getChunkSource().getChunkAtIfLoadedMainThreadNoCache(rangeX, rangeZ);
+ if (chunk == null || !chunk.areNeighboursLoaded(2)) {
+ return;
+ }
+
+ ChunkPos chunkPos = new ChunkPos(rangeX, rangeZ);
+ PlayerChunkLoader.this.chunkMap.level.getChunkSource().addTicketAtLevel(TicketType.PLAYER, chunkPos, TICK_TICKET_LEVEL, chunkPos);
+ },
+ (ServerPlayer player, int rangeX, int rangeZ, int currPosX, int currPosZ, int prevPosX, int prevPosZ,
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> newState) -> {
+ if (newState != null) {
+ return;
+ }
+ ChunkPos chunkPos = new ChunkPos(rangeX, rangeZ);
+ PlayerChunkLoader.this.chunkMap.level.getChunkSource().removeTicketAtLevel(TicketType.PLAYER, chunkPos, TICK_TICKET_LEVEL, chunkPos);
+ });
+ }
+
+ protected final LongOpenHashSet isTargetedForPlayerLoad = new LongOpenHashSet();
+ protected final LongOpenHashSet chunkTicketTracker = new LongOpenHashSet();
+
+ public boolean isChunkNearPlayers(final int chunkX, final int chunkZ) {
+ final PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> playersInSendRange = this.broadcastMap.getObjectsInRange(chunkX, chunkZ);
+
+ return playersInSendRange != null;
+ }
+
+ public void onChunkPostProcessing(final int chunkX, final int chunkZ) {
+ this.onChunkSendReady(chunkX, chunkZ);
+ }
+
+ private boolean chunkNeedsPostProcessing(final int chunkX, final int chunkZ) {
+ final long key = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+ final ChunkHolder chunk = this.chunkMap.getVisibleChunkIfPresent(key);
+
+ if (chunk == null) {
+ return false;
+ }
+
+ final LevelChunk levelChunk = chunk.getSendingChunk();
+
+ return levelChunk != null && !levelChunk.isPostProcessingDone;
+ }
+
+ // rets whether the chunk is at a loaded stage that is ready to be sent to players
+ public boolean isChunkPlayerLoaded(final int chunkX, final int chunkZ) {
+ final long key = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+ final ChunkHolder chunk = this.chunkMap.getVisibleChunkIfPresent(key);
+
+ if (chunk == null) {
+ return false;
+ }
+
+ final LevelChunk levelChunk = chunk.getSendingChunk();
+
+ return levelChunk != null && levelChunk.isPostProcessingDone && this.isTargetedForPlayerLoad.contains(key);
+ }
+
+ public boolean isChunkSent(final ServerPlayer player, final int chunkX, final int chunkZ, final boolean borderOnly) {
+ return borderOnly ? this.isChunkSentBorderOnly(player, chunkX, chunkZ) : this.isChunkSent(player, chunkX, chunkZ);
+ }
+
+ public boolean isChunkSent(final ServerPlayer player, final int chunkX, final int chunkZ) {
+ final PlayerLoaderData data = this.playerMap.get(player);
+ if (data == null) {
+ return false;
+ }
+
+ return data.hasSentChunk(chunkX, chunkZ);
+ }
+
+ public boolean isChunkSentBorderOnly(final ServerPlayer player, final int chunkX, final int chunkZ) {
+ final PlayerLoaderData data = this.playerMap.get(player);
+ if (data == null) {
+ return false;
+ }
+
+ final boolean center = data.hasSentChunk(chunkX, chunkZ);
+ if (!center) {
+ return false;
+ }
+
+ return !(data.hasSentChunk(chunkX - 1, chunkZ) && data.hasSentChunk(chunkX + 1, chunkZ) &&
+ data.hasSentChunk(chunkX, chunkZ - 1) && data.hasSentChunk(chunkX, chunkZ + 1));
+ }
+
+ protected int getMaxConcurrentChunkSends() {
+ return GlobalConfiguration.get().chunkLoading.maxConcurrentSends;
+ }
+
+ protected int getMaxChunkLoads() {
+ double config = GlobalConfiguration.get().chunkLoading.playerMaxConcurrentLoads;
+ double max = GlobalConfiguration.get().chunkLoading.globalMaxConcurrentLoads;
+ return (int)Math.ceil(Math.min(config * MinecraftServer.getServer().getPlayerCount(), max <= 1.0 ? Double.MAX_VALUE : max));
+ }
+
+ protected long getTargetSendPerPlayerAddend() {
+ return GlobalConfiguration.get().chunkLoading.targetPlayerChunkSendRate <= 1.0 ? 0L : (long)Math.round(1.0e9 / GlobalConfiguration.get().chunkLoading.targetPlayerChunkSendRate);
+ }
+
+ protected long getMaxSendAddend() {
+ return GlobalConfiguration.get().chunkLoading.globalMaxChunkSendRate <= 1.0 ? 0L : (long)Math.round(1.0e9 / GlobalConfiguration.get().chunkLoading.globalMaxChunkSendRate);
+ }
+
+ public void onChunkPlayerTickReady(final int chunkX, final int chunkZ) {
+ final ChunkPos chunkPos = new ChunkPos(chunkX, chunkZ);
+ this.chunkMap.level.getChunkSource().addTicketAtLevel(TicketType.PLAYER, chunkPos, TICK_TICKET_LEVEL, chunkPos);
+ }
+
+ public void onChunkSendReady(final int chunkX, final int chunkZ) {
+ final PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> playersInSendRange = this.broadcastMap.getObjectsInRange(chunkX, chunkZ);
+
+ if (playersInSendRange == null) {
+ return;
+ }
+
+ final Object[] rawData = playersInSendRange.getBackingSet();
+ for (int i = 0, len = rawData.length; i < len; ++i) {
+ final Object raw = rawData[i];
+
+ if (!(raw instanceof ServerPlayer)) {
+ continue;
+ }
+ this.onChunkSendReady((ServerPlayer)raw, chunkX, chunkZ);
+ }
+ }
+
+ public void onChunkSendReady(final ServerPlayer player, final int chunkX, final int chunkZ) {
+ final PlayerLoaderData data = this.playerMap.get(player);
+
+ if (data == null) {
+ return;
+ }
+
+ if (data.hasSentChunk(chunkX, chunkZ) || !this.isChunkPlayerLoaded(chunkX, chunkZ)) {
+ // if we don't have player tickets, then the load logic will pick this up and queue to send
+ return;
+ }
+
+ if (!data.chunksToBeSent.remove(CoordinateUtils.getChunkKey(chunkX, chunkZ))) {
+ // don't queue to send, we don't want the chunk
+ return;
+ }
+
+ final long playerPos = this.broadcastMap.getLastCoordinate(player);
+ final int playerChunkX = CoordinateUtils.getChunkX(playerPos);
+ final int playerChunkZ = CoordinateUtils.getChunkZ(playerPos);
+ final int manhattanDistance = Math.abs(playerChunkX - chunkX) + Math.abs(playerChunkZ - chunkZ);
+
+ final ChunkPriorityHolder holder = new ChunkPriorityHolder(chunkX, chunkZ, manhattanDistance, 0.0);
+ data.sendQueue.add(holder);
+ }
+
+ public void onChunkLoad(final int chunkX, final int chunkZ) {
+ if (this.chunkTicketTracker.remove(CoordinateUtils.getChunkKey(chunkX, chunkZ))) {
+ --this.concurrentChunkLoads;
+ }
+ }
+
+ public void onChunkLeave(final ServerPlayer player, final int chunkX, final int chunkZ) {
+ final PlayerLoaderData data = this.playerMap.get(player);
+
+ if (data == null) {
+ return;
+ }
+
+ data.unloadChunk(chunkX, chunkZ);
+ }
+
+ public void addPlayer(final ServerPlayer player) {
+ TickThread.ensureTickThread("Cannot add player async");
+ if (!player.isRealPlayer) {
+ return;
+ }
+ final PlayerLoaderData data = new PlayerLoaderData(player, this);
+ if (this.playerMap.putIfAbsent(player, data) == null) {
+ data.update();
+ }
+ }
+
+ public void removePlayer(final ServerPlayer player) {
+ TickThread.ensureTickThread("Cannot remove player async");
+ if (!player.isRealPlayer) {
+ return;
+ }
+
+ final PlayerLoaderData loaderData = this.playerMap.remove(player);
+ if (loaderData == null) {
+ return;
+ }
+ loaderData.remove();
+ this.chunkLoadQueue.remove(loaderData);
+ this.chunkSendQueue.remove(loaderData);
+ this.chunkSendWaitQueue.remove(loaderData);
+ synchronized (this.sendingChunkCounts) {
+ final int count = this.sendingChunkCounts.removeInt(loaderData);
+ if (count != 0) {
+ concurrentChunkSends.getAndAdd(-count);
+ }
+ }
+ }
+
+ public void updatePlayer(final ServerPlayer player) {
+ TickThread.ensureTickThread("Cannot update player async");
+ if (!player.isRealPlayer) {
+ return;
+ }
+ final PlayerLoaderData loaderData = this.playerMap.get(player);
+ if (loaderData != null) {
+ loaderData.update();
+ }
+ }
+
+ public PlayerLoaderData getData(final ServerPlayer player) {
+ return this.playerMap.get(player);
+ }
+
+ public void tick() {
+ TickThread.ensureTickThread("Cannot tick async");
+ for (final PlayerLoaderData data : this.playerMap.values()) {
+ data.update();
+ }
+ this.tickMidTick();
+ }
+
+ protected static final AtomicInteger concurrentChunkSends = new AtomicInteger();
+ protected final Reference2IntOpenHashMap<PlayerLoaderData> sendingChunkCounts = new Reference2IntOpenHashMap<>();
+ private static long nextChunkSend;
+ private void trySendChunks() {
+ final long time = System.nanoTime();
+ if (nextChunkSend - time > 0) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return;
+ }
+ // drain entries from wait queue
+ while (!this.chunkSendWaitQueue.isEmpty()) {
+ final PlayerLoaderData data = this.chunkSendWaitQueue.first();
+
+ if (data.nextChunkSendTarget - time > 0) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ break;
+ }
+
+ this.chunkSendWaitQueue.pollFirst();
+
+ this.chunkSendQueue.add(data);
+ }
+
+ if (this.chunkSendQueue.isEmpty()) {
+ return;
+ }
+
+ final int maxSends = this.getMaxConcurrentChunkSends();
+ final long nextPlayerDeadline = this.getTargetSendPerPlayerAddend() + time;
+ for (;;) {
+ if (this.chunkSendQueue.isEmpty()) {
+ break;
+ }
+ final int currSends = concurrentChunkSends.get();
+ if (currSends >= maxSends) {
+ break;
+ }
+
+ if (!concurrentChunkSends.compareAndSet(currSends, currSends + 1)) {
+ continue;
+ }
+
+ // send chunk
+
+ final PlayerLoaderData data = this.chunkSendQueue.removeFirst();
+
+ final ChunkPriorityHolder queuedSend = data.sendQueue.pollFirst();
+ if (queuedSend == null) {
+ concurrentChunkSends.getAndDecrement(); // we never sent, so decrease
+ // stop iterating over players who have nothing to send
+ if (this.chunkSendQueue.isEmpty()) {
+ // nothing left
+ break;
+ }
+ continue;
+ }
+
+ if (!this.isChunkPlayerLoaded(queuedSend.chunkX, queuedSend.chunkZ)) {
+ throw new IllegalStateException();
+ }
+
+ data.nextChunkSendTarget = nextPlayerDeadline;
+ this.chunkSendWaitQueue.add(data);
+
+ synchronized (this.sendingChunkCounts) {
+ this.sendingChunkCounts.addTo(data, 1);
+ }
+
+ data.sendChunk(queuedSend.chunkX, queuedSend.chunkZ, () -> {
+ synchronized (this.sendingChunkCounts) {
+ final int count = this.sendingChunkCounts.getInt(data);
+ if (count == 0) {
+ // disconnected, so we don't need to decrement: it will be decremented for us
+ return;
+ }
+ if (count == 1) {
+ this.sendingChunkCounts.removeInt(data);
+ } else {
+ this.sendingChunkCounts.put(data, count - 1);
+ }
+ }
+
+ concurrentChunkSends.getAndDecrement();
+ });
+
+ nextChunkSend = this.getMaxSendAddend() + time;
+ if (nextChunkSend - time > 0) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ break;
+ }
+ }
+ }
+
+ protected int concurrentChunkLoads;
+ // this interval prevents bursting a lot of chunk loads
+ protected static final IntervalledCounter TICKET_ADDITION_COUNTER_SHORT = new IntervalledCounter((long)(1.0e6 * 50.0)); // 50ms
+ // this interval ensures the rate is kept between ticks correctly
+ protected static final IntervalledCounter TICKET_ADDITION_COUNTER_LONG = new IntervalledCounter((long)(1.0e6 * 1000.0)); // 1000ms
+ private void tryLoadChunks() {
+ if (this.chunkLoadQueue.isEmpty()) {
+ return;
+ }
+
+ final int maxLoads = this.getMaxChunkLoads();
+ final long time = System.nanoTime();
+ boolean updatedCounters = false;
+ for (;;) {
+ final PlayerLoaderData data = this.chunkLoadQueue.pollFirst();
+
+ data.lastChunkLoad = time;
+
+ final ChunkPriorityHolder queuedLoad = data.loadQueue.peekFirst();
+ if (queuedLoad == null) {
+ if (this.chunkLoadQueue.isEmpty()) {
+ break;
+ }
+ continue;
+ }
+
+ if (!updatedCounters) {
+ updatedCounters = true;
+ TICKET_ADDITION_COUNTER_SHORT.updateCurrentTime(time);
+ TICKET_ADDITION_COUNTER_LONG.updateCurrentTime(time);
+ data.ticketAdditionCounterShort.updateCurrentTime(time);
+ data.ticketAdditionCounterLong.updateCurrentTime(time);
+ }
+
+ if (this.isChunkPlayerLoaded(queuedLoad.chunkX, queuedLoad.chunkZ)) {
+ // already loaded!
+ data.loadQueue.pollFirst(); // already loaded so we just skip
+ this.chunkLoadQueue.add(data);
+
+ // ensure the chunk is queued to send
+ this.onChunkSendReady(queuedLoad.chunkX, queuedLoad.chunkZ);
+ continue;
+ }
+
+ final long chunkKey = CoordinateUtils.getChunkKey(queuedLoad.chunkX, queuedLoad.chunkZ);
+
+ final double priority = queuedLoad.priority;
+ // while we do need to rate limit chunk loads, the logic for sending chunks requires that tickets are present.
+ // when chunks are loaded (i.e spawn) but do not have this player's tickets, they have to wait behind the
+ // load queue. To avoid this problem, we check early here if tickets are required to load the chunk - if they
+ // aren't required, it bypasses the limiter system.
+ boolean unloadedTargetChunk = false;
+ unloaded_check:
+ for (int dz = -1; dz <= 1; ++dz) {
+ for (int dx = -1; dx <= 1; ++dx) {
+ final int offX = queuedLoad.chunkX + dx;
+ final int offZ = queuedLoad.chunkZ + dz;
+ if (this.chunkMap.level.getChunkSource().getChunkAtIfLoadedMainThreadNoCache(offX, offZ) == null) {
+ unloadedTargetChunk = true;
+ break unloaded_check;
+ }
+ }
+ }
+ if (unloadedTargetChunk && priority >= 0.0) {
+ // priority >= 0.0 implies rate limited chunks
+
+ final int currentChunkLoads = this.concurrentChunkLoads;
+ if (currentChunkLoads >= maxLoads || (GlobalConfiguration.get().chunkLoading.globalMaxChunkLoadRate > 0 && (TICKET_ADDITION_COUNTER_SHORT.getRate() >= GlobalConfiguration.get().chunkLoading.globalMaxChunkLoadRate || TICKET_ADDITION_COUNTER_LONG.getRate() >= GlobalConfiguration.get().chunkLoading.globalMaxChunkLoadRate))
+ || (GlobalConfiguration.get().chunkLoading.playerMaxChunkLoadRate > 0.0 && (data.ticketAdditionCounterShort.getRate() >= GlobalConfiguration.get().chunkLoading.playerMaxChunkLoadRate || data.ticketAdditionCounterLong.getRate() >= GlobalConfiguration.get().chunkLoading.playerMaxChunkLoadRate))) {
+ // don't poll, we didn't load it
+ this.chunkLoadQueue.add(data);
+ break;
+ }
+ }
+
+ // can only poll after we decide to load
+ data.loadQueue.pollFirst();
+
+ // now that we've polled we can re-add to load queue
+ this.chunkLoadQueue.add(data);
+
+ // add necessary tickets to load chunk up to send-ready
+ for (int dz = -1; dz <= 1; ++dz) {
+ for (int dx = -1; dx <= 1; ++dx) {
+ final int offX = queuedLoad.chunkX + dx;
+ final int offZ = queuedLoad.chunkZ + dz;
+ final ChunkPos chunkPos = new ChunkPos(offX, offZ);
+
+ this.chunkMap.level.getChunkSource().addTicketAtLevel(TicketType.PLAYER, chunkPos, LOADED_TICKET_LEVEL, chunkPos);
+ if (this.chunkMap.level.getChunkSource().getChunkAtIfLoadedMainThreadNoCache(offX, offZ) != null) {
+ continue;
+ }
+
+ if (priority > 0.0 && this.chunkTicketTracker.add(CoordinateUtils.getChunkKey(offX, offZ))) {
+ // won't reach here if unloadedTargetChunk is false
+ ++this.concurrentChunkLoads;
+ TICKET_ADDITION_COUNTER_SHORT.addTime(time);
+ TICKET_ADDITION_COUNTER_LONG.addTime(time);
+ data.ticketAdditionCounterShort.addTime(time);
+ data.ticketAdditionCounterLong.addTime(time);
+ }
+ }
+ }
+
+ // mark that we've added tickets here
+ this.isTargetedForPlayerLoad.add(chunkKey);
+
+ // it's possible all we needed was the player tickets to queue up the send.
+ if (this.isChunkPlayerLoaded(queuedLoad.chunkX, queuedLoad.chunkZ)) {
+ // yup, all we needed.
+ this.onChunkSendReady(queuedLoad.chunkX, queuedLoad.chunkZ);
+ } else if (this.chunkNeedsPostProcessing(queuedLoad.chunkX, queuedLoad.chunkZ)) {
+ // requires post processing
+ this.chunkMap.mainThreadExecutor.execute(() -> {
+ final long key = CoordinateUtils.getChunkKey(queuedLoad.chunkX, queuedLoad.chunkZ);
+ final ChunkHolder holder = PlayerChunkLoader.this.chunkMap.getVisibleChunkIfPresent(key);
+
+ if (holder == null) {
+ return;
+ }
+
+ final LevelChunk chunk = holder.getSendingChunk();
+
+ if (chunk != null && !chunk.isPostProcessingDone) {
+ chunk.postProcessGeneration();
+ }
+ });
+ }
+ }
+ }
+
+ public void tickMidTick() {
+ // try to send more chunks
+ this.trySendChunks();
+
+ // try to queue more chunks to load
+ this.tryLoadChunks();
+ }
+
+ static final class ChunkPriorityHolder {
+ public final int chunkX;
+ public final int chunkZ;
+ public final int manhattanDistanceToPlayer;
+ public final double priority;
+
+ public ChunkPriorityHolder(final int chunkX, final int chunkZ, final int manhattanDistanceToPlayer, final double priority) {
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.manhattanDistanceToPlayer = manhattanDistanceToPlayer;
+ this.priority = priority;
+ }
+ }
+
+ public static final class PlayerLoaderData {
+
+ protected static final float FOV = 110.0f;
+ protected static final double PRIORITISED_DISTANCE = 12.0 * 16.0;
+
+ // Player max sprint speed is approximately 8m/s
+ protected static final double LOOK_PRIORITY_SPEED_THRESHOLD = (10.0/20.0) * (10.0/20.0);
+ protected static final double LOOK_PRIORITY_YAW_DELTA_RECALC_THRESHOLD = 3.0f;
+
+ protected double lastLocX = Double.NEGATIVE_INFINITY;
+ protected double lastLocZ = Double.NEGATIVE_INFINITY;
+
+ protected int lastChunkX = Integer.MIN_VALUE;
+ protected int lastChunkZ = Integer.MIN_VALUE;
+
+ // this is corrected so that 0 is along the positive x-axis
+ protected float lastYaw = Float.NEGATIVE_INFINITY;
+
+ protected int lastSendDistance = Integer.MIN_VALUE;
+ protected int lastLoadDistance = Integer.MIN_VALUE;
+ protected int lastTickDistance = Integer.MIN_VALUE;
+ protected boolean usingLookingPriority;
+
+ protected final ServerPlayer player;
+ protected final PlayerChunkLoader loader;
+
+ // warning: modifications of this field must be aware that the loadQueue inside PlayerChunkLoader uses this field
+ // in a comparator!
+ protected final ArrayDeque<ChunkPriorityHolder> loadQueue = new ArrayDeque<>();
+ protected final LongOpenHashSet sentChunks = new LongOpenHashSet();
+ protected final LongOpenHashSet chunksToBeSent = new LongOpenHashSet();
+
+ protected final TreeSet<ChunkPriorityHolder> sendQueue = new TreeSet<>((final ChunkPriorityHolder p1, final ChunkPriorityHolder p2) -> {
+ final int distanceCompare = Integer.compare(p1.manhattanDistanceToPlayer, p2.manhattanDistanceToPlayer);
+ if (distanceCompare != 0) {
+ return distanceCompare;
+ }
+
+ final int coordinateXCompare = Integer.compare(p1.chunkX, p2.chunkX);
+ if (coordinateXCompare != 0) {
+ return coordinateXCompare;
+ }
+
+ return Integer.compare(p1.chunkZ, p2.chunkZ);
+ });
+
+ protected int sendViewDistance = -1;
+ protected int loadViewDistance = -1;
+ protected int tickViewDistance = -1;
+
+ protected long nextChunkSendTarget;
+
+ // this interval prevents bursting a lot of chunk loads
+ protected final IntervalledCounter ticketAdditionCounterShort = new IntervalledCounter((long)(1.0e6 * 50.0)); // 50ms
+ // this ensures the rate is kept between ticks correctly
+ protected final IntervalledCounter ticketAdditionCounterLong = new IntervalledCounter((long)(1.0e6 * 1000.0)); // 1000ms
+
+ public long lastChunkLoad;
+
+ public PlayerLoaderData(final ServerPlayer player, final PlayerChunkLoader loader) {
+ this.player = player;
+ this.loader = loader;
+ }
+
+ // these view distance methods are for api
+ public int getTargetSendViewDistance() {
+ final int tickViewDistance = this.tickViewDistance == -1 ? this.loader.getTickDistance() : this.tickViewDistance;
+ final int loadViewDistance = Math.max(tickViewDistance + 1, this.loadViewDistance == -1 ? this.loader.getLoadDistance() : this.loadViewDistance);
+ final int clientViewDistance = this.getClientViewDistance();
+ final int sendViewDistance = Math.min(loadViewDistance, this.sendViewDistance == -1 ? (!GlobalConfiguration.get().chunkLoading.autoconfigSendDistance || clientViewDistance == -1 ? this.loader.getSendDistance() : clientViewDistance + 1) : this.sendViewDistance);
+ return sendViewDistance;
+ }
+
+ public void setTargetSendViewDistance(final int distance) {
+ if (distance != -1 && (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE + 1)) {
+ throw new IllegalArgumentException("Send view distance must be a number between " + MIN_VIEW_DISTANCE + " and " + (MAX_VIEW_DISTANCE + 1) + " or -1, got: " + distance);
+ }
+ this.sendViewDistance = distance;
+ }
+
+ public int getTargetNoTickViewDistance() {
+ return (this.loadViewDistance == -1 ? this.getLoadDistance() : this.loadViewDistance) - 1;
+ }
+
+ public void setTargetNoTickViewDistance(final int distance) {
+ if (distance != -1 && (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE)) {
+ throw new IllegalArgumentException("Simulation distance must be a number between " + MIN_VIEW_DISTANCE + " and " + MAX_VIEW_DISTANCE + " or -1, got: " + distance);
+ }
+ this.loadViewDistance = distance == -1 ? -1 : distance + 1;
+ }
+
+ public int getTargetTickViewDistance() {
+ return this.tickViewDistance == -1 ? this.loader.getTickDistance() : this.tickViewDistance;
+ }
+
+ public void setTargetTickViewDistance(final int distance) {
+ if (distance != -1 && (distance < MIN_VIEW_DISTANCE || distance > MAX_VIEW_DISTANCE)) {
+ throw new IllegalArgumentException("View distance must be a number between " + MIN_VIEW_DISTANCE + " and " + MAX_VIEW_DISTANCE + " or -1, got: " + distance);
+ }
+ this.tickViewDistance = distance;
+ }
+
+ protected int getLoadDistance() {
+ final int tickViewDistance = this.tickViewDistance == -1 ? this.loader.getTickDistance() : this.tickViewDistance;
+
+ return Math.max(tickViewDistance + 1, this.loadViewDistance == -1 ? this.loader.getLoadDistance() : this.loadViewDistance);
+ }
+
+ public boolean hasSentChunk(final int chunkX, final int chunkZ) {
+ return this.sentChunks.contains(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+
+ public void sendChunk(final int chunkX, final int chunkZ, final Runnable onChunkSend) {
+ if (this.sentChunks.add(CoordinateUtils.getChunkKey(chunkX, chunkZ))) {
+ this.player.getLevel().getChunkSource().chunkMap.updateChunkTracking(this.player,
+ new ChunkPos(chunkX, chunkZ), new MutableObject<>(), false, true); // unloaded, loaded
+ this.player.connection.connection.execute(onChunkSend);
+ } else {
+ throw new IllegalStateException();
+ }
+ }
+
+ public void unloadChunk(final int chunkX, final int chunkZ) {
+ if (this.sentChunks.remove(CoordinateUtils.getChunkKey(chunkX, chunkZ))) {
+ this.player.getLevel().getChunkSource().chunkMap.updateChunkTracking(this.player,
+ new ChunkPos(chunkX, chunkZ), null, true, false); // unloaded, loaded
+ }
+ }
+
+ protected static boolean wantChunkLoaded(final int centerX, final int centerZ, final int chunkX, final int chunkZ,
+ final int sendRadius) {
+ // expect sendRadius to be = 1 + target viewable radius
+ return ChunkMap.isChunkInRange(chunkX, chunkZ, centerX, centerZ, sendRadius);
+ }
+
+ protected static boolean triangleIntersects(final double p1x, final double p1z, // triangle point
+ final double p2x, final double p2z, // triangle point
+ final double p3x, final double p3z, // triangle point
+
+ final double targetX, final double targetZ) { // point
+ // from barycentric coordinates:
+ // targetX = a*p1x + b*p2x + c*p3x
+ // targetZ = a*p1z + b*p2z + c*p3z
+ // 1.0 = a*1.0 + b*1.0 + c*1.0
+ // where a, b, c >= 0.0
+ // so, if any of a, b, c are less-than zero then there is no intersection.
+
+ // d = ((p2z - p3z)(p1x - p3x) + (p3x - p2x)(p1z - p3z))
+ // a = ((p2z - p3z)(targetX - p3x) + (p3x - p2x)(targetZ - p3z)) / d
+ // b = ((p3z - p1z)(targetX - p3x) + (p1x - p3x)(targetZ - p3z)) / d
+ // c = 1.0 - a - b
+
+ final double d = (p2z - p3z)*(p1x - p3x) + (p3x - p2x)*(p1z - p3z);
+ final double a = ((p2z - p3z)*(targetX - p3x) + (p3x - p2x)*(targetZ - p3z)) / d;
+
+ if (a < 0.0 || a > 1.0) {
+ return false;
+ }
+
+ final double b = ((p3z - p1z)*(targetX - p3x) + (p1x - p3x)*(targetZ - p3z)) / d;
+ if (b < 0.0 || b > 1.0) {
+ return false;
+ }
+
+ final double c = 1.0 - a - b;
+
+ return c >= 0.0 && c <= 1.0;
+ }
+
+ public void remove() {
+ this.loader.broadcastMap.remove(this.player);
+ this.loader.loadMap.remove(this.player);
+ this.loader.loadTicketCleanup.remove(this.player);
+ this.loader.tickMap.remove(this.player);
+ }
+
+ protected int getClientViewDistance() {
+ return this.player.clientViewDistance == null ? -1 : Math.max(0, this.player.clientViewDistance.intValue());
+ }
+
+ public void update() {
+ final int tickViewDistance = this.tickViewDistance == -1 ? this.loader.getTickDistance() : this.tickViewDistance;
+ // load view cannot be less-than tick view + 1
+ final int loadViewDistance = Math.max(tickViewDistance + 1, this.loadViewDistance == -1 ? this.loader.getLoadDistance() : this.loadViewDistance);
+ // send view cannot be greater-than load view
+ final int clientViewDistance = this.getClientViewDistance();
+ final int sendViewDistance = Math.min(loadViewDistance, this.sendViewDistance == -1 ? (!GlobalConfiguration.get().chunkLoading.autoconfigSendDistance || clientViewDistance == -1 ? this.loader.getSendDistance() : clientViewDistance + 1) : this.sendViewDistance);
+
+ final double posX = this.player.getX();
+ final double posZ = this.player.getZ();
+ final float yaw = MCUtil.normalizeYaw(this.player.getYRot() + 90.0f); // mc yaw 0 is along the positive z axis, but obviously this is really dumb - offset so we are at positive x-axis
+
+ // in general, we really only want to prioritise chunks in front if we know we're moving pretty fast into them.
+ final boolean useLookPriority = GlobalConfiguration.get().chunkLoading.enableFrustumPriority && (this.player.getDeltaMovement().horizontalDistanceSqr() > LOOK_PRIORITY_SPEED_THRESHOLD ||
+ this.player.getAbilities().flying);
+
+ // make sure we're in the send queue
+ this.loader.chunkSendWaitQueue.add(this);
+
+ if (
+ // has view distance stayed the same?
+ sendViewDistance == this.lastSendDistance
+ && loadViewDistance == this.lastLoadDistance
+ && tickViewDistance == this.lastTickDistance
+
+ && (this.usingLookingPriority ? (
+ // has our block stayed the same (this also accounts for chunk change)?
+ Mth.floor(this.lastLocX) == Mth.floor(posX)
+ && Mth.floor(this.lastLocZ) == Mth.floor(posZ)
+ ) : (
+ // has our chunk stayed the same
+ (Mth.floor(this.lastLocX) >> 4) == (Mth.floor(posX) >> 4)
+ && (Mth.floor(this.lastLocZ) >> 4) == (Mth.floor(posZ) >> 4)
+ ))
+
+ // has our decision about look priority changed?
+ && this.usingLookingPriority == useLookPriority
+
+ // if we are currently using look priority, has our yaw stayed within recalc threshold?
+ && (!this.usingLookingPriority || Math.abs(yaw - this.lastYaw) <= LOOK_PRIORITY_YAW_DELTA_RECALC_THRESHOLD)
+ ) {
+ // nothing we care about changed, so we're not re-calculating
+ return;
+ }
+
+ final int centerChunkX = Mth.floor(posX) >> 4;
+ final int centerChunkZ = Mth.floor(posZ) >> 4;
+
+ final boolean needsChunkCenterUpdate = (centerChunkX != this.lastChunkX) || (centerChunkZ != this.lastChunkZ);
+ this.loader.broadcastMap.addOrUpdate(this.player, centerChunkX, centerChunkZ, sendViewDistance);
+ this.loader.loadMap.addOrUpdate(this.player, centerChunkX, centerChunkZ, loadViewDistance);
+ this.loader.loadTicketCleanup.addOrUpdate(this.player, centerChunkX, centerChunkZ, loadViewDistance + 1);
+ this.loader.tickMap.addOrUpdate(this.player, centerChunkX, centerChunkZ, tickViewDistance);
+
+ if (sendViewDistance != this.lastSendDistance) {
+ // update the view radius for client
+ // note that this should be after the map calls because the client wont expect unload calls not in its VD
+ // and it's possible we decreased VD here
+ this.player.connection.send(new ClientboundSetChunkCacheRadiusPacket(sendViewDistance));
+ }
+ if (tickViewDistance != this.lastTickDistance) {
+ this.player.connection.send(new ClientboundSetSimulationDistancePacket(tickViewDistance));
+ }
+
+ this.lastLocX = posX;
+ this.lastLocZ = posZ;
+ this.lastYaw = yaw;
+ this.lastSendDistance = sendViewDistance;
+ this.lastLoadDistance = loadViewDistance;
+ this.lastTickDistance = tickViewDistance;
+ this.usingLookingPriority = useLookPriority;
+
+ this.lastChunkX = centerChunkX;
+ this.lastChunkZ = centerChunkZ;
+
+ // points for player "view" triangle:
+
+ // obviously, the player pos is a vertex
+ final double p1x = posX;
+ final double p1z = posZ;
+
+ // to the left of the looking direction
+ final double p2x = PRIORITISED_DISTANCE * Math.cos(Math.toRadians(yaw + (double)(FOV / 2.0))) // calculate rotated vector
+ + p1x; // offset vector
+ final double p2z = PRIORITISED_DISTANCE * Math.sin(Math.toRadians(yaw + (double)(FOV / 2.0))) // calculate rotated vector
+ + p1z; // offset vector
+
+ // to the right of the looking direction
+ final double p3x = PRIORITISED_DISTANCE * Math.cos(Math.toRadians(yaw - (double)(FOV / 2.0))) // calculate rotated vector
+ + p1x; // offset vector
+ final double p3z = PRIORITISED_DISTANCE * Math.sin(Math.toRadians(yaw - (double)(FOV / 2.0))) // calculate rotated vector
+ + p1z; // offset vector
+
+ // now that we have all of our points, we can recalculate the load queue
+
+ final List<ChunkPriorityHolder> loadQueue = new ArrayList<>();
+
+ // clear send queue, we are re-sorting
+ this.sendQueue.clear();
+ // clear chunk want set, vd/position might have changed
+ this.chunksToBeSent.clear();
+
+ final int searchViewDistance = Math.max(loadViewDistance, sendViewDistance);
+
+ for (int dx = -searchViewDistance; dx <= searchViewDistance; ++dx) {
+ for (int dz = -searchViewDistance; dz <= searchViewDistance; ++dz) {
+ final int chunkX = dx + centerChunkX;
+ final int chunkZ = dz + centerChunkZ;
+ final int squareDistance = Math.max(Math.abs(dx), Math.abs(dz));
+ final boolean sendChunk = squareDistance <= sendViewDistance && wantChunkLoaded(centerChunkX, centerChunkZ, chunkX, chunkZ, sendViewDistance);
+
+ if (this.hasSentChunk(chunkX, chunkZ)) {
+ // already sent (which means it is also loaded)
+ if (!sendChunk) {
+ // have sent the chunk, but don't want it anymore
+ // unload it now
+ this.unloadChunk(chunkX, chunkZ);
+ }
+ continue;
+ }
+
+ final boolean loadChunk = squareDistance <= loadViewDistance;
+
+ final boolean prioritised = useLookPriority && triangleIntersects(
+ // prioritisation triangle
+ p1x, p1z, p2x, p2z, p3x, p3z,
+
+ // center of chunk
+ (double)((chunkX << 4) | 8), (double)((chunkZ << 4) | 8)
+ );
+
+ final int manhattanDistance = Math.abs(dx) + Math.abs(dz);
+
+ final double priority;
+
+ if (squareDistance <= GlobalConfiguration.get().chunkLoading.minLoadRadius) {
+ // priority should be negative, and we also want to order it from center outwards
+ // so we want (0,0) to be the smallest, and (minLoadedRadius,minLoadedRadius) to be the greatest
+ priority = -((2 * GlobalConfiguration.get().chunkLoading.minLoadRadius + 1) - manhattanDistance);
+ } else {
+ if (prioritised) {
+ // we don't prioritise these chunks above others because we also want to make sure some chunks
+ // will be loaded if the player changes direction
+ priority = (double)manhattanDistance / 6.0;
+ } else {
+ priority = (double)manhattanDistance;
+ }
+ }
+
+ final ChunkPriorityHolder holder = new ChunkPriorityHolder(chunkX, chunkZ, manhattanDistance, priority);
+
+ if (!this.loader.isChunkPlayerLoaded(chunkX, chunkZ)) {
+ if (loadChunk) {
+ loadQueue.add(holder);
+ if (sendChunk) {
+ this.chunksToBeSent.add(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+ }
+ } else {
+ // loaded but not sent: so queue it!
+ if (sendChunk) {
+ this.sendQueue.add(holder);
+ }
+ }
+ }
+ }
+
+ loadQueue.sort((final ChunkPriorityHolder p1, final ChunkPriorityHolder p2) -> {
+ return Double.compare(p1.priority, p2.priority);
+ });
+
+ // we're modifying loadQueue, must remove
+ this.loader.chunkLoadQueue.remove(this);
+
+ this.loadQueue.clear();
+ this.loadQueue.addAll(loadQueue);
+
+ // must re-add
+ this.loader.chunkLoadQueue.add(this);
+
+ // update the chunk center
+ // this must be done last so that the client does not ignore any of our unload chunk packets
+ if (needsChunkCenterUpdate) {
+ this.player.connection.send(new ClientboundSetChunkCacheCenterPacket(centerChunkX, centerChunkZ));
+ }
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/ChunkSystem.java b/src/main/java/io/papermc/paper/chunk/system/ChunkSystem.java
index 8a5e93961dac4d87c81c0e70b6f4124a1f1d2556..0dc94dec1317b3f86d38074c6cbe41ab828cab1d 100644
--- a/src/main/java/io/papermc/paper/chunk/system/ChunkSystem.java
+++ b/src/main/java/io/papermc/paper/chunk/system/ChunkSystem.java
@@ -31,191 +31,41 @@ public final class ChunkSystem {
}
public static void scheduleChunkTask(final ServerLevel level, final int chunkX, final int chunkZ, final Runnable run, final PrioritisedExecutor.Priority priority) {
- level.chunkSource.mainThreadProcessor.execute(run);
+ level.chunkTaskScheduler.scheduleChunkTask(chunkX, chunkZ, run, priority); // Paper - rewrite chunk system
}
public static void scheduleChunkLoad(final ServerLevel level, final int chunkX, final int chunkZ, final boolean gen,
final ChunkStatus toStatus, final boolean addTicket, final PrioritisedExecutor.Priority priority,
final Consumer<ChunkAccess> onComplete) {
- if (gen) {
- scheduleChunkLoad(level, chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
- return;
- }
- scheduleChunkLoad(level, chunkX, chunkZ, ChunkStatus.EMPTY, addTicket, priority, (final ChunkAccess chunk) -> {
- if (chunk == null) {
- onComplete.accept(null);
- } else {
- if (chunk.getStatus().isOrAfter(toStatus)) {
- scheduleChunkLoad(level, chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
- } else {
- onComplete.accept(null);
- }
- }
- });
+ level.chunkTaskScheduler.scheduleChunkLoad(chunkX, chunkZ, gen, toStatus, addTicket, priority, onComplete); // Paper - rewrite chunk system
}
- static final TicketType<Long> CHUNK_LOAD = TicketType.create("chunk_load", Long::compareTo);
-
- private static long chunkLoadCounter = 0L;
+ // Paper - rewrite chunk system
public static void scheduleChunkLoad(final ServerLevel level, final int chunkX, final int chunkZ, final ChunkStatus toStatus,
final boolean addTicket, final PrioritisedExecutor.Priority priority, final Consumer<ChunkAccess> onComplete) {
- if (!Bukkit.isPrimaryThread()) {
- scheduleChunkTask(level, chunkX, chunkZ, () -> {
- scheduleChunkLoad(level, chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
- }, priority);
- return;
- }
-
- final int minLevel = 33 + ChunkStatus.getDistance(toStatus);
- final Long chunkReference = addTicket ? Long.valueOf(++chunkLoadCounter) : null;
- final ChunkPos chunkPos = new ChunkPos(chunkX, chunkZ);
-
- if (addTicket) {
- level.chunkSource.addTicketAtLevel(CHUNK_LOAD, chunkPos, minLevel, chunkReference);
- }
- level.chunkSource.runDistanceManagerUpdates();
-
- final Consumer<ChunkAccess> loadCallback = (final ChunkAccess chunk) -> {
- try {
- if (onComplete != null) {
- onComplete.accept(chunk);
- }
- } catch (final ThreadDeath death) {
- throw death;
- } catch (final Throwable thr) {
- LOGGER.error("Exception handling chunk load callback", thr);
- SneakyThrow.sneaky(thr);
- } finally {
- if (addTicket) {
- level.chunkSource.addTicketAtLevel(TicketType.UNKNOWN, chunkPos, minLevel, chunkPos);
- level.chunkSource.removeTicketAtLevel(CHUNK_LOAD, chunkPos, minLevel, chunkReference);
- }
- }
- };
-
- final ChunkHolder holder = level.chunkSource.chunkMap.getUpdatingChunkIfPresent(CoordinateUtils.getChunkKey(chunkX, chunkZ));
-
- if (holder == null || holder.getTicketLevel() > minLevel) {
- loadCallback.accept(null);
- return;
- }
-
- final CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> loadFuture = holder.getOrScheduleFuture(toStatus, level.chunkSource.chunkMap);
-
- if (loadFuture.isDone()) {
- loadCallback.accept(loadFuture.join().left().orElse(null));
- return;
- }
-
- loadFuture.whenCompleteAsync((final Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either, final Throwable thr) -> {
- if (thr != null) {
- loadCallback.accept(null);
- return;
- }
- loadCallback.accept(either.left().orElse(null));
- }, (final Runnable r) -> {
- scheduleChunkTask(level, chunkX, chunkZ, r, PrioritisedExecutor.Priority.HIGHEST);
- });
+ level.chunkTaskScheduler.scheduleChunkLoad(chunkX, chunkZ, toStatus, addTicket, priority, onComplete); // Paper - rewrite chunk system
}
public static void scheduleTickingState(final ServerLevel level, final int chunkX, final int chunkZ,
final ChunkHolder.FullChunkStatus toStatus, final boolean addTicket,
final PrioritisedExecutor.Priority priority, final Consumer<LevelChunk> onComplete) {
- if (toStatus == ChunkHolder.FullChunkStatus.INACCESSIBLE) {
- throw new IllegalArgumentException("Cannot wait for INACCESSIBLE status");
- }
-
- if (!Bukkit.isPrimaryThread()) {
- scheduleChunkTask(level, chunkX, chunkZ, () -> {
- scheduleTickingState(level, chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
- }, priority);
- return;
- }
-
- final int minLevel = 33 - (toStatus.ordinal() - 1);
- final int radius = toStatus.ordinal() - 1;
- final Long chunkReference = addTicket ? Long.valueOf(++chunkLoadCounter) : null;
- final ChunkPos chunkPos = new ChunkPos(chunkX, chunkZ);
-
- if (addTicket) {
- level.chunkSource.addTicketAtLevel(CHUNK_LOAD, chunkPos, minLevel, chunkReference);
- }
- level.chunkSource.runDistanceManagerUpdates();
-
- final Consumer<LevelChunk> loadCallback = (final LevelChunk chunk) -> {
- try {
- if (onComplete != null) {
- onComplete.accept(chunk);
- }
- } catch (final ThreadDeath death) {
- throw death;
- } catch (final Throwable thr) {
- LOGGER.error("Exception handling chunk load callback", thr);
- SneakyThrow.sneaky(thr);
- } finally {
- if (addTicket) {
- level.chunkSource.addTicketAtLevel(TicketType.UNKNOWN, chunkPos, minLevel, chunkPos);
- level.chunkSource.removeTicketAtLevel(CHUNK_LOAD, chunkPos, minLevel, chunkReference);
- }
- }
- };
-
- final ChunkHolder holder = level.chunkSource.chunkMap.getUpdatingChunkIfPresent(CoordinateUtils.getChunkKey(chunkX, chunkZ));
-
- if (holder == null || holder.getTicketLevel() > minLevel) {
- loadCallback.accept(null);
- return;
- }
-
- final CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> tickingState;
- switch (toStatus) {
- case BORDER: {
- tickingState = holder.getFullChunkFuture();
- break;
- }
- case TICKING: {
- tickingState = holder.getTickingChunkFuture();
- break;
- }
- case ENTITY_TICKING: {
- tickingState = holder.getEntityTickingChunkFuture();
- break;
- }
- default: {
- throw new IllegalStateException("Cannot reach here");
- }
- }
-
- if (tickingState.isDone()) {
- loadCallback.accept(tickingState.join().left().orElse(null));
- return;
- }
-
- tickingState.whenCompleteAsync((final Either<LevelChunk, ChunkHolder.ChunkLoadingFailure> either, final Throwable thr) -> {
- if (thr != null) {
- loadCallback.accept(null);
- return;
- }
- loadCallback.accept(either.left().orElse(null));
- }, (final Runnable r) -> {
- scheduleChunkTask(level, chunkX, chunkZ, r, PrioritisedExecutor.Priority.HIGHEST);
- });
+ level.chunkTaskScheduler.scheduleTickingState(chunkX, chunkZ, toStatus, addTicket, priority, onComplete); // Paper - rewrite chunk system
}
public static List<ChunkHolder> getVisibleChunkHolders(final ServerLevel level) {
- return new ArrayList<>(level.chunkSource.chunkMap.visibleChunkMap.values());
+ return level.chunkTaskScheduler.chunkHolderManager.getOldChunkHolders(); // Paper - rewrite chunk system
}
public static List<ChunkHolder> getUpdatingChunkHolders(final ServerLevel level) {
- return new ArrayList<>(level.chunkSource.chunkMap.updatingChunkMap.values());
+ return level.chunkTaskScheduler.chunkHolderManager.getOldChunkHolders(); // Paper - rewrite chunk system
}
public static int getVisibleChunkHolderCount(final ServerLevel level) {
- return level.chunkSource.chunkMap.visibleChunkMap.size();
+ return level.chunkTaskScheduler.chunkHolderManager.size(); // Paper - rewrite chunk system
}
public static int getUpdatingChunkHolderCount(final ServerLevel level) {
- return level.chunkSource.chunkMap.updatingChunkMap.size();
+ return level.chunkTaskScheduler.chunkHolderManager.size(); // Paper - rewrite chunk system
}
public static boolean hasAnyChunkHolders(final ServerLevel level) {
@@ -269,23 +119,15 @@ public final class ChunkSystem {
}
public static int getSendViewDistance(final ServerPlayer player) {
- return getLoadViewDistance(player);
+ return io.papermc.paper.chunk.PlayerChunkLoader.getSendViewDistance(player);
}
public static int getLoadViewDistance(final ServerPlayer player) {
- final ServerLevel level = player.getLevel();
- if (level == null) {
- return Bukkit.getViewDistance() + 1;
- }
- return level.chunkSource.chunkMap.getEffectiveViewDistance() + 1;
+ return io.papermc.paper.chunk.PlayerChunkLoader.getLoadViewDistance(player);
}
public static int getTickViewDistance(final ServerPlayer player) {
- final ServerLevel level = player.getLevel();
- if (level == null) {
- return Bukkit.getSimulationDistance();
- }
- return level.chunkSource.chunkMap.distanceManager.getSimulationDistance();
+ return io.papermc.paper.chunk.PlayerChunkLoader.getTickViewDistance(player);
}
private ChunkSystem() {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
diff --git a/src/main/java/io/papermc/paper/chunk/system/entity/EntityLookup.java b/src/main/java/io/papermc/paper/chunk/system/entity/EntityLookup.java
new file mode 100644
2022-12-08 05:24:59 +01:00
index 0000000000000000000000000000000000000000..61c170555c8854b102c640b0b6a615f9f732edbf
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/entity/EntityLookup.java
@@ -0,0 +1,839 @@
+package io.papermc.paper.chunk.system.entity;
+
+import com.destroystokyo.paper.util.maplist.EntityList;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import io.papermc.paper.util.WorldUtil;
+import io.papermc.paper.world.ChunkEntitySlices;
+import it.unimi.dsi.fastutil.ints.Int2ReferenceOpenHashMap;
+import it.unimi.dsi.fastutil.longs.Long2ObjectOpenHashMap;
+import it.unimi.dsi.fastutil.objects.Object2ReferenceOpenHashMap;
+import net.minecraft.core.BlockPos;
+import io.papermc.paper.chunk.system.ChunkSystem;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ServerLevel;
2022-12-08 05:24:59 +01:00
+import net.minecraft.util.AbortableIterationConsumer;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.util.Mth;
+import net.minecraft.world.entity.Entity;
+import net.minecraft.world.entity.EntityType;
+import net.minecraft.world.level.entity.EntityInLevelCallback;
+import net.minecraft.world.level.entity.EntityTypeTest;
+import net.minecraft.world.level.entity.LevelCallback;
+import net.minecraft.world.level.entity.LevelEntityGetter;
+import net.minecraft.world.level.entity.Visibility;
+import net.minecraft.world.phys.AABB;
+import org.jetbrains.annotations.NotNull;
+import org.jetbrains.annotations.Nullable;
+import org.slf4j.Logger;
+import java.util.ArrayList;
+import java.util.Iterator;
+import java.util.List;
+import java.util.NoSuchElementException;
+import java.util.UUID;
+import java.util.concurrent.locks.StampedLock;
+import java.util.function.Consumer;
+import java.util.function.Predicate;
+
+public final class EntityLookup implements LevelEntityGetter<Entity> {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ protected static final int REGION_SHIFT = 5;
+ protected static final int REGION_MASK = (1 << REGION_SHIFT) - 1;
+ protected static final int REGION_SIZE = 1 << REGION_SHIFT;
+
+ public final ServerLevel world;
+
+ private final StampedLock stateLock = new StampedLock();
+ protected final Long2ObjectOpenHashMap<ChunkSlicesRegion> regions = new Long2ObjectOpenHashMap<>(128, 0.5f);
+
+ private final int minSection; // inclusive
+ private final int maxSection; // inclusive
+ private final LevelCallback<Entity> worldCallback;
+
+ private final StampedLock entityByLock = new StampedLock();
+ private final Int2ReferenceOpenHashMap<Entity> entityById = new Int2ReferenceOpenHashMap<>();
+ private final Object2ReferenceOpenHashMap<UUID, Entity> entityByUUID = new Object2ReferenceOpenHashMap<>();
+ private final EntityList accessibleEntities = new EntityList();
+
+ public EntityLookup(final ServerLevel world, final LevelCallback<Entity> worldCallback) {
+ this.world = world;
+ this.minSection = WorldUtil.getMinSection(world);
+ this.maxSection = WorldUtil.getMaxSection(world);
+ this.worldCallback = worldCallback;
+ }
+
+ private static Entity maskNonAccessible(final Entity entity) {
+ if (entity == null) {
+ return null;
+ }
+ final Visibility visibility = EntityLookup.getEntityStatus(entity);
+ return visibility.isAccessible() ? entity : null;
+ }
+
+ @Nullable
+ @Override
+ public Entity get(final int id) {
+ final long attempt = this.entityByLock.tryOptimisticRead();
+ if (attempt != 0L) {
+ try {
+ final Entity ret = this.entityById.get(id);
+
+ if (this.entityByLock.validate(attempt)) {
+ return maskNonAccessible(ret);
+ }
+ } catch (final Error error) {
+ throw error;
+ } catch (final Throwable thr) {
+ // ignore
+ }
+ }
+
+ this.entityByLock.readLock();
+ try {
+ return maskNonAccessible(this.entityById.get(id));
+ } finally {
+ this.entityByLock.tryUnlockRead();
+ }
+ }
+
+ @Nullable
+ @Override
+ public Entity get(final UUID id) {
+ final long attempt = this.entityByLock.tryOptimisticRead();
+ if (attempt != 0L) {
+ try {
+ final Entity ret = this.entityByUUID.get(id);
+
+ if (this.entityByLock.validate(attempt)) {
+ return maskNonAccessible(ret);
+ }
+ } catch (final Error error) {
+ throw error;
+ } catch (final Throwable thr) {
+ // ignore
+ }
+ }
+
+ this.entityByLock.readLock();
+ try {
+ return maskNonAccessible(this.entityByUUID.get(id));
+ } finally {
+ this.entityByLock.tryUnlockRead();
+ }
+ }
+
+ public boolean hasEntity(final UUID uuid) {
+ return this.get(uuid) != null;
+ }
+
+ public String getDebugInfo() {
+ return "count_id:" + this.entityById.size() + ",count_uuid:" + this.entityByUUID.size() + ",region_count:" + this.regions.size();
+ }
+
+ static final class ArrayIterable<T> implements Iterable<T> {
+
+ private final T[] array;
+ private final int off;
+ private final int length;
+
+ public ArrayIterable(final T[] array, final int off, final int length) {
+ this.array = array;
+ this.off = off;
+ this.length = length;
+ if (length > array.length) {
+ throw new IllegalArgumentException("Length must be no greater-than the array length");
+ }
+ }
+
+ @NotNull
+ @Override
+ public Iterator<T> iterator() {
+ return new ArrayIterator<>(this.array, this.off, this.length);
+ }
+
+ static final class ArrayIterator<T> implements Iterator<T> {
+
+ private final T[] array;
+ private int off;
+ private final int length;
+
+ public ArrayIterator(final T[] array, final int off, final int length) {
+ this.array = array;
+ this.off = off;
+ this.length = length;
+ }
+
+ @Override
+ public boolean hasNext() {
+ return this.off < this.length;
+ }
+
+ @Override
+ public T next() {
+ if (this.off >= this.length) {
+ throw new NoSuchElementException();
+ }
+ return this.array[this.off++];
+ }
+
+ @Override
+ public void remove() {
+ throw new UnsupportedOperationException();
+ }
+ }
+ }
+
+ @Override
+ public Iterable<Entity> getAll() {
+ return new ArrayIterable<>(this.accessibleEntities.getRawData(), 0, this.accessibleEntities.size());
+ }
+
+ @Override
2022-12-08 05:24:59 +01:00
+ public <U extends Entity> void get(final EntityTypeTest<Entity, U> filter, final AbortableIterationConsumer<U> action) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ for (final Entity entity : this.entityById.values()) {
+ final Visibility visibility = EntityLookup.getEntityStatus(entity);
+ if (!visibility.isAccessible()) {
+ continue;
+ }
+ final U casted = filter.tryCast(entity);
2022-12-08 05:24:59 +01:00
+ if (casted != null && action.accept(casted).shouldAbort()) {
+ break;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+ }
+ }
+
+ @Override
+ public void get(final AABB box, final Consumer<Entity> action) {
+ List<Entity> entities = new ArrayList<>();
+ this.getEntitiesWithoutDragonParts(null, box, entities, null);
+ for (int i = 0, len = entities.size(); i < len; ++i) {
+ action.accept(entities.get(i));
+ }
+ }
+
+ @Override
2022-12-08 05:24:59 +01:00
+ public <U extends Entity> void get(final EntityTypeTest<Entity, U> filter, final AABB box, final AbortableIterationConsumer<U> action) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ List<Entity> entities = new ArrayList<>();
+ this.getEntitiesWithoutDragonParts(null, box, entities, null);
+ for (int i = 0, len = entities.size(); i < len; ++i) {
+ final U casted = filter.tryCast(entities.get(i));
2022-12-08 05:24:59 +01:00
+ if (casted != null && action.accept(casted).shouldAbort()) {
+ break;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+ }
+ }
+
+ public void entityStatusChange(final Entity entity, final ChunkEntitySlices slices, final Visibility oldVisibility, final Visibility newVisibility, final boolean moved,
+ final boolean created, final boolean destroyed) {
+ TickThread.ensureTickThread(entity, "Entity status change must only happen on the main thread");
+
+ if (entity.updatingSectionStatus) {
+ // recursive status update
+ LOGGER.error("Cannot recursively update entity chunk status for entity " + entity, new Throwable());
+ return;
+ }
+
+ final boolean entityStatusUpdateBefore = slices == null ? false : slices.startPreventingStatusUpdates();
+
+ if (entityStatusUpdateBefore) {
+ LOGGER.error("Cannot update chunk status for entity " + entity + " since entity chunk (" + slices.chunkX + "," + slices.chunkZ + ") is receiving update", new Throwable());
+ return;
+ }
+
+ try {
+ final Boolean ticketBlockBefore = this.world.chunkTaskScheduler.chunkHolderManager.blockTicketUpdates();
+ try {
+ entity.updatingSectionStatus = true;
+ try {
+ if (created) {
+ EntityLookup.this.worldCallback.onCreated(entity);
+ }
+
+ if (oldVisibility == newVisibility) {
+ if (moved && newVisibility.isAccessible()) {
+ EntityLookup.this.worldCallback.onSectionChange(entity);
+ }
+ return;
+ }
+
+ if (newVisibility.ordinal() > oldVisibility.ordinal()) {
+ // status upgrade
+ if (!oldVisibility.isAccessible() && newVisibility.isAccessible()) {
+ this.accessibleEntities.add(entity);
+ EntityLookup.this.worldCallback.onTrackingStart(entity);
+ }
+
+ if (!oldVisibility.isTicking() && newVisibility.isTicking()) {
+ EntityLookup.this.worldCallback.onTickingStart(entity);
+ }
+ } else {
+ // status downgrade
+ if (oldVisibility.isTicking() && !newVisibility.isTicking()) {
+ EntityLookup.this.worldCallback.onTickingEnd(entity);
+ }
+
+ if (oldVisibility.isAccessible() && !newVisibility.isAccessible()) {
+ this.accessibleEntities.remove(entity);
+ EntityLookup.this.worldCallback.onTrackingEnd(entity);
+ }
+ }
+
+ if (moved && newVisibility.isAccessible()) {
+ EntityLookup.this.worldCallback.onSectionChange(entity);
+ }
+
+ if (destroyed) {
+ EntityLookup.this.worldCallback.onDestroyed(entity);
+ }
+ } finally {
+ entity.updatingSectionStatus = false;
+ }
+ } finally {
+ this.world.chunkTaskScheduler.chunkHolderManager.unblockTicketUpdates(ticketBlockBefore);
+ }
+ } finally {
+ if (slices != null) {
+ slices.stopPreventingStatusUpdates(false);
+ }
+ }
+ }
+
+ public void chunkStatusChange(final int x, final int z, final ChunkHolder.FullChunkStatus newStatus) {
+ this.getChunk(x, z).updateStatus(newStatus, this);
+ }
+
+ public void addLegacyChunkEntities(final List<Entity> entities) {
+ for (int i = 0, len = entities.size(); i < len; ++i) {
+ this.addEntity(entities.get(i), true);
+ }
+ }
+
+ public void addEntityChunkEntities(final List<Entity> entities) {
+ for (int i = 0, len = entities.size(); i < len; ++i) {
+ this.addEntity(entities.get(i), true);
+ }
+ }
+
+ public void addWorldGenChunkEntities(final List<Entity> entities) {
+ for (int i = 0, len = entities.size(); i < len; ++i) {
+ this.addEntity(entities.get(i), false);
+ }
+ }
+
+ public boolean addNewEntity(final Entity entity) {
+ return this.addEntity(entity, false);
+ }
+
+ public static Visibility getEntityStatus(final Entity entity) {
+ if (entity.isAlwaysTicking()) {
+ return Visibility.TICKING;
+ }
+ final ChunkHolder.FullChunkStatus entityStatus = entity.chunkStatus;
+ return Visibility.fromFullChunkStatus(entityStatus == null ? ChunkHolder.FullChunkStatus.INACCESSIBLE : entityStatus);
+ }
+
+ private boolean addEntity(final Entity entity, final boolean fromDisk) {
+ final BlockPos pos = entity.blockPosition();
+ final int sectionX = pos.getX() >> 4;
+ final int sectionY = Mth.clamp(pos.getY() >> 4, this.minSection, this.maxSection);
+ final int sectionZ = pos.getZ() >> 4;
+ TickThread.ensureTickThread(this.world, sectionX, sectionZ, "Cannot add entity off-main thread");
+
+ if (entity.isRemoved()) {
+ LOGGER.warn("Refusing to add removed entity: " + entity);
+ return false;
+ }
+
+ if (entity.updatingSectionStatus) {
+ LOGGER.warn("Entity " + entity + " is currently prevented from being added/removed to world since it is processing section status updates", new Throwable());
+ return false;
+ }
+
+ if (fromDisk) {
+ ChunkSystem.onEntityPreAdd(this.world, entity);
+ if (entity.isRemoved()) {
+ // removed from checkDupeUUID call
+ return false;
+ }
+ }
+
+ this.entityByLock.writeLock();
+ try {
+ if (this.entityById.containsKey(entity.getId())) {
+ LOGGER.warn("Entity id already exists: " + entity.getId() + ", mapped to " + this.entityById.get(entity.getId()) + ", can't add " + entity);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return false;
+ }
+ if (this.entityByUUID.containsKey(entity.getUUID())) {
+ LOGGER.warn("Entity uuid already exists: " + entity.getUUID() + ", mapped to " + this.entityByUUID.get(entity.getUUID()) + ", can't add " + entity);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return false;
+ }
+ this.entityById.put(entity.getId(), entity);
+ this.entityByUUID.put(entity.getUUID(), entity);
+ } finally {
+ this.entityByLock.tryUnlockWrite();
+ }
+
+ entity.sectionX = sectionX;
+ entity.sectionY = sectionY;
+ entity.sectionZ = sectionZ;
+ final ChunkEntitySlices slices = this.getOrCreateChunk(sectionX, sectionZ);
+ if (!slices.addEntity(entity, sectionY)) {
+ LOGGER.warn("Entity " + entity + " added to world '" + this.world.getWorld().getName() + "', but was already contained in entity chunk (" + sectionX + "," + sectionZ + ")");
+ }
+
+ entity.setLevelCallback(new EntityCallback(entity));
+
+ this.entityStatusChange(entity, slices, Visibility.HIDDEN, getEntityStatus(entity), false, !fromDisk, false);
+
+ return true;
+ }
+
+ private void removeEntity(final Entity entity) {
+ final int sectionX = entity.sectionX;
+ final int sectionY = entity.sectionY;
+ final int sectionZ = entity.sectionZ;
+ TickThread.ensureTickThread(this.world, sectionX, sectionZ, "Cannot remove entity off-main");
+ if (!entity.isRemoved()) {
+ throw new IllegalStateException("Only call Entity#setRemoved to remove an entity");
+ }
+ final ChunkEntitySlices slices = this.getChunk(sectionX, sectionZ);
+ // all entities should be in a chunk
+ if (slices == null) {
+ LOGGER.warn("Cannot remove entity " + entity + " from null entity slices (" + sectionX + "," + sectionZ + ")");
+ } else {
+ if (!slices.removeEntity(entity, sectionY)) {
+ LOGGER.warn("Failed to remove entity " + entity + " from entity slices (" + sectionX + "," + sectionZ + ")");
+ }
+ }
+ entity.sectionX = entity.sectionY = entity.sectionZ = Integer.MIN_VALUE;
+
+ this.entityByLock.writeLock();
+ try {
+ if (!this.entityById.remove(entity.getId(), entity)) {
+ LOGGER.warn("Failed to remove entity " + entity + " by id, current entity mapped: " + this.entityById.get(entity.getId()));
+ }
+ if (!this.entityByUUID.remove(entity.getUUID(), entity)) {
+ LOGGER.warn("Failed to remove entity " + entity + " by uuid, current entity mapped: " + this.entityByUUID.get(entity.getUUID()));
+ }
+ } finally {
+ this.entityByLock.tryUnlockWrite();
+ }
+ }
+
+ private ChunkEntitySlices moveEntity(final Entity entity) {
+ // ensure we own the entity
+ TickThread.ensureTickThread(entity, "Cannot move entity off-main");
+
+ final BlockPos newPos = entity.blockPosition();
+ final int newSectionX = newPos.getX() >> 4;
+ final int newSectionY = Mth.clamp(newPos.getY() >> 4, this.minSection, this.maxSection);
+ final int newSectionZ = newPos.getZ() >> 4;
+
+ if (newSectionX == entity.sectionX && newSectionY == entity.sectionY && newSectionZ == entity.sectionZ) {
+ return null;
+ }
+
+ // ensure the new section is owned by this tick thread
+ TickThread.ensureTickThread(this.world, newSectionX, newSectionZ, "Cannot move entity off-main");
+
+ // ensure the old section is owned by this tick thread
+ TickThread.ensureTickThread(this.world, entity.sectionX, entity.sectionZ, "Cannot move entity off-main");
+
+ final ChunkEntitySlices old = this.getChunk(entity.sectionX, entity.sectionZ);
+ final ChunkEntitySlices slices = this.getOrCreateChunk(newSectionX, newSectionZ);
+
+ if (!old.removeEntity(entity, entity.sectionY)) {
+ LOGGER.warn("Could not remove entity " + entity + " from its old chunk section (" + entity.sectionX + "," + entity.sectionY + "," + entity.sectionZ + ") since it was not contained in the section");
+ }
+
+ if (!slices.addEntity(entity, newSectionY)) {
+ LOGGER.warn("Could not add entity " + entity + " to its new chunk section (" + newSectionX + "," + newSectionY + "," + newSectionZ + ") as it is already contained in the section");
+ }
+
+ entity.sectionX = newSectionX;
+ entity.sectionY = newSectionY;
+ entity.sectionZ = newSectionZ;
+
+ return slices;
+ }
+
+ public void getEntitiesWithoutDragonParts(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ final int minChunkX = (Mth.floor(box.minX) - 2) >> 4;
+ final int minChunkZ = (Mth.floor(box.minZ) - 2) >> 4;
+ final int maxChunkX = (Mth.floor(box.maxX) + 2) >> 4;
+ final int maxChunkZ = (Mth.floor(box.maxZ) + 2) >> 4;
+
+ final int minRegionX = minChunkX >> REGION_SHIFT;
+ final int minRegionZ = minChunkZ >> REGION_SHIFT;
+ final int maxRegionX = maxChunkX >> REGION_SHIFT;
+ final int maxRegionZ = maxChunkZ >> REGION_SHIFT;
+
+ for (int currRegionZ = minRegionZ; currRegionZ <= maxRegionZ; ++currRegionZ) {
+ final int minZ = currRegionZ == minRegionZ ? minChunkZ & REGION_MASK : 0;
+ final int maxZ = currRegionZ == maxRegionZ ? maxChunkZ & REGION_MASK : REGION_MASK;
+
+ for (int currRegionX = minRegionX; currRegionX <= maxRegionX; ++currRegionX) {
+ final ChunkSlicesRegion region = this.getRegion(currRegionX, currRegionZ);
+
+ if (region == null) {
+ continue;
+ }
+
+ final int minX = currRegionX == minRegionX ? minChunkX & REGION_MASK : 0;
+ final int maxX = currRegionX == maxRegionX ? maxChunkX & REGION_MASK : REGION_MASK;
+
+ for (int currZ = minZ; currZ <= maxZ; ++currZ) {
+ for (int currX = minX; currX <= maxX; ++currX) {
+ final ChunkEntitySlices chunk = region.get(currX | (currZ << REGION_SHIFT));
+ if (chunk == null || !chunk.status.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ continue;
+ }
+
+ chunk.getEntitiesWithoutDragonParts(except, box, into, predicate);
+ }
+ }
+ }
+ }
+ }
+
+ public void getEntities(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ final int minChunkX = (Mth.floor(box.minX) - 2) >> 4;
+ final int minChunkZ = (Mth.floor(box.minZ) - 2) >> 4;
+ final int maxChunkX = (Mth.floor(box.maxX) + 2) >> 4;
+ final int maxChunkZ = (Mth.floor(box.maxZ) + 2) >> 4;
+
+ final int minRegionX = minChunkX >> REGION_SHIFT;
+ final int minRegionZ = minChunkZ >> REGION_SHIFT;
+ final int maxRegionX = maxChunkX >> REGION_SHIFT;
+ final int maxRegionZ = maxChunkZ >> REGION_SHIFT;
+
+ for (int currRegionZ = minRegionZ; currRegionZ <= maxRegionZ; ++currRegionZ) {
+ final int minZ = currRegionZ == minRegionZ ? minChunkZ & REGION_MASK : 0;
+ final int maxZ = currRegionZ == maxRegionZ ? maxChunkZ & REGION_MASK : REGION_MASK;
+
+ for (int currRegionX = minRegionX; currRegionX <= maxRegionX; ++currRegionX) {
+ final ChunkSlicesRegion region = this.getRegion(currRegionX, currRegionZ);
+
+ if (region == null) {
+ continue;
+ }
+
+ final int minX = currRegionX == minRegionX ? minChunkX & REGION_MASK : 0;
+ final int maxX = currRegionX == maxRegionX ? maxChunkX & REGION_MASK : REGION_MASK;
+
+ for (int currZ = minZ; currZ <= maxZ; ++currZ) {
+ for (int currX = minX; currX <= maxX; ++currX) {
+ final ChunkEntitySlices chunk = region.get(currX | (currZ << REGION_SHIFT));
+ if (chunk == null || !chunk.status.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ continue;
+ }
+
+ chunk.getEntities(except, box, into, predicate);
+ }
+ }
+ }
+ }
+ }
+
+ public void getHardCollidingEntities(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ final int minChunkX = (Mth.floor(box.minX) - 2) >> 4;
+ final int minChunkZ = (Mth.floor(box.minZ) - 2) >> 4;
+ final int maxChunkX = (Mth.floor(box.maxX) + 2) >> 4;
+ final int maxChunkZ = (Mth.floor(box.maxZ) + 2) >> 4;
+
+ final int minRegionX = minChunkX >> REGION_SHIFT;
+ final int minRegionZ = minChunkZ >> REGION_SHIFT;
+ final int maxRegionX = maxChunkX >> REGION_SHIFT;
+ final int maxRegionZ = maxChunkZ >> REGION_SHIFT;
+
+ for (int currRegionZ = minRegionZ; currRegionZ <= maxRegionZ; ++currRegionZ) {
+ final int minZ = currRegionZ == minRegionZ ? minChunkZ & REGION_MASK : 0;
+ final int maxZ = currRegionZ == maxRegionZ ? maxChunkZ & REGION_MASK : REGION_MASK;
+
+ for (int currRegionX = minRegionX; currRegionX <= maxRegionX; ++currRegionX) {
+ final ChunkSlicesRegion region = this.getRegion(currRegionX, currRegionZ);
+
+ if (region == null) {
+ continue;
+ }
+
+ final int minX = currRegionX == minRegionX ? minChunkX & REGION_MASK : 0;
+ final int maxX = currRegionX == maxRegionX ? maxChunkX & REGION_MASK : REGION_MASK;
+
+ for (int currZ = minZ; currZ <= maxZ; ++currZ) {
+ for (int currX = minX; currX <= maxX; ++currX) {
+ final ChunkEntitySlices chunk = region.get(currX | (currZ << REGION_SHIFT));
+ if (chunk == null || !chunk.status.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ continue;
+ }
+
+ chunk.getHardCollidingEntities(except, box, into, predicate);
+ }
+ }
+ }
+ }
+ }
+
+ public <T extends Entity> void getEntities(final EntityType<?> type, final AABB box, final List<? super T> into,
+ final Predicate<? super T> predicate) {
+ final int minChunkX = (Mth.floor(box.minX) - 2) >> 4;
+ final int minChunkZ = (Mth.floor(box.minZ) - 2) >> 4;
+ final int maxChunkX = (Mth.floor(box.maxX) + 2) >> 4;
+ final int maxChunkZ = (Mth.floor(box.maxZ) + 2) >> 4;
+
+ final int minRegionX = minChunkX >> REGION_SHIFT;
+ final int minRegionZ = minChunkZ >> REGION_SHIFT;
+ final int maxRegionX = maxChunkX >> REGION_SHIFT;
+ final int maxRegionZ = maxChunkZ >> REGION_SHIFT;
+
+ for (int currRegionZ = minRegionZ; currRegionZ <= maxRegionZ; ++currRegionZ) {
+ final int minZ = currRegionZ == minRegionZ ? minChunkZ & REGION_MASK : 0;
+ final int maxZ = currRegionZ == maxRegionZ ? maxChunkZ & REGION_MASK : REGION_MASK;
+
+ for (int currRegionX = minRegionX; currRegionX <= maxRegionX; ++currRegionX) {
+ final ChunkSlicesRegion region = this.getRegion(currRegionX, currRegionZ);
+
+ if (region == null) {
+ continue;
+ }
+
+ final int minX = currRegionX == minRegionX ? minChunkX & REGION_MASK : 0;
+ final int maxX = currRegionX == maxRegionX ? maxChunkX & REGION_MASK : REGION_MASK;
+
+ for (int currZ = minZ; currZ <= maxZ; ++currZ) {
+ for (int currX = minX; currX <= maxX; ++currX) {
+ final ChunkEntitySlices chunk = region.get(currX | (currZ << REGION_SHIFT));
+ if (chunk == null || !chunk.status.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ continue;
+ }
+
+ chunk.getEntities(type, box, (List)into, (Predicate)predicate);
+ }
+ }
+ }
+ }
+ }
+
+ public <T extends Entity> void getEntities(final Class<? extends T> clazz, final Entity except, final AABB box, final List<? super T> into,
+ final Predicate<? super T> predicate) {
+ final int minChunkX = (Mth.floor(box.minX) - 2) >> 4;
+ final int minChunkZ = (Mth.floor(box.minZ) - 2) >> 4;
+ final int maxChunkX = (Mth.floor(box.maxX) + 2) >> 4;
+ final int maxChunkZ = (Mth.floor(box.maxZ) + 2) >> 4;
+
+ final int minRegionX = minChunkX >> REGION_SHIFT;
+ final int minRegionZ = minChunkZ >> REGION_SHIFT;
+ final int maxRegionX = maxChunkX >> REGION_SHIFT;
+ final int maxRegionZ = maxChunkZ >> REGION_SHIFT;
+
+ for (int currRegionZ = minRegionZ; currRegionZ <= maxRegionZ; ++currRegionZ) {
+ final int minZ = currRegionZ == minRegionZ ? minChunkZ & REGION_MASK : 0;
+ final int maxZ = currRegionZ == maxRegionZ ? maxChunkZ & REGION_MASK : REGION_MASK;
+
+ for (int currRegionX = minRegionX; currRegionX <= maxRegionX; ++currRegionX) {
+ final ChunkSlicesRegion region = this.getRegion(currRegionX, currRegionZ);
+
+ if (region == null) {
+ continue;
+ }
+
+ final int minX = currRegionX == minRegionX ? minChunkX & REGION_MASK : 0;
+ final int maxX = currRegionX == maxRegionX ? maxChunkX & REGION_MASK : REGION_MASK;
+
+ for (int currZ = minZ; currZ <= maxZ; ++currZ) {
+ for (int currX = minX; currX <= maxX; ++currX) {
+ final ChunkEntitySlices chunk = region.get(currX | (currZ << REGION_SHIFT));
+ if (chunk == null || !chunk.status.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ continue;
+ }
+
+ chunk.getEntities(clazz, except, box, into, predicate);
+ }
+ }
+ }
+ }
+ }
+
+ public void entitySectionLoad(final int chunkX, final int chunkZ, final ChunkEntitySlices slices) {
+ TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Cannot load in entity section off-main");
+ synchronized (this) {
+ final ChunkEntitySlices curr = this.getChunk(chunkX, chunkZ);
+ if (curr != null) {
+ this.removeChunk(chunkX, chunkZ);
+
+ curr.mergeInto(slices);
+
+ this.addChunk(chunkX, chunkZ, slices);
+ } else {
+ this.addChunk(chunkX, chunkZ, slices);
+ }
+ }
+ }
+
+ public void entitySectionUnload(final int chunkX, final int chunkZ) {
+ TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Cannot unload entity section off-main");
+ this.removeChunk(chunkX, chunkZ);
+ }
+
+ public ChunkEntitySlices getChunk(final int chunkX, final int chunkZ) {
+ final ChunkSlicesRegion region = this.getRegion(chunkX >> REGION_SHIFT, chunkZ >> REGION_SHIFT);
+ if (region == null) {
+ return null;
+ }
+
+ return region.get((chunkX & REGION_MASK) | ((chunkZ & REGION_MASK) << REGION_SHIFT));
+ }
+
+ public ChunkEntitySlices getOrCreateChunk(final int chunkX, final int chunkZ) {
+ final ChunkSlicesRegion region = this.getRegion(chunkX >> REGION_SHIFT, chunkZ >> REGION_SHIFT);
+ ChunkEntitySlices ret;
+ if (region == null || (ret = region.get((chunkX & REGION_MASK) | ((chunkZ & REGION_MASK) << REGION_SHIFT))) == null) {
+ // loadInEntityChunk will call addChunk for us
+ return this.world.chunkTaskScheduler.chunkHolderManager.getOrCreateEntityChunk(chunkX, chunkZ, true);
+ }
+
+ return ret;
+ }
+
+ public ChunkSlicesRegion getRegion(final int regionX, final int regionZ) {
+ final long key = CoordinateUtils.getChunkKey(regionX, regionZ);
+ final long attempt = this.stateLock.tryOptimisticRead();
+ if (attempt != 0L) {
+ try {
+ final ChunkSlicesRegion ret = this.regions.get(key);
+
+ if (this.stateLock.validate(attempt)) {
+ return ret;
+ }
+ } catch (final Error error) {
+ throw error;
+ } catch (final Throwable thr) {
+ // ignore
+ }
+ }
+
+ this.stateLock.readLock();
+ try {
+ return this.regions.get(key);
+ } finally {
+ this.stateLock.tryUnlockRead();
+ }
+ }
+
+ private synchronized void removeChunk(final int chunkX, final int chunkZ) {
+ final long key = CoordinateUtils.getChunkKey(chunkX >> REGION_SHIFT, chunkZ >> REGION_SHIFT);
+ final int relIndex = (chunkX & REGION_MASK) | ((chunkZ & REGION_MASK) << REGION_SHIFT);
+
+ final ChunkSlicesRegion region = this.regions.get(key);
+ final int remaining = region.remove(relIndex);
+
+ if (remaining == 0) {
+ this.stateLock.writeLock();
+ try {
+ this.regions.remove(key);
+ } finally {
+ this.stateLock.tryUnlockWrite();
+ }
+ }
+ }
+
+ public synchronized void addChunk(final int chunkX, final int chunkZ, final ChunkEntitySlices slices) {
+ final long key = CoordinateUtils.getChunkKey(chunkX >> REGION_SHIFT, chunkZ >> REGION_SHIFT);
+ final int relIndex = (chunkX & REGION_MASK) | ((chunkZ & REGION_MASK) << REGION_SHIFT);
+
+ ChunkSlicesRegion region = this.regions.get(key);
+ if (region != null) {
+ region.add(relIndex, slices);
+ } else {
+ region = new ChunkSlicesRegion();
+ region.add(relIndex, slices);
+ this.stateLock.writeLock();
+ try {
+ this.regions.put(key, region);
+ } finally {
+ this.stateLock.tryUnlockWrite();
+ }
+ }
+ }
+
+ public static final class ChunkSlicesRegion {
+
+ protected final ChunkEntitySlices[] slices = new ChunkEntitySlices[REGION_SIZE * REGION_SIZE];
+ protected int sliceCount;
+
+ public ChunkEntitySlices get(final int index) {
+ return this.slices[index];
+ }
+
+ public int remove(final int index) {
+ final ChunkEntitySlices slices = this.slices[index];
+ if (slices == null) {
+ throw new IllegalStateException();
+ }
+
+ this.slices[index] = null;
+
+ return --this.sliceCount;
+ }
+
+ public void add(final int index, final ChunkEntitySlices slices) {
+ final ChunkEntitySlices curr = this.slices[index];
+ if (curr != null) {
+ throw new IllegalStateException();
+ }
+
+ this.slices[index] = slices;
+
+ ++this.sliceCount;
+ }
+ }
+
+ private final class EntityCallback implements EntityInLevelCallback {
+
+ public final Entity entity;
+
+ public EntityCallback(final Entity entity) {
+ this.entity = entity;
+ }
+
+ @Override
+ public void onMove() {
+ final Entity entity = this.entity;
+ final Visibility oldVisibility = getEntityStatus(entity);
+ final ChunkEntitySlices newSlices = EntityLookup.this.moveEntity(this.entity);
+ if (newSlices == null) {
+ // no new section, so didn't change sections
+ return;
+ }
+ final Visibility newVisibility = getEntityStatus(entity);
+
+ EntityLookup.this.entityStatusChange(entity, newSlices, oldVisibility, newVisibility, true, false, false);
+ }
+
+ @Override
+ public void onRemove(final Entity.RemovalReason reason) {
+ final Entity entity = this.entity;
+ TickThread.ensureTickThread(entity, "Cannot remove entity off-main"); // Paper - rewrite chunk system
+ final Visibility tickingState = EntityLookup.getEntityStatus(entity);
+
+ EntityLookup.this.removeEntity(entity);
+
+ EntityLookup.this.entityStatusChange(entity, null, tickingState, Visibility.HIDDEN, false, false, reason.shouldDestroy());
+
+ this.entity.setLevelCallback(NoOpCallback.INSTANCE);
+ }
+ }
+
+ private static final class NoOpCallback implements EntityInLevelCallback {
+
+ public static final NoOpCallback INSTANCE = new NoOpCallback();
+
+ @Override
+ public void onMove() {}
+
+ @Override
+ public void onRemove(final Entity.RemovalReason reason) {}
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/io/RegionFileIOThread.java b/src/main/java/io/papermc/paper/chunk/system/io/RegionFileIOThread.java
new file mode 100644
index 0000000000000000000000000000000000000000..a08cde4eefe879adcee7c4118bc38f98c5097ed0
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/io/RegionFileIOThread.java
@@ -0,0 +1,1328 @@
+package io.papermc.paper.chunk.system.io;
+
+import ca.spottedleaf.concurrentutil.collection.MultiThreadedQueue;
+import ca.spottedleaf.concurrentutil.executor.Cancellable;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedQueueExecutorThread;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedThreadedTaskQueue;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import it.unimi.dsi.fastutil.HashCommon;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.storage.RegionFile;
+import net.minecraft.world.level.chunk.storage.RegionFileStorage;
+import org.slf4j.Logger;
+import java.io.IOException;
+import java.lang.invoke.VarHandle;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.CompletionException;
+import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.BiConsumer;
+import java.util.function.BiFunction;
+import java.util.function.Consumer;
+import java.util.function.Function;
+
+/**
+ * Prioritised RegionFile I/O executor, responsible for all RegionFile access.
+ * <p>
+ * All functions provided are MT-Safe, however certain ordering constraints are recommended:
+ * <li>
+ * Chunk saves may not occur for unloaded chunks.
+ * </li>
+ * <li>
+ * Tasks must be scheduled on the chunk scheduler thread.
+ * </li>
+ * By following these constraints, no chunk data loss should occur with the exception of underlying I/O problems.
+ * </p>
+ */
+public final class RegionFileIOThread extends PrioritisedQueueExecutorThread {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ /**
+ * The kinds of region files controlled by the region file thread. Add more when needed, and ensure
+ * getControllerFor is updated.
+ */
+ public static enum RegionFileType {
+ CHUNK_DATA,
+ POI_DATA,
+ ENTITY_DATA;
+ }
+
+ protected static final RegionFileType[] CACHED_REGIONFILE_TYPES = RegionFileType.values();
+
+ private ChunkDataController getControllerFor(final ServerLevel world, final RegionFileType type) {
+ switch (type) {
+ case CHUNK_DATA:
+ return world.chunkDataControllerNew;
+ case POI_DATA:
+ return world.poiDataControllerNew;
+ case ENTITY_DATA:
+ return world.entityDataControllerNew;
+ default:
+ throw new IllegalStateException("Unknown controller type " + type);
+ }
+ }
+
+ /**
+ * Collects regionfile data for a certain chunk.
+ */
+ public static final class RegionFileData {
+
+ private final boolean[] hasResult = new boolean[CACHED_REGIONFILE_TYPES.length];
+ private final CompoundTag[] data = new CompoundTag[CACHED_REGIONFILE_TYPES.length];
+ private final Throwable[] throwables = new Throwable[CACHED_REGIONFILE_TYPES.length];
+
+ /**
+ * Sets the result associated with the specified regionfile type. Note that
+ * results can only be set once per regionfile type.
+ *
+ * @param type The regionfile type.
+ * @param data The result to set.
+ */
+ public void setData(final RegionFileType type, final CompoundTag data) {
+ final int index = type.ordinal();
+
+ if (this.hasResult[index]) {
+ throw new IllegalArgumentException("Result already exists for type " + type);
+ }
+ this.hasResult[index] = true;
+ this.data[index] = data;
+ }
+
+ /**
+ * Sets the result associated with the specified regionfile type. Note that
+ * results can only be set once per regionfile type.
+ *
+ * @param type The regionfile type.
+ * @param throwable The result to set.
+ */
+ public void setThrowable(final RegionFileType type, final Throwable throwable) {
+ final int index = type.ordinal();
+
+ if (this.hasResult[index]) {
+ throw new IllegalArgumentException("Result already exists for type " + type);
+ }
+ this.hasResult[index] = true;
+ this.throwables[index] = throwable;
+ }
+
+ /**
+ * Returns whether there is a result for the specified regionfile type.
+ *
+ * @param type Specified regionfile type.
+ *
+ * @return Whether a result exists for {@code type}.
+ */
+ public boolean hasResult(final RegionFileType type) {
+ return this.hasResult[type.ordinal()];
+ }
+
+ /**
+ * Returns the data result for the regionfile type.
+ *
+ * @param type Specified regionfile type.
+ *
+ * @throws IllegalArgumentException If the result has not been set for {@code type}.
+ * @return The data result for the specified type. If the result is a {@code Throwable},
+ * then returns {@code null}.
+ */
+ public CompoundTag getData(final RegionFileType type) {
+ final int index = type.ordinal();
+
+ if (!this.hasResult[index]) {
+ throw new IllegalArgumentException("Result does not exist for type " + type);
+ }
+
+ return this.data[index];
+ }
+
+ /**
+ * Returns the throwable result for the regionfile type.
+ *
+ * @param type Specified regionfile type.
+ *
+ * @throws IllegalArgumentException If the result has not been set for {@code type}.
+ * @return The throwable result for the specified type. If the result is an {@code CompoundTag},
+ * then returns {@code null}.
+ */
+ public Throwable getThrowable(final RegionFileType type) {
+ final int index = type.ordinal();
+
+ if (!this.hasResult[index]) {
+ throw new IllegalArgumentException("Result does not exist for type " + type);
+ }
+
+ return this.throwables[index];
+ }
+ }
+
+ private static final Object INIT_LOCK = new Object();
+
+ static RegionFileIOThread[] threads;
+
+ /* needs to be consistent given a set of parameters */
+ static RegionFileIOThread selectThread(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type) {
+ if (threads == null) {
+ throw new IllegalStateException("Threads not initialised");
+ }
+
+ final int regionX = chunkX >> 5;
+ final int regionZ = chunkZ >> 5;
+ final int typeOffset = type.ordinal();
+
+ return threads[(System.identityHashCode(world) + regionX + regionZ + typeOffset) % threads.length];
+ }
+
+ /**
+ * Shuts down the I/O executor(s). Watis for all tasks to complete if specified.
+ * Tasks queued during this call might not be accepted, and tasks queued after will not be accepted.
+ *
+ * @param wait Whether to wait until all tasks have completed.
+ */
+ public static void close(final boolean wait) {
+ for (int i = 0, len = threads.length; i < len; ++i) {
+ threads[i].close(false, true);
+ }
+ if (wait) {
+ RegionFileIOThread.flush();
+ }
+ }
+
+ public static long[] getExecutedTasks() {
+ final long[] ret = new long[threads.length];
+ for (int i = 0, len = threads.length; i < len; ++i) {
+ ret[i] = threads[i].getTotalTasksExecuted();
+ }
+
+ return ret;
+ }
+
+ public static long[] getTasksScheduled() {
+ final long[] ret = new long[threads.length];
+ for (int i = 0, len = threads.length; i < len; ++i) {
+ ret[i] = threads[i].getTotalTasksScheduled();
+ }
+ return ret;
+ }
+
+ public static void flush() {
+ for (int i = 0, len = threads.length; i < len; ++i) {
+ threads[i].waitUntilAllExecuted();
+ }
+ }
+
+ public static void partialFlush(final int totalTasksRemaining) {
+ long failures = 1L; // start out at 0.25ms
+
+ for (;;) {
+ final long[] executed = getExecutedTasks();
+ final long[] scheduled = getTasksScheduled();
+
+ long sum = 0;
+ for (int i = 0; i < executed.length; ++i) {
+ sum += scheduled[i] - executed[i];
+ }
+
+ if (sum <= totalTasksRemaining) {
+ break;
+ }
+
+ failures = ConcurrentUtil.linearLongBackoff(failures, 250_000L, 5_000_000L); // 500us, 5ms
+ }
+ }
+
+ /**
+ * Inits the executor with the specified number of threads.
+ *
+ * @param threads Specified number of threads.
+ */
+ public static void init(final int threads) {
+ synchronized (INIT_LOCK) {
+ if (RegionFileIOThread.threads != null) {
+ throw new IllegalStateException("Already initialised threads");
+ }
+
+ RegionFileIOThread.threads = new RegionFileIOThread[threads];
+
+ for (int i = 0; i < threads; ++i) {
+ RegionFileIOThread.threads[i] = new RegionFileIOThread(i);
+ RegionFileIOThread.threads[i].start();
+ }
+ }
+ }
+
+ private RegionFileIOThread(final int threadNumber) {
+ super(new PrioritisedThreadedTaskQueue(), (int)(1.0e6)); // 1.0ms spinwait time
+ this.setName("RegionFile I/O Thread #" + threadNumber);
+ this.setPriority(Thread.NORM_PRIORITY - 2); // we keep priority close to normal because threads can wait on us
+ this.setUncaughtExceptionHandler((final Thread thread, final Throwable thr) -> {
+ LOGGER.error("Uncaught exception thrown from I/O thread, report this! Thread: " + thread.getName(), thr);
+ });
+ }
+
+ /**
+ * Returns whether the current thread is a regionfile I/O executor.
+ * @return Whether the current thread is a regionfile I/O executor.
+ */
+ public static boolean isRegionFileThread() {
+ return Thread.currentThread() instanceof RegionFileIOThread;
+ }
+
+ /**
+ * Returns the priority associated with blocking I/O based on the current thread. The goal is to avoid
+ * dumb plugins from taking away priority from threads we consider crucial.
+ * @return The priroity to use with blocking I/O on the current thread.
+ */
+ public static PrioritisedExecutor.Priority getIOBlockingPriorityForCurrentThread() {
+ if (TickThread.isTickThread()) {
+ return PrioritisedExecutor.Priority.BLOCKING;
+ }
+ return PrioritisedExecutor.Priority.HIGHEST;
+ }
+
+ /**
+ * Returns the current {@code CompoundTag} pending for write for the specified chunk & regionfile type.
+ * Note that this does not copy the result, so do not modify the result returned.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param type Specified regionfile type.
+ *
+ * @return The compound tag associated for the specified chunk. {@code null} if no write was pending, or if {@code null} is the write pending.
+ */
+ public static CompoundTag getPendingWrite(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ return thread.getPendingWriteInternal(world, chunkX, chunkZ, type);
+ }
+
+ CompoundTag getPendingWriteInternal(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+ final ChunkDataTask task = taskController.tasks.get(Long.valueOf(CoordinateUtils.getChunkKey(chunkX, chunkZ)));
+
+ if (task == null) {
+ return null;
+ }
+
+ final CompoundTag ret = task.inProgressWrite;
+
+ return ret == ChunkDataTask.NOTHING_TO_WRITE ? null : ret;
+ }
+
+ /**
+ * Returns the priority for the specified regionfile type for the specified chunk.
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param type Specified regionfile type.
+ * @return The priority for the chunk
+ */
+ public static PrioritisedExecutor.Priority getPriority(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ return thread.getPriorityInternal(world, chunkX, chunkZ, type);
+ }
+
+ PrioritisedExecutor.Priority getPriorityInternal(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+ final ChunkDataTask task = taskController.tasks.get(Long.valueOf(CoordinateUtils.getChunkKey(chunkX, chunkZ)));
+
+ if (task == null) {
+ return PrioritisedExecutor.Priority.COMPLETING;
+ }
+
+ return task.prioritisedTask.getPriority();
+ }
+
+ /**
+ * Sets the priority for all regionfile types for the specified chunk. Note that great care should
+ * be taken using this method, as there can be multiple tasks tied to the same chunk that want different
+ * priorities.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param priority New priority.
+ *
+ * @see #raisePriority(ServerLevel, int, int, Priority)
+ * @see #raisePriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void setPriority(final ServerLevel world, final int chunkX, final int chunkZ,
+ final PrioritisedExecutor.Priority priority) {
+ for (final RegionFileType type : CACHED_REGIONFILE_TYPES) {
+ RegionFileIOThread.setPriority(world, chunkX, chunkZ, type, priority);
+ }
+ }
+
+ /**
+ * Sets the priority for the specified regionfile type for the specified chunk. Note that great care should
+ * be taken using this method, as there can be multiple tasks tied to the same chunk that want different
+ * priorities.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param type Specified regionfile type.
+ * @param priority New priority.
+ *
+ * @see #raisePriority(ServerLevel, int, int, Priority)
+ * @see #raisePriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void setPriority(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ thread.setPriorityInternal(world, chunkX, chunkZ, type, priority);
+ }
+
+ void setPriorityInternal(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+ final ChunkDataTask task = taskController.tasks.get(Long.valueOf(CoordinateUtils.getChunkKey(chunkX, chunkZ)));
+
+ if (task != null) {
+ task.prioritisedTask.setPriority(priority);
+ }
+ }
+
+ /**
+ * Raises the priority for all regionfile types for the specified chunk.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param priority New priority.
+ *
+ * @see #setPriority(ServerLevel, int, int, Priority)
+ * @see #setPriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void raisePriority(final ServerLevel world, final int chunkX, final int chunkZ,
+ final PrioritisedExecutor.Priority priority) {
+ for (final RegionFileType type : CACHED_REGIONFILE_TYPES) {
+ RegionFileIOThread.raisePriority(world, chunkX, chunkZ, type, priority);
+ }
+ }
+
+ /**
+ * Raises the priority for the specified regionfile type for the specified chunk.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param type Specified regionfile type.
+ * @param priority New priority.
+ *
+ * @see #setPriority(ServerLevel, int, int, Priority)
+ * @see #setPriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, Priority)
+ * @see #lowerPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void raisePriority(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ thread.raisePriorityInternal(world, chunkX, chunkZ, type, priority);
+ }
+
+ void raisePriorityInternal(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+ final ChunkDataTask task = taskController.tasks.get(Long.valueOf(CoordinateUtils.getChunkKey(chunkX, chunkZ)));
+
+ if (task != null) {
+ task.prioritisedTask.raisePriority(priority);
+ }
+ }
+
+ /**
+ * Lowers the priority for all regionfile types for the specified chunk.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param priority New priority.
+ *
+ * @see #raisePriority(ServerLevel, int, int, Priority)
+ * @see #raisePriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #setPriority(ServerLevel, int, int, Priority)
+ * @see #setPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void lowerPriority(final ServerLevel world, final int chunkX, final int chunkZ,
+ final PrioritisedExecutor.Priority priority) {
+ for (final RegionFileType type : CACHED_REGIONFILE_TYPES) {
+ RegionFileIOThread.lowerPriority(world, chunkX, chunkZ, type, priority);
+ }
+ }
+
+ /**
+ * Lowers the priority for the specified regionfile type for the specified chunk.
+ *
+ * @param world Specified world.
+ * @param chunkX Specified chunk x.
+ * @param chunkZ Specified chunk z.
+ * @param type Specified regionfile type.
+ * @param priority New priority.
+ *
+ * @see #raisePriority(ServerLevel, int, int, Priority)
+ * @see #raisePriority(ServerLevel, int, int, RegionFileType, Priority)
+ * @see #setPriority(ServerLevel, int, int, Priority)
+ * @see #setPriority(ServerLevel, int, int, RegionFileType, Priority)
+ */
+ public static void lowerPriority(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ thread.lowerPriorityInternal(world, chunkX, chunkZ, type, priority);
+ }
+
+ void lowerPriorityInternal(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+ final ChunkDataTask task = taskController.tasks.get(Long.valueOf(CoordinateUtils.getChunkKey(chunkX, chunkZ)));
+
+ if (task != null) {
+ task.prioritisedTask.lowerPriority(priority);
+ }
+ }
+
+ /**
+ * Schedules the chunk data to be written asynchronously.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * This function presumes a chunk load for the coordinates is not called during this function (anytime after is OK). This means
+ * saves must be scheduled before a chunk is unloaded.
+ * </li>
+ * <li>
+ * Writes may be called concurrently, although only the "later" write will go through.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param data Chunk's data
+ * @param type The regionfile type to write to.
+ *
+ * @throws IllegalStateException If the file io thread has shutdown.
+ */
+ public static void scheduleSave(final ServerLevel world, final int chunkX, final int chunkZ, final CompoundTag data,
+ final RegionFileType type) {
+ RegionFileIOThread.scheduleSave(world, chunkX, chunkZ, data, type, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ /**
+ * Schedules the chunk data to be written asynchronously.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * This function presumes a chunk load for the coordinates is not called during this function (anytime after is OK). This means
+ * saves must be scheduled before a chunk is unloaded.
+ * </li>
+ * <li>
+ * Writes may be called concurrently, although only the "later" write will go through.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param data Chunk's data
+ * @param type The regionfile type to write to.
+ * @param priority The minimum priority to schedule at.
+ *
+ * @throws IllegalStateException If the file io thread has shutdown.
+ */
+ public static void scheduleSave(final ServerLevel world, final int chunkX, final int chunkZ, final CompoundTag data,
+ final RegionFileType type, final PrioritisedExecutor.Priority priority) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ thread.scheduleSaveInternal(world, chunkX, chunkZ, data, type, priority);
+ }
+
+ void scheduleSaveInternal(final ServerLevel world, final int chunkX, final int chunkZ, final CompoundTag data,
+ final RegionFileType type, final PrioritisedExecutor.Priority priority) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+
+ final boolean[] created = new boolean[1];
+ final ChunkCoordinate key = new ChunkCoordinate(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ final ChunkDataTask task = taskController.tasks.compute(key, (final ChunkCoordinate keyInMap, final ChunkDataTask taskRunning) -> {
+ if (taskRunning == null || taskRunning.failedWrite) {
+ // no task is scheduled or the previous write failed - meaning we need to overwrite it
+
+ // create task
+ final ChunkDataTask newTask = new ChunkDataTask(world, chunkX, chunkZ, taskController, RegionFileIOThread.this, priority);
+ newTask.inProgressWrite = data;
+ created[0] = true;
+
+ return newTask;
+ }
+
+ taskRunning.inProgressWrite = data;
+
+ return taskRunning;
+ });
+
+ if (created[0]) {
+ task.prioritisedTask.queue();
+ } else {
+ task.prioritisedTask.raisePriority(priority);
+ }
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load all regionfile types, and then call
+ * {@code onComplete}. This is a bulk load operation, see {@link #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)}
+ * for single load.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, RegionFileType...)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, Priority, RegionFileType...)
+ */
+ public static Cancellable loadAllChunkData(final ServerLevel world, final int chunkX, final int chunkZ,
+ final Consumer<RegionFileData> onComplete, final boolean intendingToBlock) {
+ return RegionFileIOThread.loadAllChunkData(world, chunkX, chunkZ, onComplete, intendingToBlock, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load all regionfile types, and then call
+ * {@code onComplete}. This is a bulk load operation, see {@link #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)}
+ * for single load.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ * @param priority The minimum priority to load the data at.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, RegionFileType...)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, Priority, RegionFileType...)
+ */
+ public static Cancellable loadAllChunkData(final ServerLevel world, final int chunkX, final int chunkZ,
+ final Consumer<RegionFileData> onComplete, final boolean intendingToBlock,
+ final PrioritisedExecutor.Priority priority) {
+ return RegionFileIOThread.loadChunkData(world, chunkX, chunkZ, onComplete, intendingToBlock, priority, CACHED_REGIONFILE_TYPES);
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load data for the specified regionfile type(s), and
+ * then call {@code onComplete}. This is a bulk load operation, see {@link #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)}
+ * for single load.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ * @param types The regionfile type(s) to load.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean, Priority)
+ */
+ public static Cancellable loadChunkData(final ServerLevel world, final int chunkX, final int chunkZ,
+ final Consumer<RegionFileData> onComplete, final boolean intendingToBlock,
+ final RegionFileType... types) {
+ return RegionFileIOThread.loadChunkData(world, chunkX, chunkZ, onComplete, intendingToBlock, PrioritisedExecutor.Priority.NORMAL, types);
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load data for the specified regionfile type(s), and
+ * then call {@code onComplete}. This is a bulk load operation, see {@link #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)}
+ * for single load.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ * @param types The regionfile type(s) to load.
+ * @param priority The minimum priority to load the data at.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean)
+ * @see #loadDataAsync(ServerLevel, int, int, RegionFileType, BiConsumer, boolean, Priority)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean, Priority)
+ */
+ public static Cancellable loadChunkData(final ServerLevel world, final int chunkX, final int chunkZ,
+ final Consumer<RegionFileData> onComplete, final boolean intendingToBlock,
+ final PrioritisedExecutor.Priority priority, final RegionFileType... types) {
+ if (types == null) {
+ throw new NullPointerException("Types cannot be null");
+ }
+ if (types.length == 0) {
+ throw new IllegalArgumentException("Types cannot be empty");
+ }
+
+ final RegionFileData ret = new RegionFileData();
+
+ final Cancellable[] reads = new CancellableRead[types.length];
+ final AtomicInteger completions = new AtomicInteger();
+ final int expectedCompletions = types.length;
+
+ for (int i = 0; i < expectedCompletions; ++i) {
+ final RegionFileType type = types[i];
+ reads[i] = RegionFileIOThread.loadDataAsync(world, chunkX, chunkZ, type,
+ (final CompoundTag data, final Throwable throwable) -> {
+ if (throwable != null) {
+ ret.setThrowable(type, throwable);
+ } else {
+ ret.setData(type, data);
+ }
+
+ if (completions.incrementAndGet() == expectedCompletions) {
+ onComplete.accept(ret);
+ }
+ }, intendingToBlock, priority);
+ }
+
+ return new CancellableReads(reads);
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load the specified regionfile type, and then call
+ * {@code onComplete}.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, RegionFileType...)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, Priority, RegionFileType...)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean, Priority)
+ */
+ public static Cancellable loadDataAsync(final ServerLevel world, final int chunkX, final int chunkZ,
+ final RegionFileType type, final BiConsumer<CompoundTag, Throwable> onComplete,
+ final boolean intendingToBlock) {
+ return RegionFileIOThread.loadDataAsync(world, chunkX, chunkZ, type, onComplete, intendingToBlock, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ /**
+ * Schedules a load to be executed asynchronously. This task will load the specified regionfile type, and then call
+ * {@code onComplete}.
+ * <p>
+ * Impl notes:
+ * </p>
+ * <li>
+ * The {@code onComplete} parameter may be completed during the execution of this function synchronously or it may
+ * be completed asynchronously on this file io thread. Interacting with the file IO thread in the completion of
+ * data is undefined behaviour, and can cause deadlock.
+ * </li>
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param onComplete Consumer to execute once this task has completed
+ * @param intendingToBlock Whether the caller is intending to block on completion. This only affects the cost
+ * of this call.
+ * @param priority Minimum priority to load the data at.
+ *
+ * @return The {@link Cancellable} for this chunk load. Cancelling it will not affect other loads for the same chunk data.
+ *
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, RegionFileType...)
+ * @see #loadChunkData(ServerLevel, int, int, Consumer, boolean, Priority, RegionFileType...)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean)
+ * @see #loadAllChunkData(ServerLevel, int, int, Consumer, boolean, Priority)
+ */
+ public static Cancellable loadDataAsync(final ServerLevel world, final int chunkX, final int chunkZ,
+ final RegionFileType type, final BiConsumer<CompoundTag, Throwable> onComplete,
+ final boolean intendingToBlock, final PrioritisedExecutor.Priority priority) {
+ final RegionFileIOThread thread = RegionFileIOThread.selectThread(world, chunkX, chunkZ, type);
+ return thread.loadDataAsyncInternal(world, chunkX, chunkZ, type, onComplete, intendingToBlock, priority);
+ }
+
+ private static Boolean doesRegionFileExist(final int chunkX, final int chunkZ, final boolean intendingToBlock,
+ final ChunkDataController taskController) {
+ final ChunkPos chunkPos = new ChunkPos(chunkX, chunkZ);
+ if (intendingToBlock) {
+ return taskController.computeForRegionFile(chunkX, chunkZ, true, (final RegionFile file) -> {
+ if (file == null) { // null if no regionfile exists
+ return Boolean.FALSE;
+ }
+
+ return file.hasChunk(chunkPos) ? Boolean.TRUE : Boolean.FALSE;
+ });
+ } else {
+ return taskController.computeForRegionFileIfLoaded(chunkX, chunkZ, (final RegionFile file) -> {
+ if (file == null) { // null if not loaded
+ return Boolean.TRUE;
+ }
+
+ return file.hasChunk(chunkPos) ? Boolean.TRUE : Boolean.FALSE;
+ });
+ }
+ }
+
+ Cancellable loadDataAsyncInternal(final ServerLevel world, final int chunkX, final int chunkZ,
+ final RegionFileType type, final BiConsumer<CompoundTag, Throwable> onComplete,
+ final boolean intendingToBlock, final PrioritisedExecutor.Priority priority) {
+ final ChunkDataController taskController = this.getControllerFor(world, type);
+
+ final ImmediateCallbackCompletion callbackInfo = new ImmediateCallbackCompletion();
+
+ final ChunkCoordinate key = new ChunkCoordinate(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ final BiFunction<ChunkCoordinate, ChunkDataTask, ChunkDataTask> compute = (final ChunkCoordinate keyInMap, final ChunkDataTask running) -> {
+ if (running == null) {
+ // not scheduled
+
+ if (callbackInfo.regionFileCalculation == null) {
+ // caller will compute this outside of compute(), to avoid holding the bin lock
+ callbackInfo.needsRegionFileTest = true;
+ return null;
+ }
+
+ if (callbackInfo.regionFileCalculation == Boolean.FALSE) {
+ // not on disk
+ callbackInfo.data = null;
+ callbackInfo.throwable = null;
+ callbackInfo.completeNow = true;
+ return null;
+ }
+
+ // set up task
+ final ChunkDataTask newTask = new ChunkDataTask(
+ world, chunkX, chunkZ, taskController, RegionFileIOThread.this, priority
+ );
+ newTask.inProgressRead = new RegionFileIOThread.InProgressRead();
+ newTask.inProgressRead.waiters.add(onComplete);
+
+ callbackInfo.tasksNeedsScheduling = true;
+ return newTask;
+ }
+
+ final CompoundTag pendingWrite = running.inProgressWrite;
+
+ if (pendingWrite == ChunkDataTask.NOTHING_TO_WRITE) {
+ // need to add to waiters here, because the regionfile thread will use compute() to lock and check for cancellations
+ if (!running.inProgressRead.addToWaiters(onComplete)) {
+ callbackInfo.data = running.inProgressRead.value;
+ callbackInfo.throwable = running.inProgressRead.throwable;
+ callbackInfo.completeNow = true;
+ }
+ return running;
+ }
+ // using the result sync here - don't bump priority
+
+ // at this stage we have to use the in progress write's data to avoid an order issue
+ callbackInfo.data = pendingWrite;
+ callbackInfo.throwable = null;
+ callbackInfo.completeNow = true;
+ return running;
+ };
+
+ ChunkDataTask curr = taskController.tasks.get(key);
+ if (curr == null) {
+ callbackInfo.regionFileCalculation = doesRegionFileExist(chunkX, chunkZ, intendingToBlock, taskController);
+ }
+ ChunkDataTask ret = taskController.tasks.compute(key, compute);
+ if (callbackInfo.needsRegionFileTest) {
+ // curr isn't null but when we went into compute() it was
+ callbackInfo.regionFileCalculation = doesRegionFileExist(chunkX, chunkZ, intendingToBlock, taskController);
+ // now it should be fine
+ ret = taskController.tasks.compute(key, compute);
+ }
+
+ // needs to be scheduled
+ if (callbackInfo.tasksNeedsScheduling) {
+ ret.prioritisedTask.queue();
+ } else if (callbackInfo.completeNow) {
+ try {
+ onComplete.accept(callbackInfo.data, callbackInfo.throwable);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Callback " + ConcurrentUtil.genericToString(onComplete) + " synchronously failed to handle chunk data for task " + ret.toString(), thr);
+ }
+ } else {
+ // we're waiting on a task we didn't schedule, so raise its priority to what we want
+ ret.prioritisedTask.raisePriority(priority);
+ }
+
+ return new CancellableRead(onComplete, ret);
+ }
+
+ /**
+ * Schedules a load task to be executed asynchronously, and blocks on that task.
+ *
+ * @param world Chunk's world
+ * @param chunkX Chunk's x coordinate
+ * @param chunkZ Chunk's z coordinate
+ * @param type Regionfile type
+ * @param priority Minimum priority to load the data at.
+ *
+ * @return The chunk data for the chunk. Note that a {@code null} result means the chunk or regionfile does not exist on disk.
+ *
+ * @throws IOException If the load fails for any reason
+ */
+ public static CompoundTag loadData(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileType type,
+ final PrioritisedExecutor.Priority priority) throws IOException {
+ final CompletableFuture<CompoundTag> ret = new CompletableFuture<>();
+
+ RegionFileIOThread.loadDataAsync(world, chunkX, chunkZ, type, (final CompoundTag compound, final Throwable thr) -> {
+ if (thr != null) {
+ ret.completeExceptionally(thr);
+ } else {
+ ret.complete(compound);
+ }
+ }, true, priority);
+
+ try {
+ return ret.join();
+ } catch (final CompletionException ex) {
+ throw new IOException(ex);
+ }
+ }
+
+ private static final class ImmediateCallbackCompletion {
+
+ public CompoundTag data;
+ public Throwable throwable;
+ public boolean completeNow;
+ public boolean tasksNeedsScheduling;
+ public boolean needsRegionFileTest;
+ public Boolean regionFileCalculation;
+
+ }
+
+ static final class CancellableRead implements Cancellable {
+
+ private BiConsumer<CompoundTag, Throwable> callback;
+ private RegionFileIOThread.ChunkDataTask task;
+
+ CancellableRead(final BiConsumer<CompoundTag, Throwable> callback, final RegionFileIOThread.ChunkDataTask task) {
+ this.callback = callback;
+ this.task = task;
+ }
+
+ @Override
+ public boolean cancel() {
+ final BiConsumer<CompoundTag, Throwable> callback = this.callback;
+ final RegionFileIOThread.ChunkDataTask task = this.task;
+
+ if (callback == null || task == null) {
+ return false;
+ }
+
+ this.callback = null;
+ this.task = null;
+
+ final RegionFileIOThread.InProgressRead read = task.inProgressRead;
+
+ // read can be null if no read was scheduled (i.e no regionfile existed or chunk in regionfile didn't)
+ return (read != null && read.waiters.remove(callback));
+ }
+ }
+
+ static final class CancellableReads implements Cancellable {
+
+ private Cancellable[] reads;
+
+ protected static final VarHandle READS_HANDLE = ConcurrentUtil.getVarHandle(CancellableReads.class, "reads", Cancellable[].class);
+
+ CancellableReads(final Cancellable[] reads) {
+ this.reads = reads;
+ }
+
+ @Override
+ public boolean cancel() {
+ final Cancellable[] reads = (Cancellable[])READS_HANDLE.getAndSet((CancellableReads)this, (Cancellable[])null);
+
+ if (reads == null) {
+ return false;
+ }
+
+ boolean ret = false;
+
+ for (final Cancellable read : reads) {
+ ret |= read.cancel();
+ }
+
+ return ret;
+ }
+ }
+
+ static final class InProgressRead {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ CompoundTag value;
+ Throwable throwable;
+ final MultiThreadedQueue<BiConsumer<CompoundTag, Throwable>> waiters = new MultiThreadedQueue<>();
+
+ // rets false if already completed (callback not invoked), true if callback was added
+ boolean addToWaiters(final BiConsumer<CompoundTag, Throwable> callback) {
+ return this.waiters.add(callback);
+ }
+
+ void complete(final RegionFileIOThread.ChunkDataTask task, final CompoundTag value, final Throwable throwable) {
+ this.value = value;
+ this.throwable = throwable;
+
+ BiConsumer<CompoundTag, Throwable> consumer;
+ while ((consumer = this.waiters.pollOrBlockAdds()) != null) {
+ try {
+ consumer.accept(value, throwable);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Callback " + ConcurrentUtil.genericToString(consumer) + " failed to handle chunk data for task " + task.toString(), thr);
+ }
+ }
+ }
+ }
+
+ /**
+ * Class exists to replace {@link Long} usages as keys inside non-fastutil hashtables. The hash for some Long {@code x}
+ * is defined as {@code (x >>> 32) ^ x}. Chunk keys as long values are defined as {@code ((chunkX & 0xFFFFFFFFL) | (chunkZ << 32))},
+ * which means the hashcode as a Long value will be {@code chunkX ^ chunkZ}. Given that most chunks are created within a radius arounds players,
+ * this will lead to many hash collisions. So, this class uses a better hashing algorithm so that usage of
+ * non-fastutil collections is not degraded.
+ */
+ public static final class ChunkCoordinate implements Comparable<ChunkCoordinate> {
+
+ public final long key;
+
+ public ChunkCoordinate(final long key) {
+ this.key = key;
+ }
+
+ @Override
+ public int hashCode() {
+ return (int)HashCommon.mix(this.key);
+ }
+
+ @Override
+ public boolean equals(final Object obj) {
+ if (this == obj) {
+ return true;
+ }
+
+ if (!(obj instanceof ChunkCoordinate)) {
+ return false;
+ }
+
+ final ChunkCoordinate other = (ChunkCoordinate)obj;
+
+ return this.key == other.key;
+ }
+
+ // This class is intended for HashMap/ConcurrentHashMap usage, which do treeify bin nodes if the chain
+ // is too large. So we should implement compareTo to help.
+ @Override
+ public int compareTo(final RegionFileIOThread.ChunkCoordinate other) {
+ return Long.compare(this.key, other.key);
+ }
+
+ @Override
+ public String toString() {
+ return new ChunkPos(this.key).toString();
+ }
+ }
+
+ public static abstract class ChunkDataController {
+
+ // ConcurrentHashMap synchronizes per chain, so reduce the chance of task's hashes colliding.
+ protected final ConcurrentHashMap<ChunkCoordinate, ChunkDataTask> tasks = new ConcurrentHashMap<>(8192, 0.10f);
+
+ public final RegionFileType type;
+
+ public ChunkDataController(final RegionFileType type) {
+ this.type = type;
+ }
+
+ public abstract RegionFileStorage getCache();
+
+ public abstract void writeData(final int chunkX, final int chunkZ, final CompoundTag compound) throws IOException;
+
+ public abstract CompoundTag readData(final int chunkX, final int chunkZ) throws IOException;
+
+ public boolean hasTasks() {
+ return !this.tasks.isEmpty();
+ }
+
+ public <T> T computeForRegionFile(final int chunkX, final int chunkZ, final boolean existingOnly, final Function<RegionFile, T> function) {
+ final RegionFileStorage cache = this.getCache();
+ final RegionFile regionFile;
+ synchronized (cache) {
+ try {
+ regionFile = cache.getRegionFile(new ChunkPos(chunkX, chunkZ), existingOnly, true);
+ } catch (final IOException ex) {
+ throw new RuntimeException(ex);
+ }
+ }
+
+ try {
+ return function.apply(regionFile);
+ } finally {
+ if (regionFile != null) {
+ regionFile.fileLock.unlock();
+ }
+ }
+ }
+
+ public <T> T computeForRegionFileIfLoaded(final int chunkX, final int chunkZ, final Function<RegionFile, T> function) {
+ final RegionFileStorage cache = this.getCache();
+ final RegionFile regionFile;
+
+ synchronized (cache) {
+ regionFile = cache.getRegionFileIfLoaded(new ChunkPos(chunkX, chunkZ));
+ if (regionFile != null) {
+ regionFile.fileLock.lock();
+ }
+ }
+
+ try {
+ return function.apply(regionFile);
+ } finally {
+ if (regionFile != null) {
+ regionFile.fileLock.unlock();
+ }
+ }
+ }
+ }
+
+ static final class ChunkDataTask implements Runnable {
+
+ protected static final CompoundTag NOTHING_TO_WRITE = new CompoundTag();
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ RegionFileIOThread.InProgressRead inProgressRead;
+ volatile CompoundTag inProgressWrite = NOTHING_TO_WRITE; // only needs to be acquire/release
+
+ boolean failedWrite;
+
+ final ServerLevel world;
+ final int chunkX;
+ final int chunkZ;
+ final RegionFileIOThread.ChunkDataController taskController;
+
+ final PrioritisedExecutor.PrioritisedTask prioritisedTask;
+
+ /*
+ * IO thread will perform reads before writes for a given chunk x and z
+ *
+ * How reads/writes are scheduled:
+ *
+ * If read is scheduled while scheduling write, take no special action and just schedule write
+ * If read is scheduled while scheduling read and no write is scheduled, chain the read task
+ *
+ *
+ * If write is scheduled while scheduling read, use the pending write data and ret immediately (so no read is scheduled)
+ * If write is scheduled while scheduling write (ignore read in progress), overwrite the write in progress data
+ *
+ * This allows the reads and writes to act as if they occur synchronously to the thread scheduling them, however
+ * it fails to properly propagate write failures thanks to writes overwriting each other
+ */
+
+ public ChunkDataTask(final ServerLevel world, final int chunkX, final int chunkZ, final RegionFileIOThread.ChunkDataController taskController,
+ final PrioritisedExecutor executor, final PrioritisedExecutor.Priority priority) {
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.taskController = taskController;
+ this.prioritisedTask = executor.createTask(this, priority);
+ }
+
+ @Override
+ public String toString() {
+ return "Task for world: '" + this.world.getWorld().getName() + "' at (" + this.chunkX + "," + this.chunkZ +
+ ") type: " + this.taskController.type.name() + ", hash: " + this.hashCode();
+ }
+
+ @Override
+ public void run() {
+ final RegionFileIOThread.InProgressRead read = this.inProgressRead;
+ final ChunkCoordinate chunkKey = new ChunkCoordinate(CoordinateUtils.getChunkKey(this.chunkX, this.chunkZ));
+
+ if (read != null) {
+ final boolean[] canRead = new boolean[] { true };
+
+ if (read.waiters.isEmpty()) {
+ // cancelled read? go to task controller to confirm
+ final ChunkDataTask inMap = this.taskController.tasks.compute(chunkKey, (final ChunkCoordinate keyInMap, final ChunkDataTask valueInMap) -> {
+ if (valueInMap == null) {
+ throw new IllegalStateException("Write completed concurrently, expected this task: " + ChunkDataTask.this.toString() + ", report this!");
+ }
+ if (valueInMap != ChunkDataTask.this) {
+ throw new IllegalStateException("Chunk task mismatch, expected this task: " + ChunkDataTask.this.toString() + ", got: " + valueInMap.toString() + ", report this!");
+ }
+
+ if (!read.waiters.isEmpty()) { // as per usual IntelliJ is unable to figure out that there are concurrent accesses.
+ return valueInMap;
+ } else {
+ canRead[0] = false;
+ }
+
+ return valueInMap.inProgressWrite == NOTHING_TO_WRITE ? null : valueInMap;
+ });
+
+ if (inMap == null) {
+ // read is cancelled - and no write pending, so we're done
+ return;
+ }
+ // if there is a write in progress, we don't actually have to worry about waiters gaining new entries -
+ // the readers will just use the in progress write, so the value in canRead is good to use without
+ // further synchronisation.
+ }
+
+ if (canRead[0]) {
+ CompoundTag compound = null;
+ Throwable throwable = null;
+
+ try {
+ compound = this.taskController.readData(this.chunkX, this.chunkZ);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ throwable = thr;
+ LOGGER.error("Failed to read chunk data for task: " + this.toString(), thr);
+ }
+ read.complete(this, compound, throwable);
+ }
+ }
+
+ CompoundTag write = this.inProgressWrite;
+
+ if (write == NOTHING_TO_WRITE) {
+ final ChunkDataTask inMap = this.taskController.tasks.compute(chunkKey, (final ChunkCoordinate keyInMap, final ChunkDataTask valueInMap) -> {
+ if (valueInMap == null) {
+ throw new IllegalStateException("Write completed concurrently, expected this task: " + ChunkDataTask.this.toString() + ", report this!");
+ }
+ if (valueInMap != ChunkDataTask.this) {
+ throw new IllegalStateException("Chunk task mismatch, expected this task: " + ChunkDataTask.this.toString() + ", got: " + valueInMap.toString() + ", report this!");
+ }
+ return valueInMap.inProgressWrite == NOTHING_TO_WRITE ? null : valueInMap;
+ });
+
+ if (inMap == null) {
+ return; // set the task value to null, indicating we're done
+ } // else: inProgressWrite changed, so now we have something to write
+ }
+
+ for (;;) {
+ write = this.inProgressWrite;
+ final CompoundTag dataWritten = write;
+
+ boolean failedWrite = false;
+
+ try {
+ this.taskController.writeData(this.chunkX, this.chunkZ, write);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ if (thr instanceof RegionFileStorage.RegionFileSizeException) {
+ final int maxSize = RegionFile.MAX_CHUNK_SIZE / (1024 * 1024);
+ LOGGER.error("Chunk at (" + this.chunkX + "," + this.chunkZ + ") in '" + this.world.getWorld().getName() + "' exceeds max size of " + maxSize + "MiB, it has been deleted from disk.");
+ } else {
+ failedWrite = thr instanceof IOException;
+ LOGGER.error("Failed to write chunk data for task: " + this.toString(), thr);
+ }
+ }
+
+ final boolean finalFailWrite = failedWrite;
+ final boolean[] done = new boolean[] { false };
+
+ this.taskController.tasks.compute(chunkKey, (final ChunkCoordinate keyInMap, final ChunkDataTask valueInMap) -> {
+ if (valueInMap == null) {
+ throw new IllegalStateException("Write completed concurrently, expected this task: " + ChunkDataTask.this.toString() + ", report this!");
+ }
+ if (valueInMap != ChunkDataTask.this) {
+ throw new IllegalStateException("Chunk task mismatch, expected this task: " + ChunkDataTask.this.toString() + ", got: " + valueInMap.toString() + ", report this!");
+ }
+ if (valueInMap.inProgressWrite == dataWritten) {
+ valueInMap.failedWrite = finalFailWrite;
+ done[0] = true;
+ // keep the data in map if we failed the write so we can try to prevent data loss
+ return finalFailWrite ? valueInMap : null;
+ }
+ // different data than expected, means we need to retry write
+ return valueInMap;
+ });
+
+ if (done[0]) {
+ return;
+ }
+
+ // fetch & write new data
+ continue;
+ }
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/light/LightQueue.java b/src/main/java/io/papermc/paper/chunk/system/light/LightQueue.java
new file mode 100644
index 0000000000000000000000000000000000000000..0b7a2b0ead4f3bc07bfd9a38c2b7cf024bd140c6
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/light/LightQueue.java
@@ -0,0 +1,280 @@
+package io.papermc.paper.chunk.system.light;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.starlight.common.light.BlockStarLightEngine;
+import ca.spottedleaf.starlight.common.light.SkyStarLightEngine;
+import ca.spottedleaf.starlight.common.light.StarLightInterface;
+import io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler;
+import io.papermc.paper.util.CoordinateUtils;
+import it.unimi.dsi.fastutil.longs.Long2ObjectOpenHashMap;
+import it.unimi.dsi.fastutil.shorts.ShortCollection;
+import it.unimi.dsi.fastutil.shorts.ShortOpenHashSet;
+import net.minecraft.core.BlockPos;
+import net.minecraft.core.SectionPos;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.ChunkPos;
+import java.util.ArrayList;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Set;
+import java.util.concurrent.CompletableFuture;
+import java.util.function.BooleanSupplier;
+
+public final class LightQueue {
+
+ protected final Long2ObjectOpenHashMap<ChunkTasks> chunkTasks = new Long2ObjectOpenHashMap<>();
+ protected final StarLightInterface manager;
+ protected final ServerLevel world;
+
+ public LightQueue(final StarLightInterface manager) {
+ this.manager = manager;
+ this.world = ((ServerLevel)manager.getWorld());
+ }
+
+ public void lowerPriority(final int chunkX, final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ final ChunkTasks task;
+ synchronized (this) {
+ task = this.chunkTasks.get(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+ if (task != null) {
+ task.lowerPriority(priority);
+ }
+ }
+
+ public void setPriority(final int chunkX, final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ final ChunkTasks task;
+ synchronized (this) {
+ task = this.chunkTasks.get(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+ if (task != null) {
+ task.setPriority(priority);
+ }
+ }
+
+ public void raisePriority(final int chunkX, final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ final ChunkTasks task;
+ synchronized (this) {
+ task = this.chunkTasks.get(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+ if (task != null) {
+ task.raisePriority(priority);
+ }
+ }
+
+ public PrioritisedExecutor.Priority getPriority(final int chunkX, final int chunkZ) {
+ final ChunkTasks task;
+ synchronized (this) {
+ task = this.chunkTasks.get(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+ if (task != null) {
+ return task.getPriority();
+ }
+
+ return PrioritisedExecutor.Priority.COMPLETING;
+ }
+
+ public boolean isEmpty() {
+ synchronized (this) {
+ return this.chunkTasks.isEmpty();
+ }
+ }
+
+ public CompletableFuture<Void> queueBlockChange(final BlockPos pos) {
+ final ChunkTasks tasks;
+ synchronized (this) {
+ tasks = this.chunkTasks.computeIfAbsent(CoordinateUtils.getChunkKey(pos), (final long keyInMap) -> {
+ return new ChunkTasks(keyInMap, LightQueue.this.manager, LightQueue.this);
+ });
+ tasks.changedPositions.add(pos.immutable());
+ }
+
+ tasks.schedule();
+
+ return tasks.onComplete;
+ }
+
+ public CompletableFuture<Void> queueSectionChange(final SectionPos pos, final boolean newEmptyValue) {
+ final ChunkTasks tasks;
+ synchronized (this) {
+ tasks = this.chunkTasks.computeIfAbsent(CoordinateUtils.getChunkKey(pos), (final long keyInMap) -> {
+ return new ChunkTasks(keyInMap, LightQueue.this.manager, LightQueue.this);
+ });
+
+ if (tasks.changedSectionSet == null) {
+ tasks.changedSectionSet = new Boolean[this.manager.maxSection - this.manager.minSection + 1];
+ }
+ tasks.changedSectionSet[pos.getY() - this.manager.minSection] = Boolean.valueOf(newEmptyValue);
+ }
+
+ tasks.schedule();
+
+ return tasks.onComplete;
+ }
+
+ public CompletableFuture<Void> queueChunkLightTask(final ChunkPos pos, final BooleanSupplier lightTask, final PrioritisedExecutor.Priority priority) {
+ final ChunkTasks tasks;
+ synchronized (this) {
+ tasks = this.chunkTasks.computeIfAbsent(CoordinateUtils.getChunkKey(pos), (final long keyInMap) -> {
+ return new ChunkTasks(keyInMap, LightQueue.this.manager, LightQueue.this, priority);
+ });
+ if (tasks.lightTasks == null) {
+ tasks.lightTasks = new ArrayList<>();
+ }
+ tasks.lightTasks.add(lightTask);
+ }
+
+ tasks.schedule();
+
+ return tasks.onComplete;
+ }
+
+ public CompletableFuture<Void> queueChunkSkylightEdgeCheck(final SectionPos pos, final ShortCollection sections) {
+ final ChunkTasks tasks;
+ synchronized (this) {
+ tasks = this.chunkTasks.computeIfAbsent(CoordinateUtils.getChunkKey(pos), (final long keyInMap) -> {
+ return new ChunkTasks(keyInMap, LightQueue.this.manager, LightQueue.this);
+ });
+
+ ShortOpenHashSet queuedEdges = tasks.queuedEdgeChecksSky;
+ if (queuedEdges == null) {
+ queuedEdges = tasks.queuedEdgeChecksSky = new ShortOpenHashSet();
+ }
+ queuedEdges.addAll(sections);
+ }
+
+ tasks.schedule();
+
+ return tasks.onComplete;
+ }
+
+ public CompletableFuture<Void> queueChunkBlocklightEdgeCheck(final SectionPos pos, final ShortCollection sections) {
+ final ChunkTasks tasks;
+
+ synchronized (this) {
+ tasks = this.chunkTasks.computeIfAbsent(CoordinateUtils.getChunkKey(pos), (final long keyInMap) -> {
+ return new ChunkTasks(keyInMap, LightQueue.this.manager, LightQueue.this);
+ });
+
+ ShortOpenHashSet queuedEdges = tasks.queuedEdgeChecksBlock;
+ if (queuedEdges == null) {
+ queuedEdges = tasks.queuedEdgeChecksBlock = new ShortOpenHashSet();
+ }
+ queuedEdges.addAll(sections);
+ }
+
+ tasks.schedule();
+
+ return tasks.onComplete;
+ }
+
+ public void removeChunk(final ChunkPos pos) {
+ final ChunkTasks tasks;
+ synchronized (this) {
+ tasks = this.chunkTasks.remove(CoordinateUtils.getChunkKey(pos));
+ }
+ if (tasks != null && tasks.cancel()) {
+ tasks.onComplete.complete(null);
+ }
+ }
+
+ protected static final class ChunkTasks implements Runnable {
+
+ final Set<BlockPos> changedPositions = new HashSet<>();
+ Boolean[] changedSectionSet;
+ ShortOpenHashSet queuedEdgeChecksSky;
+ ShortOpenHashSet queuedEdgeChecksBlock;
+ List<BooleanSupplier> lightTasks;
+
+ final CompletableFuture<Void> onComplete = new CompletableFuture<>();
+
+ public final long chunkCoordinate;
+ private final StarLightInterface lightEngine;
+ private final LightQueue queue;
+ private final PrioritisedExecutor.PrioritisedTask task;
+
+ public ChunkTasks(final long chunkCoordinate, final StarLightInterface lightEngine, final LightQueue queue) {
+ this(chunkCoordinate, lightEngine, queue, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ public ChunkTasks(final long chunkCoordinate, final StarLightInterface lightEngine, final LightQueue queue,
+ final PrioritisedExecutor.Priority priority) {
+ this.chunkCoordinate = chunkCoordinate;
+ this.lightEngine = lightEngine;
+ this.queue = queue;
+ this.task = queue.world.chunkTaskScheduler.lightExecutor.createTask(this, priority);
+ }
+
+ public void schedule() {
+ this.task.queue();
+ }
+
+ public boolean cancel() {
+ return this.task.cancel();
+ }
+
+ public PrioritisedExecutor.Priority getPriority() {
+ return this.task.getPriority();
+ }
+
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ this.task.lowerPriority(priority);
+ }
+
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ this.task.setPriority(priority);
+ }
+
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ this.task.raisePriority(priority);
+ }
+
+ @Override
+ public void run() {
+ final SkyStarLightEngine skyEngine = this.lightEngine.getSkyLightEngine();
+ final BlockStarLightEngine blockEngine = this.lightEngine.getBlockLightEngine();
+ try {
+ synchronized (this.queue) {
+ this.queue.chunkTasks.remove(this.chunkCoordinate);
+ }
+
+ boolean litChunk = false;
+ if (this.lightTasks != null) {
+ for (final BooleanSupplier run : this.lightTasks) {
+ if (run.getAsBoolean()) {
+ litChunk = true;
+ break;
+ }
+ }
+ }
+
+ final long coordinate = this.chunkCoordinate;
+ final int chunkX = CoordinateUtils.getChunkX(coordinate);
+ final int chunkZ = CoordinateUtils.getChunkZ(coordinate);
+
+ final Set<BlockPos> positions = this.changedPositions;
+ final Boolean[] sectionChanges = this.changedSectionSet;
+
+ if (!litChunk) {
+ if (skyEngine != null && (!positions.isEmpty() || sectionChanges != null)) {
+ skyEngine.blocksChangedInChunk(this.lightEngine.getLightAccess(), chunkX, chunkZ, positions, sectionChanges);
+ }
+ if (blockEngine != null && (!positions.isEmpty() || sectionChanges != null)) {
+ blockEngine.blocksChangedInChunk(this.lightEngine.getLightAccess(), chunkX, chunkZ, positions, sectionChanges);
+ }
+
+ if (skyEngine != null && this.queuedEdgeChecksSky != null) {
+ skyEngine.checkChunkEdges(this.lightEngine.getLightAccess(), chunkX, chunkZ, this.queuedEdgeChecksSky);
+ }
+ if (blockEngine != null && this.queuedEdgeChecksBlock != null) {
+ blockEngine.checkChunkEdges(this.lightEngine.getLightAccess(), chunkX, chunkZ, this.queuedEdgeChecksBlock);
+ }
+ }
+
+ this.onComplete.complete(null);
+ } finally {
+ this.lightEngine.releaseSkyLightEngine(skyEngine);
+ this.lightEngine.releaseBlockLightEngine(blockEngine);
+ }
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/poi/PoiChunk.java b/src/main/java/io/papermc/paper/chunk/system/poi/PoiChunk.java
new file mode 100644
2023-03-15 20:17:42 +01:00
index 0000000000000000000000000000000000000000..d72041aa814ff179e6e29a45dcd359a91d426d47
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/poi/PoiChunk.java
@@ -0,0 +1,213 @@
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+package io.papermc.paper.chunk.system.poi;
+
+import com.mojang.logging.LogUtils;
+import com.mojang.serialization.Codec;
+import com.mojang.serialization.DataResult;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import io.papermc.paper.util.WorldUtil;
+import net.minecraft.SharedConstants;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.nbt.NbtOps;
+import net.minecraft.nbt.Tag;
+import net.minecraft.resources.RegistryOps;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.entity.ai.village.poi.PoiManager;
+import net.minecraft.world.entity.ai.village.poi.PoiSection;
+import org.slf4j.Logger;
+
+import java.util.Optional;
+
+public final class PoiChunk {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ public final ServerLevel world;
+ public final int chunkX;
+ public final int chunkZ;
+ public final int minSection;
+ public final int maxSection;
+
+ protected final PoiSection[] sections;
+
+ private boolean isDirty;
+ private boolean loaded;
+
+ public PoiChunk(final ServerLevel world, final int chunkX, final int chunkZ, final int minSection, final int maxSection) {
+ this(world, chunkX, chunkZ, minSection, maxSection, new PoiSection[maxSection - minSection + 1]);
+ }
+
+ public PoiChunk(final ServerLevel world, final int chunkX, final int chunkZ, final int minSection, final int maxSection, final PoiSection[] sections) {
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.minSection = minSection;
+ this.maxSection = maxSection;
+ this.sections = sections;
+ if (this.sections.length != (maxSection - minSection + 1)) {
+ throw new IllegalStateException("Incorrect length used, expected " + (maxSection - minSection + 1) + ", got " + this.sections.length);
+ }
+ }
+
+ public void load() {
+ TickThread.ensureTickThread(this.world, this.chunkX, this.chunkZ, "Loading in poi chunk off-main");
+ if (this.loaded) {
+ return;
+ }
+ this.loaded = true;
+ this.world.chunkSource.getPoiManager().loadInPoiChunk(this);
+ }
+
+ public boolean isLoaded() {
+ return this.loaded;
+ }
+
+ public boolean isEmpty() {
+ for (final PoiSection section : this.sections) {
+ if (section != null && !section.isEmpty()) {
+ return false;
+ }
+ }
+
+ return true;
+ }
+
+ public PoiSection getOrCreateSection(final int chunkY) {
+ if (chunkY >= this.minSection && chunkY <= this.maxSection) {
+ final int idx = chunkY - this.minSection;
+ final PoiSection ret = this.sections[idx];
+ if (ret != null) {
+ return ret;
+ }
+
+ final PoiManager poiManager = this.world.getPoiManager();
+ final long key = CoordinateUtils.getChunkSectionKey(this.chunkX, chunkY, this.chunkZ);
+
+ return this.sections[idx] = new PoiSection(() -> {
+ poiManager.setDirty(key);
+ });
+ }
+ throw new IllegalArgumentException("chunkY is out of bounds, chunkY: " + chunkY + " outside [" + this.minSection + "," + this.maxSection + "]");
+ }
+
+ public PoiSection getSection(final int chunkY) {
+ if (chunkY >= this.minSection && chunkY <= this.maxSection) {
+ return this.sections[chunkY - this.minSection];
+ }
+ return null;
+ }
+
+ public Optional<PoiSection> getSectionForVanilla(final int chunkY) {
+ if (chunkY >= this.minSection && chunkY <= this.maxSection) {
+ final PoiSection ret = this.sections[chunkY - this.minSection];
+ return ret == null ? Optional.empty() : ret.noAllocateOptional;
+ }
+ return Optional.empty();
+ }
+
+ public boolean isDirty() {
+ return this.isDirty;
+ }
+
+ public void setDirty(final boolean dirty) {
+ this.isDirty = dirty;
+ }
+
+ // returns null if empty
+ public CompoundTag save() {
+ final RegistryOps<Tag> registryOps = RegistryOps.create(NbtOps.INSTANCE, world.getPoiManager().registryAccess);
+
+ final CompoundTag ret = new CompoundTag();
+ final CompoundTag sections = new CompoundTag();
+ ret.put("Sections", sections);
+
2023-03-15 20:17:42 +01:00
+ ret.putInt("DataVersion", SharedConstants.getCurrentVersion().getDataVersion().getVersion());
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ final ServerLevel world = this.world;
+ final PoiManager poiManager = world.getPoiManager();
+ final int chunkX = this.chunkX;
+ final int chunkZ = this.chunkZ;
+
+ for (int sectionY = this.minSection; sectionY <= this.maxSection; ++sectionY) {
+ final PoiSection chunk = this.sections[sectionY - this.minSection];
+ if (chunk == null || chunk.isEmpty()) {
+ continue;
+ }
+
+ final long key = CoordinateUtils.getChunkSectionKey(chunkX, sectionY, chunkZ);
+ // codecs are honestly such a fucking disaster. What the fuck is this trash?
+ final Codec<PoiSection> codec = PoiSection.codec(() -> {
+ poiManager.setDirty(key);
+ });
+
+ final DataResult<Tag> serializedResult = codec.encodeStart(registryOps, chunk);
+ final int finalSectionY = sectionY;
+ final Tag serialized = serializedResult.resultOrPartial((final String description) -> {
+ LOGGER.error("Failed to serialize poi chunk for world: " + world.getWorld().getName() + ", chunk: (" + chunkX + "," + finalSectionY + "," + chunkZ + "); description: " + description);
+ }).orElse(null);
+ if (serialized == null) {
+ // failed, should be logged from the resultOrPartial
+ continue;
+ }
+
+ sections.put(Integer.toString(sectionY), serialized);
+ }
+
+ return sections.isEmpty() ? null : ret;
+ }
+
+ public static PoiChunk empty(final ServerLevel world, final int chunkX, final int chunkZ) {
+ final PoiChunk ret = new PoiChunk(world, chunkX, chunkZ, WorldUtil.getMinSection(world), WorldUtil.getMaxSection(world));
+ ret.loaded = true;
+ return ret;
+ }
+
+ public static PoiChunk parse(final ServerLevel world, final int chunkX, final int chunkZ, final CompoundTag data) {
+ final PoiChunk ret = empty(world, chunkX, chunkZ);
+
+ final RegistryOps<Tag> registryOps = RegistryOps.create(NbtOps.INSTANCE, world.getPoiManager().registryAccess);
+
+ final CompoundTag sections = data.getCompound("Sections");
+
+ if (sections.isEmpty()) {
+ // nothing to parse
+ return ret;
+ }
+
+ final PoiManager poiManager = world.getPoiManager();
+
+ boolean readAnything = false;
+
+ for (int sectionY = ret.minSection; sectionY <= ret.maxSection; ++sectionY) {
+ final String key = Integer.toString(sectionY);
+ if (!sections.contains(key)) {
+ continue;
+ }
+
+ final long coordinateKey = CoordinateUtils.getChunkSectionKey(chunkX, sectionY, chunkZ);
+ // codecs are honestly such a fucking disaster. What the fuck is this trash?
+ final Codec<PoiSection> codec = PoiSection.codec(() -> {
+ poiManager.setDirty(coordinateKey);
+ });
+
+ final CompoundTag section = sections.getCompound(key);
+ final DataResult<PoiSection> deserializeResult = codec.parse(registryOps, section);
+ final int finalSectionY = sectionY;
+ final PoiSection deserialized = deserializeResult.resultOrPartial((final String description) -> {
+ LOGGER.error("Failed to deserialize poi chunk for world: " + world.getWorld().getName() + ", chunk: (" + chunkX + "," + finalSectionY + "," + chunkZ + "); description: " + description);
+ }).orElse(null);
+
+ if (deserialized == null || deserialized.isEmpty()) {
+ // completely empty, no point in storing this
+ continue;
+ }
+
+ readAnything = true;
+ ret.sections[sectionY - ret.minSection] = deserialized;
+ }
+
+ ret.loaded = !readAnything; // Set loaded to false if we read anything to ensure proper callbacks to PoiManager are made on #load
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return ret;
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkFullTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkFullTask.java
new file mode 100644
index 0000000000000000000000000000000000000000..fb42d776f15f735fb59e972e00e2b512c23a8387
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkFullTask.java
@@ -0,0 +1,121 @@
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.ImposterProtoChunk;
+import net.minecraft.world.level.chunk.LevelChunk;
+import net.minecraft.world.level.chunk.ProtoChunk;
+import java.lang.invoke.VarHandle;
+
+public final class ChunkFullTask extends ChunkProgressionTask implements Runnable {
+
+ protected final NewChunkHolder chunkHolder;
+ protected final ChunkAccess fromChunk;
+ protected final PrioritisedExecutor.PrioritisedTask convertToFullTask;
+
+ public ChunkFullTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX, final int chunkZ,
+ final NewChunkHolder chunkHolder, final ChunkAccess fromChunk, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ);
+ this.chunkHolder = chunkHolder;
+ this.fromChunk = fromChunk;
+ this.convertToFullTask = scheduler.createChunkTask(chunkX, chunkZ, this, priority);
+ }
+
+ @Override
+ public ChunkStatus getTargetStatus() {
+ return ChunkStatus.FULL;
+ }
+
+ @Override
+ public void run() {
+ // See Vanilla protoChunkToFullChunk for what this function should be doing
+ final LevelChunk chunk;
+ try {
+ if (this.fromChunk instanceof ImposterProtoChunk wrappedFull) {
+ chunk = wrappedFull.getWrapped();
+ } else {
+ final ServerLevel world = this.world;
+ final ProtoChunk protoChunk = (ProtoChunk)this.fromChunk;
+ chunk = new LevelChunk(this.world, protoChunk, (final LevelChunk unused) -> {
+ ChunkMap.postLoadProtoChunk(world, protoChunk.getEntities());
+ });
+ }
+
+ chunk.setChunkHolder(this.scheduler.chunkHolderManager.getChunkHolder(this.chunkX, this.chunkZ)); // replaces setFullStatus
+ chunk.runPostLoad();
+ // Unlike Vanilla, we load the entity chunk here, as we load the NBT in empty status (unlike Vanilla)
+ // This brings entity addition back in line with older versions of the game
+ // Since we load the NBT in the empty status, this will never block for I/O
+ this.world.chunkTaskScheduler.chunkHolderManager.getOrCreateEntityChunk(this.chunkX, this.chunkZ, false);
+
+ // we don't need the entitiesInLevel trash, this system doesn't double run callbacks
+ chunk.setLoaded(true);
+ chunk.registerAllBlockEntitiesAfterLevelLoad();
+ chunk.registerTickContainerInLevel(this.world);
+ } catch (final Throwable throwable) {
+ this.complete(null, throwable);
+
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ return;
+ }
+ this.complete(chunk, null);
+ }
+
+ protected volatile boolean scheduled;
+ protected static final VarHandle SCHEDULED_HANDLE = ConcurrentUtil.getVarHandle(ChunkFullTask.class, "scheduled", boolean.class);
+
+ @Override
+ public boolean isScheduled() {
+ return this.scheduled;
+ }
+
+ @Override
+ public void schedule() {
+ if ((boolean)SCHEDULED_HANDLE.getAndSet((ChunkFullTask)this, true)) {
+ throw new IllegalStateException("Cannot double call schedule()");
+ }
+ this.convertToFullTask.queue();
+ }
+
+ @Override
+ public void cancel() {
+ if (this.convertToFullTask.cancel()) {
+ this.complete(null, null);
+ }
+ }
+
+ @Override
+ public PrioritisedExecutor.Priority getPriority() {
+ return this.convertToFullTask.getPriority();
+ }
+
+ @Override
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.convertToFullTask.lowerPriority(priority);
+ }
+
+ @Override
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.convertToFullTask.setPriority(priority);
+ }
+
+ @Override
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.convertToFullTask.raisePriority(priority);
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkHolderManager.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkHolderManager.java
new file mode 100644
index 0000000000000000000000000000000000000000..e4dcadc24b3d73178ee1a4b64b8c6343e5285e59
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkHolderManager.java
@@ -0,0 +1,1204 @@
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.map.SWMRLong2ObjectHashTable;
+import co.aikar.timings.Timing;
+import com.google.common.collect.ImmutableList;
+import com.google.gson.JsonArray;
+import com.google.gson.JsonObject;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.chunk.system.io.RegionFileIOThread;
+import io.papermc.paper.chunk.system.poi.PoiChunk;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import io.papermc.paper.util.misc.Delayed8WayDistancePropagator2D;
+import io.papermc.paper.world.ChunkEntitySlices;
+import it.unimi.dsi.fastutil.longs.Long2IntLinkedOpenHashMap;
+import it.unimi.dsi.fastutil.longs.Long2IntMap;
+import it.unimi.dsi.fastutil.longs.Long2IntOpenHashMap;
+import it.unimi.dsi.fastutil.longs.Long2ObjectMap;
+import it.unimi.dsi.fastutil.longs.Long2ObjectOpenHashMap;
+import it.unimi.dsi.fastutil.longs.LongArrayList;
+import it.unimi.dsi.fastutil.longs.LongIterator;
+import it.unimi.dsi.fastutil.objects.ObjectRBTreeSet;
+import it.unimi.dsi.fastutil.objects.ReferenceLinkedOpenHashSet;
+import net.minecraft.nbt.CompoundTag;
+import io.papermc.paper.chunk.system.ChunkSystem;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.server.MinecraftServer;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.server.level.Ticket;
+import net.minecraft.server.level.TicketType;
+import net.minecraft.util.SortedArraySet;
+import net.minecraft.util.Unit;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import org.bukkit.plugin.Plugin;
+import org.slf4j.Logger;
+import java.io.IOException;
+import java.text.DecimalFormat;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.Collection;
+import java.util.Collections;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Objects;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicReference;
+import java.util.concurrent.locks.LockSupport;
+import java.util.concurrent.locks.ReentrantLock;
+import java.util.function.Predicate;
+
+public final class ChunkHolderManager {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ public static final int FULL_LOADED_TICKET_LEVEL = 33;
+ public static final int BLOCK_TICKING_TICKET_LEVEL = 32;
+ public static final int ENTITY_TICKING_TICKET_LEVEL = 31;
+ public static final int MAX_TICKET_LEVEL = ChunkMap.MAX_CHUNK_DISTANCE; // inclusive
+
+ private static final long NO_TIMEOUT_MARKER = -1L;
+
+ final ReentrantLock ticketLock = new ReentrantLock();
+
+ private final SWMRLong2ObjectHashTable<NewChunkHolder> chunkHolders = new SWMRLong2ObjectHashTable<>(16384, 0.25f);
+ private final Long2ObjectOpenHashMap<SortedArraySet<Ticket<?>>> tickets = new Long2ObjectOpenHashMap<>(8192, 0.25f);
+ // what a disaster of a name
+ // this is a map of removal tick to a map of chunks and the number of tickets a chunk has that are to expire that tick
+ private final Long2ObjectOpenHashMap<Long2IntOpenHashMap> removeTickToChunkExpireTicketCount = new Long2ObjectOpenHashMap<>();
+ private final ServerLevel world;
+ private final ChunkTaskScheduler taskScheduler;
+ private long currentTick;
+
+ private final ArrayDeque<NewChunkHolder> pendingFullLoadUpdate = new ArrayDeque<>();
+ private final ObjectRBTreeSet<NewChunkHolder> autoSaveQueue = new ObjectRBTreeSet<>((final NewChunkHolder c1, final NewChunkHolder c2) -> {
+ if (c1 == c2) {
+ return 0;
+ }
+
+ final int saveTickCompare = Long.compare(c1.lastAutoSave, c2.lastAutoSave);
+
+ if (saveTickCompare != 0) {
+ return saveTickCompare;
+ }
+
+ final long coord1 = CoordinateUtils.getChunkKey(c1.chunkX, c1.chunkZ);
+ final long coord2 = CoordinateUtils.getChunkKey(c2.chunkX, c2.chunkZ);
+
+ if (coord1 == coord2) {
+ throw new IllegalStateException("Duplicate chunkholder in auto save queue");
+ }
+
+ return Long.compare(coord1, coord2);
+ });
+
+ public ChunkHolderManager(final ServerLevel world, final ChunkTaskScheduler taskScheduler) {
+ this.world = world;
+ this.taskScheduler = taskScheduler;
+ }
+
+ private long statusUpgradeId;
+
+ long getNextStatusUpgradeId() {
+ return ++this.statusUpgradeId;
+ }
+
+ public List<ChunkHolder> getOldChunkHolders() {
+ final List<NewChunkHolder> holders = this.getChunkHolders();
+ final List<ChunkHolder> ret = new ArrayList<>(holders.size());
+ for (final NewChunkHolder holder : holders) {
+ ret.add(holder.vanillaChunkHolder);
+ }
+ return ret;
+ }
+
+ public List<NewChunkHolder> getChunkHolders() {
+ final List<NewChunkHolder> ret = new ArrayList<>(this.chunkHolders.size());
+ this.chunkHolders.forEachValue(ret::add);
+ return ret;
+ }
+
+ public int size() {
+ return this.chunkHolders.size();
+ }
+
+ public void close(final boolean save, final boolean halt) {
+ TickThread.ensureTickThread("Closing world off-main");
+ if (halt) {
+ LOGGER.info("Waiting 60s for chunk system to halt for world '" + this.world.getWorld().getName() + "'");
+ if (!this.taskScheduler.halt(true, TimeUnit.SECONDS.toNanos(60L))) {
+ LOGGER.warn("Failed to halt world generation/loading tasks for world '" + this.world.getWorld().getName() + "'");
+ } else {
+ LOGGER.info("Halted chunk system for world '" + this.world.getWorld().getName() + "'");
+ }
+ }
+
+ if (save) {
+ this.saveAllChunks(true, true, true);
+ }
+
+ if (this.world.chunkDataControllerNew.hasTasks() || this.world.entityDataControllerNew.hasTasks() || this.world.poiDataControllerNew.hasTasks()) {
+ RegionFileIOThread.flush();
+ }
+
+ // kill regionfile cache
+ try {
+ this.world.chunkDataControllerNew.getCache().close();
+ } catch (final IOException ex) {
+ LOGGER.error("Failed to close chunk regionfile cache for world '" + this.world.getWorld().getName() + "'", ex);
+ }
+ try {
+ this.world.entityDataControllerNew.getCache().close();
+ } catch (final IOException ex) {
+ LOGGER.error("Failed to close entity regionfile cache for world '" + this.world.getWorld().getName() + "'", ex);
+ }
+ try {
+ this.world.poiDataControllerNew.getCache().close();
+ } catch (final IOException ex) {
+ LOGGER.error("Failed to close poi regionfile cache for world '" + this.world.getWorld().getName() + "'", ex);
+ }
+ }
+
+ void ensureInAutosave(final NewChunkHolder holder) {
+ if (!this.autoSaveQueue.contains(holder)) {
+ holder.lastAutoSave = MinecraftServer.currentTick;
+ this.autoSaveQueue.add(holder);
+ }
+ }
+
+ public void autoSave() {
+ final List<NewChunkHolder> reschedule = new ArrayList<>();
+ final long currentTick = MinecraftServer.currentTickLong;
+ final long maxSaveTime = currentTick - this.world.paperConfig().chunks.autoSaveInterval.value();
+ for (int autoSaved = 0; autoSaved < this.world.paperConfig().chunks.maxAutoSaveChunksPerTick && !this.autoSaveQueue.isEmpty();) {
+ final NewChunkHolder holder = this.autoSaveQueue.first();
+
+ if (holder.lastAutoSave > maxSaveTime) {
+ break;
+ }
+
+ this.autoSaveQueue.remove(holder);
+
+ holder.lastAutoSave = currentTick;
+ if (holder.save(false, false) != null) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ++autoSaved;
+ }
+
+ if (holder.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ reschedule.add(holder);
+ }
+ }
+
+ for (final NewChunkHolder holder : reschedule) {
+ if (holder.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ this.autoSaveQueue.add(holder);
+ }
+ }
+ }
+
+ public void saveAllChunks(final boolean flush, final boolean shutdown, final boolean logProgress) {
+ final List<NewChunkHolder> holders = this.getChunkHolders();
+
+ if (logProgress) {
+ LOGGER.info("Saving all chunkholders for world '" + this.world.getWorld().getName() + "'");
+ }
+
+ final DecimalFormat format = new DecimalFormat("#0.00");
+
+ int saved = 0;
+
+ long start = System.nanoTime();
+ long lastLog = start;
+ boolean needsFlush = false;
+ final int flushInterval = 50;
+
+ int savedChunk = 0;
+ int savedEntity = 0;
+ int savedPoi = 0;
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ for (int i = 0, len = holders.size(); i < len; ++i) {
+ final NewChunkHolder holder = holders.get(i);
+ try {
+ final NewChunkHolder.SaveStat saveStat = holder.save(shutdown, false);
+ if (saveStat != null) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ++saved;
+ needsFlush = flush;
+ if (saveStat.savedChunk()) {
+ ++savedChunk;
+ }
+ if (saveStat.savedEntityChunk()) {
+ ++savedEntity;
+ }
+ if (saveStat.savedPoiChunk()) {
+ ++savedPoi;
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to save chunk (" + holder.chunkX + "," + holder.chunkZ + ") in world '" + this.world.getWorld().getName() + "'", thr);
+ }
+ if (needsFlush && (saved % flushInterval) == 0) {
+ needsFlush = false;
+ RegionFileIOThread.partialFlush(flushInterval / 2);
+ }
+ if (logProgress) {
+ final long currTime = System.nanoTime();
+ if ((currTime - lastLog) > TimeUnit.SECONDS.toNanos(10L)) {
+ lastLog = currTime;
+ LOGGER.info("Saved " + saved + " chunks (" + format.format((double)(i+1)/(double)len * 100.0) + "%) in world '" + this.world.getWorld().getName() + "'");
+ }
+ }
+ }
+ if (flush) {
+ RegionFileIOThread.flush();
+ }
+ if (logProgress) {
+ LOGGER.info("Saved " + savedChunk + " block chunks, " + savedEntity + " entity chunks, " + savedPoi + " poi chunks in world '" + this.world.getWorld().getName() + "' in " + format.format(1.0E-9 * (System.nanoTime() - start)) + "s");
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+ }
+
+ protected final Long2IntLinkedOpenHashMap ticketLevelUpdates = new Long2IntLinkedOpenHashMap() {
+ @Override
+ protected void rehash(final int newN) {
+ // no downsizing allowed
+ if (newN < this.n) {
+ return;
+ }
+ super.rehash(newN);
+ }
+ };
+
+ protected final Delayed8WayDistancePropagator2D ticketLevelPropagator = new Delayed8WayDistancePropagator2D(
+ (final long coordinate, final byte oldLevel, final byte newLevel) -> {
+ ChunkHolderManager.this.ticketLevelUpdates.putAndMoveToLast(coordinate, convertBetweenTicketLevels(newLevel));
+ }
+ );
+ // function for converting between ticket levels and propagator levels and vice versa
+ // the problem is the ticket level propagator will propagate from a set source down to zero, whereas mojang expects
+ // levels to propagate from a set value up to a maximum value. so we need to convert the levels we put into the propagator
+ // and the levels we get out of the propagator
+
+ public static int convertBetweenTicketLevels(final int level) {
+ return ChunkMap.MAX_CHUNK_DISTANCE - level + 1;
+ }
+
+ public boolean hasTickets() {
+ this.ticketLock.lock();
+ try {
+ return !this.tickets.isEmpty();
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public String getTicketDebugString(final long coordinate) {
+ this.ticketLock.lock();
+ try {
+ final SortedArraySet<Ticket<?>> tickets = this.tickets.get(coordinate);
+
+ return tickets != null ? tickets.first().toString() : "no_ticket";
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public Long2ObjectOpenHashMap<SortedArraySet<Ticket<?>>> getTicketsCopy() {
+ this.ticketLock.lock();
+ try {
+ return this.tickets.clone();
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public Collection<Plugin> getPluginChunkTickets(int x, int z) {
+ ImmutableList.Builder<Plugin> ret;
+ this.ticketLock.lock();
+ try {
+ SortedArraySet<Ticket<?>> tickets = this.tickets.get(ChunkPos.asLong(x, z));
+
+ if (tickets == null) {
+ return Collections.emptyList();
+ }
+
+ ret = ImmutableList.builder();
+ for (Ticket<?> ticket : tickets) {
+ if (ticket.getType() == TicketType.PLUGIN_TICKET) {
+ ret.add((Plugin)ticket.key);
+ }
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+
+ return ret.build();
+ }
+
+ protected final int getPropagatedTicketLevel(final long coordinate) {
+ return convertBetweenTicketLevels(this.ticketLevelPropagator.getLevel(coordinate));
+ }
+
+ protected final void updateTicketLevel(final long coordinate, final int ticketLevel) {
+ if (ticketLevel > ChunkMap.MAX_CHUNK_DISTANCE) {
+ this.ticketLevelPropagator.removeSource(coordinate);
+ } else {
+ this.ticketLevelPropagator.setSource(coordinate, convertBetweenTicketLevels(ticketLevel));
+ }
+ }
+
+ private static int getTicketLevelAt(SortedArraySet<Ticket<?>> tickets) {
+ return !tickets.isEmpty() ? tickets.first().getTicketLevel() : MAX_TICKET_LEVEL + 1;
+ }
+
+ public <T> boolean addTicketAtLevel(final TicketType<T> type, final ChunkPos chunkPos, final int level,
+ final T identifier) {
+ return this.addTicketAtLevel(type, CoordinateUtils.getChunkKey(chunkPos), level, identifier);
+ }
+
+ public <T> boolean addTicketAtLevel(final TicketType<T> type, final int chunkX, final int chunkZ, final int level,
+ final T identifier) {
+ return this.addTicketAtLevel(type, CoordinateUtils.getChunkKey(chunkX, chunkZ), level, identifier);
+ }
+
+ // supposed to return true if the ticket was added and did not replace another
+ // but, we always return false if the ticket cannot be added
+ public <T> boolean addTicketAtLevel(final TicketType<T> type, final long chunk, final int level, final T identifier) {
+ final long removeDelay = Math.max(0, type.timeout);
+ if (level > MAX_TICKET_LEVEL) {
+ return false;
+ }
+
+ this.ticketLock.lock();
+ try {
+ final long removeTick = removeDelay == 0 ? NO_TIMEOUT_MARKER : this.currentTick + removeDelay;
+ final Ticket<T> ticket = new Ticket<>(type, level, identifier, removeTick);
+
+ final SortedArraySet<Ticket<?>> ticketsAtChunk = this.tickets.computeIfAbsent(chunk, (final long keyInMap) -> {
+ return SortedArraySet.create(4);
+ });
+
+ final int levelBefore = getTicketLevelAt(ticketsAtChunk);
+ final Ticket<T> current = (Ticket<T>)ticketsAtChunk.replace(ticket);
+ final int levelAfter = getTicketLevelAt(ticketsAtChunk);
+
+ if (current != ticket) {
+ final long oldRemovalTick = current.removalTick;
+ if (removeTick != oldRemovalTick) {
+ if (oldRemovalTick != NO_TIMEOUT_MARKER) {
+ final Long2IntOpenHashMap removeCounts = this.removeTickToChunkExpireTicketCount.get(oldRemovalTick);
+ final int prevCount = removeCounts.addTo(chunk, -1);
+
+ if (prevCount == 1) {
+ removeCounts.remove(chunk);
+ if (removeCounts.isEmpty()) {
+ this.removeTickToChunkExpireTicketCount.remove(oldRemovalTick);
+ }
+ }
+ }
+ if (removeTick != NO_TIMEOUT_MARKER) {
+ this.removeTickToChunkExpireTicketCount.computeIfAbsent(removeTick, (final long keyInMap) -> {
+ return new Long2IntOpenHashMap();
+ }).addTo(chunk, 1);
+ }
+ }
+ } else {
+ if (removeTick != NO_TIMEOUT_MARKER) {
+ this.removeTickToChunkExpireTicketCount.computeIfAbsent(removeTick, (final long keyInMap) -> {
+ return new Long2IntOpenHashMap();
+ }).addTo(chunk, 1);
+ }
+ }
+
+ if (levelBefore != levelAfter) {
+ this.updateTicketLevel(chunk, levelAfter);
+ }
+
+ return current == ticket;
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public <T> boolean removeTicketAtLevel(final TicketType<T> type, final ChunkPos chunkPos, final int level, final T identifier) {
+ return this.removeTicketAtLevel(type, CoordinateUtils.getChunkKey(chunkPos), level, identifier);
+ }
+
+ public <T> boolean removeTicketAtLevel(final TicketType<T> type, final int chunkX, final int chunkZ, final int level, final T identifier) {
+ return this.removeTicketAtLevel(type, CoordinateUtils.getChunkKey(chunkX, chunkZ), level, identifier);
+ }
+
+ public <T> boolean removeTicketAtLevel(final TicketType<T> type, final long chunk, final int level, final T identifier) {
+ if (level > MAX_TICKET_LEVEL) {
+ return false;
+ }
+
+ this.ticketLock.lock();
+ try {
+ final SortedArraySet<Ticket<?>> ticketsAtChunk = this.tickets.get(chunk);
+ if (ticketsAtChunk == null) {
+ return false;
+ }
+
+ final int oldLevel = getTicketLevelAt(ticketsAtChunk);
+ final Ticket<T> ticket = (Ticket<T>)ticketsAtChunk.removeAndGet(new Ticket<>(type, level, identifier, -2L));
+
+ if (ticket == null) {
+ return false;
+ }
+
+ if (ticketsAtChunk.isEmpty()) {
+ this.tickets.remove(chunk);
+ }
+
+ final int newLevel = getTicketLevelAt(ticketsAtChunk);
+
+ final long removeTick = ticket.removalTick;
+ if (removeTick != NO_TIMEOUT_MARKER) {
+ final Long2IntOpenHashMap removeCounts = this.removeTickToChunkExpireTicketCount.get(removeTick);
+ final int currCount = removeCounts.addTo(chunk, -1);
+
+ if (currCount == 1) {
+ removeCounts.remove(chunk);
+ if (removeCounts.isEmpty()) {
+ this.removeTickToChunkExpireTicketCount.remove(removeTick);
+ }
+ }
+ }
+
+ if (oldLevel != newLevel) {
+ this.updateTicketLevel(chunk, newLevel);
+ }
+
+ return true;
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ // atomic with respect to all add/remove/addandremove ticket calls for the given chunk
+ public <T, V> void addAndRemoveTickets(final long chunk, final TicketType<T> addType, final int addLevel, final T addIdentifier,
+ final TicketType<V> removeType, final int removeLevel, final V removeIdentifier) {
+ this.ticketLock.lock();
+ try {
+ this.addTicketAtLevel(addType, chunk, addLevel, addIdentifier);
+ this.removeTicketAtLevel(removeType, chunk, removeLevel, removeIdentifier);
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public <T> void removeAllTicketsFor(final TicketType<T> ticketType, final int ticketLevel, final T ticketIdentifier) {
+ if (ticketLevel > MAX_TICKET_LEVEL) {
+ return;
+ }
+
+ this.ticketLock.lock();
+ try {
+ for (final LongIterator iterator = new LongArrayList(this.tickets.keySet()).longIterator(); iterator.hasNext();) {
+ final long chunk = iterator.nextLong();
+
+ this.removeTicketAtLevel(ticketType, chunk, ticketLevel, ticketIdentifier);
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ public void tick() {
+ TickThread.ensureTickThread("Cannot tick ticket manager off-main");
+
+ this.ticketLock.lock();
+ try {
+ final long tick = ++this.currentTick;
+
+ final Long2IntOpenHashMap toRemove = this.removeTickToChunkExpireTicketCount.remove(tick);
+
+ if (toRemove == null) {
+ return;
+ }
+
+ final Predicate<Ticket<?>> expireNow = (final Ticket<?> ticket) -> {
+ return ticket.removalTick == tick;
+ };
+
+ for (final LongIterator iterator = toRemove.keySet().longIterator(); iterator.hasNext();) {
+ final long chunk = iterator.nextLong();
+
+ final SortedArraySet<Ticket<?>> tickets = this.tickets.get(chunk);
+ tickets.removeIf(expireNow);
+ if (tickets.isEmpty()) {
+ this.tickets.remove(chunk);
+ this.ticketLevelPropagator.removeSource(chunk);
+ } else {
+ this.ticketLevelPropagator.setSource(chunk, convertBetweenTicketLevels(tickets.first().getTicketLevel()));
+ }
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+
+ this.processTicketUpdates();
+ }
+
+ public NewChunkHolder getChunkHolder(final int chunkX, final int chunkZ) {
+ return this.chunkHolders.get(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+
+ public NewChunkHolder getChunkHolder(final long position) {
+ return this.chunkHolders.get(position);
+ }
+
+ public void raisePriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ final NewChunkHolder chunkHolder = this.getChunkHolder(x, z);
+ if (chunkHolder != null) {
+ chunkHolder.raisePriority(priority);
+ }
+ }
+
+ public void setPriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ final NewChunkHolder chunkHolder = this.getChunkHolder(x, z);
+ if (chunkHolder != null) {
+ chunkHolder.setPriority(priority);
+ }
+ }
+
+ public void lowerPriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ final NewChunkHolder chunkHolder = this.getChunkHolder(x, z);
+ if (chunkHolder != null) {
+ chunkHolder.lowerPriority(priority);
+ }
+ }
+
+ private NewChunkHolder createChunkHolder(final long position) {
+ final NewChunkHolder ret = new NewChunkHolder(this.world, CoordinateUtils.getChunkX(position), CoordinateUtils.getChunkZ(position), this.taskScheduler);
+
+ ChunkSystem.onChunkHolderCreate(this.world, ret.vanillaChunkHolder);
+ ret.vanillaChunkHolder.onChunkAdd();
+
+ return ret;
+ }
+
+ // because this function creates the chunk holder without a ticket, it is the caller's responsibility to ensure
+ // the chunk holder eventually unloads. this should only be used to avoid using processTicketUpdates to create chunkholders,
+ // as processTicketUpdates may call plugin logic; in every other case a ticket is appropriate
+ private NewChunkHolder getOrCreateChunkHolder(final int chunkX, final int chunkZ) {
+ return this.getOrCreateChunkHolder(CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ }
+
+ private NewChunkHolder getOrCreateChunkHolder(final long position) {
+ if (!this.ticketLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Must hold ticket level update lock!");
+ }
+ if (!this.taskScheduler.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Must hold scheduler lock!!");
+ }
+
+ // we could just acquire these locks, but...
+ // must own the locks because the caller needs to ensure that no unload can occur AFTER this function returns
+
+ NewChunkHolder current = this.chunkHolders.get(position);
+ if (current != null) {
+ return current;
+ }
+
+ current = this.createChunkHolder(position);
+ this.chunkHolders.put(position, current);
+
+ return current;
+ }
+
+ private long entityLoadCounter;
+
+ public ChunkEntitySlices getOrCreateEntityChunk(final int chunkX, final int chunkZ, final boolean transientChunk) {
+ TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Cannot create entity chunk off-main");
+ ChunkEntitySlices ret;
+
+ NewChunkHolder current = this.getChunkHolder(chunkX, chunkZ);
+ if (current != null && (ret = current.getEntityChunk()) != null && (transientChunk || !ret.isTransient())) {
+ return ret;
+ }
+
+ final AtomicBoolean isCompleted = new AtomicBoolean();
+ final Thread waiter = Thread.currentThread();
+ final Long entityLoadId;
+ NewChunkHolder.GenericDataLoadTaskCallback loadTask = null;
+ this.ticketLock.lock();
+ try {
+ entityLoadId = Long.valueOf(this.entityLoadCounter++);
+ this.addTicketAtLevel(TicketType.ENTITY_LOAD, chunkX, chunkZ, MAX_TICKET_LEVEL, entityLoadId);
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ current = this.getOrCreateChunkHolder(chunkX, chunkZ);
+ if ((ret = current.getEntityChunk()) != null && (transientChunk || !ret.isTransient())) {
+ this.removeTicketAtLevel(TicketType.ENTITY_LOAD, chunkX, chunkZ, MAX_TICKET_LEVEL, entityLoadId);
+ return ret;
+ }
+
+ if (current.isEntityChunkNBTLoaded()) {
+ isCompleted.setPlain(true);
+ } else {
+ loadTask = current.getOrLoadEntityData((final GenericDataLoadTask.TaskResult<CompoundTag, Throwable> result) -> {
+ if (!transientChunk) {
+ isCompleted.set(true);
+ LockSupport.unpark(waiter);
+ }
+ });
+ final ChunkLoadTask.EntityDataLoadTask entityLoad = current.getEntityDataLoadTask();
+
+ if (entityLoad != null && !transientChunk) {
+ entityLoad.raisePriority(PrioritisedExecutor.Priority.BLOCKING);
+ }
+ }
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+
+ if (loadTask != null) {
+ loadTask.schedule();
+ }
+
+ if (!transientChunk) {
+ // Note: no need to busy wait on the chunk queue, entity load will complete off-main
+ boolean interrupted = false;
+ while (!isCompleted.get()) {
+ interrupted |= Thread.interrupted();
+ LockSupport.park();
+ }
+
+ if (interrupted) {
+ Thread.currentThread().interrupt();
+ }
+ }
+
+ // now that the entity data is loaded, we can load it into the world
+
+ ret = current.loadInEntityChunk(transientChunk);
+
+ final long chunkKey = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+ this.addAndRemoveTickets(chunkKey,
+ TicketType.UNKNOWN, MAX_TICKET_LEVEL, new ChunkPos(chunkX, chunkZ),
+ TicketType.ENTITY_LOAD, MAX_TICKET_LEVEL, entityLoadId
+ );
+
+ return ret;
+ }
+
+ public PoiChunk getPoiChunkIfLoaded(final int chunkX, final int chunkZ, final boolean checkLoadInCallback) {
+ final NewChunkHolder holder = this.getChunkHolder(chunkX, chunkZ);
+ if (holder != null) {
+ final PoiChunk ret = holder.getPoiChunk();
+ return ret == null || (checkLoadInCallback && !ret.isLoaded()) ? null : ret;
+ }
+ return null;
+ }
+
+ private long poiLoadCounter;
+
+ public PoiChunk loadPoiChunk(final int chunkX, final int chunkZ) {
+ TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Cannot create poi chunk off-main");
+ PoiChunk ret;
+
+ NewChunkHolder current = this.getChunkHolder(chunkX, chunkZ);
+ if (current != null && (ret = current.getPoiChunk()) != null) {
+ if (!ret.isLoaded()) {
+ ret.load();
+ }
+ return ret;
+ }
+
+ final AtomicReference<PoiChunk> completed = new AtomicReference<>();
+ final AtomicBoolean isCompleted = new AtomicBoolean();
+ final Thread waiter = Thread.currentThread();
+ final Long poiLoadId;
+ NewChunkHolder.GenericDataLoadTaskCallback loadTask = null;
+ this.ticketLock.lock();
+ try {
+ poiLoadId = Long.valueOf(this.poiLoadCounter++);
+ this.addTicketAtLevel(TicketType.POI_LOAD, chunkX, chunkZ, MAX_TICKET_LEVEL, poiLoadId);
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ current = this.getOrCreateChunkHolder(chunkX, chunkZ);
+ if (current.isPoiChunkLoaded()) {
+ this.removeTicketAtLevel(TicketType.POI_LOAD, chunkX, chunkZ, MAX_TICKET_LEVEL, poiLoadId);
+ return current.getPoiChunk();
+ }
+
+ loadTask = current.getOrLoadPoiData((final GenericDataLoadTask.TaskResult<PoiChunk, Throwable> result) -> {
+ completed.setPlain(result.left());
+ isCompleted.set(true);
+ LockSupport.unpark(waiter);
+ });
+ final ChunkLoadTask.PoiDataLoadTask poiLoad = current.getPoiDataLoadTask();
+
+ if (poiLoad != null) {
+ poiLoad.raisePriority(PrioritisedExecutor.Priority.BLOCKING);
+ }
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+
+ if (loadTask != null) {
+ loadTask.schedule();
+ }
+
+ // Note: no need to busy wait on the chunk queue, poi load will complete off-main
+
+ boolean interrupted = false;
+ while (!isCompleted.get()) {
+ interrupted |= Thread.interrupted();
+ LockSupport.park();
+ }
+
+ if (interrupted) {
+ Thread.currentThread().interrupt();
+ }
+
+ ret = completed.getPlain();
+
+ ret.load();
+
+ final long chunkKey = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+ this.addAndRemoveTickets(chunkKey,
+ TicketType.UNKNOWN, MAX_TICKET_LEVEL, new ChunkPos(chunkX, chunkZ),
+ TicketType.POI_LOAD, MAX_TICKET_LEVEL, poiLoadId
+ );
+
+ return ret;
+ }
+
+ void addChangedStatuses(final List<NewChunkHolder> changedFullStatus) {
+ if (changedFullStatus.isEmpty()) {
+ return;
+ }
+ if (!TickThread.isTickThread()) {
+ this.taskScheduler.scheduleChunkTask(() -> {
+ final ArrayDeque<NewChunkHolder> pendingFullLoadUpdate = ChunkHolderManager.this.pendingFullLoadUpdate;
+ for (int i = 0, len = changedFullStatus.size(); i < len; ++i) {
+ pendingFullLoadUpdate.add(changedFullStatus.get(i));
+ }
+
+ ChunkHolderManager.this.processPendingFullUpdate();
+ }, PrioritisedExecutor.Priority.HIGHEST);
+ } else {
+ final ArrayDeque<NewChunkHolder> pendingFullLoadUpdate = this.pendingFullLoadUpdate;
+ for (int i = 0, len = changedFullStatus.size(); i < len; ++i) {
+ pendingFullLoadUpdate.add(changedFullStatus.get(i));
+ }
+ }
+ }
+
+ final ReferenceLinkedOpenHashSet<NewChunkHolder> unloadQueue = new ReferenceLinkedOpenHashSet<>();
+
+ private void removeChunkHolder(final NewChunkHolder holder) {
+ holder.killed = true;
+ holder.vanillaChunkHolder.onChunkRemove();
+ this.autoSaveQueue.remove(holder);
+ ChunkSystem.onChunkHolderDelete(this.world, holder.vanillaChunkHolder);
+ this.chunkHolders.remove(CoordinateUtils.getChunkKey(holder.chunkX, holder.chunkZ));
+ }
+
+ // note: never call while inside the chunk system, this will absolutely break everything
+ public void processUnloads() {
+ TickThread.ensureTickThread("Cannot unload chunks off-main");
+
+ if (BLOCK_TICKET_UPDATES.get() == Boolean.TRUE) {
+ throw new IllegalStateException("Cannot unload chunks recursively");
+ }
+ if (this.ticketLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot hold ticket update lock while calling processUnloads");
+ }
+ if (this.taskScheduler.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot hold scheduling lock while calling processUnloads");
+ }
+
+ final List<NewChunkHolder.UnloadState> unloadQueue;
+ final List<ChunkProgressionTask> scheduleList = new ArrayList<>();
+ this.ticketLock.lock();
+ try {
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ if (this.unloadQueue.isEmpty()) {
+ return;
+ }
+ // in order to ensure all chunks in the unload queue do not have a pending ticket level update,
+ // process them now
+ this.processTicketUpdates(false, false, scheduleList);
+ unloadQueue = new ArrayList<>((int)(this.unloadQueue.size() * 0.05) + 1);
+
+ final int unloadCount = Math.max(50, (int)(this.unloadQueue.size() * 0.05));
+ for (int i = 0; i < unloadCount && !this.unloadQueue.isEmpty(); ++i) {
+ final NewChunkHolder chunkHolder = this.unloadQueue.removeFirst();
+ if (chunkHolder.isSafeToUnload() != null) {
+ LOGGER.error("Chunkholder " + chunkHolder + " is not safe to unload but is inside the unload queue?");
+ continue;
+ }
+ final NewChunkHolder.UnloadState state = chunkHolder.unloadStage1();
+ if (state == null) {
+ // can unload immediately
+ this.removeChunkHolder(chunkHolder);
+ continue;
+ }
+ unloadQueue.add(state);
+ }
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+ // schedule tasks, we can't let processTicketUpdates do this because we call it holding the schedule lock
+ for (int i = 0, len = scheduleList.size(); i < len; ++i) {
+ scheduleList.get(i).schedule();
+ }
+
+ final List<NewChunkHolder> toRemove = new ArrayList<>(unloadQueue.size());
+
+ final Boolean before = this.blockTicketUpdates();
+ try {
+ for (int i = 0, len = unloadQueue.size(); i < len; ++i) {
+ final NewChunkHolder.UnloadState state = unloadQueue.get(i);
+ final NewChunkHolder holder = state.holder();
+
+ holder.unloadStage2(state);
+ toRemove.add(holder);
+ }
+ } finally {
+ this.unblockTicketUpdates(before);
+ }
+
+ this.ticketLock.lock();
+ try {
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ for (int i = 0, len = toRemove.size(); i < len; ++i) {
+ final NewChunkHolder holder = toRemove.get(i);
+
+ if (holder.unloadStage3()) {
+ this.removeChunkHolder(holder);
+ } else {
+ // add cooldown so the next unload check is not immediately next tick
+ this.addTicketAtLevel(TicketType.UNLOAD_COOLDOWN, holder.chunkX, holder.chunkZ, MAX_TICKET_LEVEL, Unit.INSTANCE);
+ }
+ }
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+
+ private final ThreadLocal<Boolean> BLOCK_TICKET_UPDATES = ThreadLocal.withInitial(() -> {
+ return Boolean.FALSE;
+ });
+
+ public Boolean blockTicketUpdates() {
+ final Boolean ret = BLOCK_TICKET_UPDATES.get();
+ BLOCK_TICKET_UPDATES.set(Boolean.TRUE);
+ return ret;
+ }
+
+ public void unblockTicketUpdates(final Boolean before) {
+ BLOCK_TICKET_UPDATES.set(before);
+ }
+
+ public boolean processTicketUpdates() {
+ return this.processTicketUpdates(true, true, null);
+ }
+
+ private static final ThreadLocal<List<ChunkProgressionTask>> CURRENT_TICKET_UPDATE_SCHEDULING = new ThreadLocal<>();
+
+ static List<ChunkProgressionTask> getCurrentTicketUpdateScheduling() {
+ return CURRENT_TICKET_UPDATE_SCHEDULING.get();
+ }
+
+ private boolean processTicketUpdates(final boolean checkLocks, final boolean processFullUpdates, List<ChunkProgressionTask> scheduledTasks) {
+ TickThread.ensureTickThread("Cannot process ticket levels off-main");
+ if (BLOCK_TICKET_UPDATES.get() == Boolean.TRUE) {
+ throw new IllegalStateException("Cannot update ticket level while unloading chunks or updating entity manager");
+ }
+ if (checkLocks && this.ticketLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Illegal recursive processTicketUpdates!");
+ }
+ if (checkLocks && this.taskScheduler.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot update ticket levels from a scheduler context!");
+ }
+
+ List<NewChunkHolder> changedFullStatus = null;
+
+ final boolean isTickThread = TickThread.isTickThread();
+
+ boolean ret = false;
+ final boolean canProcessFullUpdates = processFullUpdates & isTickThread;
+ final boolean canProcessScheduling = scheduledTasks == null;
+
+ this.ticketLock.lock();
+ try {
+ final boolean levelsUpdated = this.ticketLevelPropagator.propagateUpdates();
+ if (levelsUpdated) {
+ // Unlike CB, ticket level updates cannot happen recursively. Thank god.
+ if (!this.ticketLevelUpdates.isEmpty()) {
+ ret = true;
+
+ // first the necessary chunkholders must be created, so just update the ticket levels
+ for (final Iterator<Long2IntMap.Entry> iterator = this.ticketLevelUpdates.long2IntEntrySet().fastIterator(); iterator.hasNext();) {
+ final Long2IntMap.Entry entry = iterator.next();
+ final long key = entry.getLongKey();
+ final int newLevel = entry.getIntValue();
+
+ NewChunkHolder current = this.chunkHolders.get(key);
+ if (current == null && newLevel > MAX_TICKET_LEVEL) {
+ // not loaded and it shouldn't be loaded!
+ iterator.remove();
+ continue;
+ }
+
+ final int currentLevel = current == null ? MAX_TICKET_LEVEL + 1 : current.getCurrentTicketLevel();
+ if (currentLevel == newLevel) {
+ // nothing to do
+ iterator.remove();
+ continue;
+ }
+
+ if (current == null) {
+ // must create
+ current = this.createChunkHolder(key);
+ this.chunkHolders.put(key, current);
+ current.updateTicketLevel(newLevel);
+ } else {
+ current.updateTicketLevel(newLevel);
+ }
+ }
+
+ if (scheduledTasks == null) {
+ scheduledTasks = new ArrayList<>();
+ }
+ changedFullStatus = new ArrayList<>();
+
+ // allow the chunkholders to process ticket level updates without needing to acquire the schedule lock every time
+ final List<ChunkProgressionTask> prev = CURRENT_TICKET_UPDATE_SCHEDULING.get();
+ CURRENT_TICKET_UPDATE_SCHEDULING.set(scheduledTasks);
+ try {
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ for (final Iterator<Long2IntMap.Entry> iterator = this.ticketLevelUpdates.long2IntEntrySet().fastIterator(); iterator.hasNext();) {
+ final Long2IntMap.Entry entry = iterator.next();
+ final long key = entry.getLongKey();
+ final NewChunkHolder current = this.chunkHolders.get(key);
+
+ if (current == null) {
+ throw new IllegalStateException("Expected chunk holder to be created");
+ }
+
+ current.processTicketLevelUpdate(scheduledTasks, changedFullStatus);
+ }
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ CURRENT_TICKET_UPDATE_SCHEDULING.set(prev);
+ }
+
+ this.ticketLevelUpdates.clear();
+ }
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+
+ if (changedFullStatus != null) {
+ this.addChangedStatuses(changedFullStatus);
+ }
+
+ if (canProcessScheduling && scheduledTasks != null) {
+ for (int i = 0, len = scheduledTasks.size(); i < len; ++i) {
+ scheduledTasks.get(i).schedule();
+ }
+ }
+
+ if (canProcessFullUpdates) {
+ ret |= this.processPendingFullUpdate();
+ }
+
+ return ret;
+ }
+
+ // only call on tick thread
+ protected final boolean processPendingFullUpdate() {
+ final ArrayDeque<NewChunkHolder> pendingFullLoadUpdate = this.pendingFullLoadUpdate;
+
+ boolean ret = false;
+
+ List<NewChunkHolder> changedFullStatus = new ArrayList<>();
+
+ NewChunkHolder holder;
+ while ((holder = pendingFullLoadUpdate.poll()) != null) {
+ ret |= holder.handleFullStatusChange(changedFullStatus);
+
+ if (!changedFullStatus.isEmpty()) {
+ for (int i = 0, len = changedFullStatus.size(); i < len; ++i) {
+ pendingFullLoadUpdate.add(changedFullStatus.get(i));
+ }
+ changedFullStatus.clear();
+ }
+ }
+
+ return ret;
+ }
+
+ public JsonObject getDebugJsonForWatchdog() {
+ // try and detect any potential deadlock that would require us to read unlocked
+ try {
+ if (this.ticketLock.tryLock(10, TimeUnit.SECONDS)) {
+ try {
+ if (this.taskScheduler.schedulingLock.tryLock(10, TimeUnit.SECONDS)) {
+ try {
+ return this.getDebugJsonNoLock();
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+ }
+ } catch (final InterruptedException ignore) {}
+
+ LOGGER.error("Failed to acquire ticket and scheduling lock before timeout for world " + this.world.getWorld().getName());
+
+ // because we read without locks, it may throw exceptions for fastutil maps
+ // so just try until it works...
+ Throwable lastException = null;
+ for (int count = 0;count < 1000;++count) {
+ try {
+ return this.getDebugJsonNoLock();
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr) {
+ lastException = thr;
+ Thread.yield();
+ LockSupport.parkNanos(10_000L);
+ }
+ }
+
+ // failed, return
+ LOGGER.error("Failed to retrieve debug json for watchdog thread without locking", lastException);
+ return null;
+ }
+
+ private JsonObject getDebugJsonNoLock() {
+ final JsonObject ret = new JsonObject();
+ ret.addProperty("current_tick", Long.valueOf(this.currentTick));
+
+ final JsonArray unloadQueue = new JsonArray();
+ ret.add("unload_queue", unloadQueue);
+ for (final NewChunkHolder holder : this.unloadQueue) {
+ final JsonObject coordinate = new JsonObject();
+ unloadQueue.add(coordinate);
+
+ coordinate.addProperty("chunkX", Integer.valueOf(holder.chunkX));
+ coordinate.addProperty("chunkZ", Integer.valueOf(holder.chunkZ));
+ }
+
+ final JsonArray holders = new JsonArray();
+ ret.add("chunkholders", holders);
+
+ for (final NewChunkHolder holder : this.getChunkHolders()) {
+ holders.add(holder.getDebugJson());
+ }
+
+ final JsonArray removeTickToChunkExpireTicketCount = new JsonArray();
+ ret.add("remove_tick_to_chunk_expire_ticket_count", removeTickToChunkExpireTicketCount);
+
+ for (final Long2ObjectMap.Entry<Long2IntOpenHashMap> tickEntry : this.removeTickToChunkExpireTicketCount.long2ObjectEntrySet()) {
+ final long tick = tickEntry.getLongKey();
+ final Long2IntOpenHashMap coordinateToCount = tickEntry.getValue();
+
+ final JsonObject tickJson = new JsonObject();
+ removeTickToChunkExpireTicketCount.add(tickJson);
+
+ tickJson.addProperty("tick", Long.valueOf(tick));
+
+ final JsonArray tickEntries = new JsonArray();
+ tickJson.add("entries", tickEntries);
+
+ for (final Long2IntMap.Entry entry : coordinateToCount.long2IntEntrySet()) {
+ final long coordinate = entry.getLongKey();
+ final int count = entry.getIntValue();
+
+ final JsonObject entryJson = new JsonObject();
+ tickEntries.add(entryJson);
+
+ entryJson.addProperty("chunkX", Long.valueOf(CoordinateUtils.getChunkX(coordinate)));
+ entryJson.addProperty("chunkZ", Long.valueOf(CoordinateUtils.getChunkZ(coordinate)));
+ entryJson.addProperty("count", Integer.valueOf(count));
+ }
+ }
+
+ final JsonArray allTicketsJson = new JsonArray();
+ ret.add("tickets", allTicketsJson);
+
+ for (final Long2ObjectMap.Entry<SortedArraySet<Ticket<?>>> coordinateTickets : this.tickets.long2ObjectEntrySet()) {
+ final long coordinate = coordinateTickets.getLongKey();
+ final SortedArraySet<Ticket<?>> tickets = coordinateTickets.getValue();
+
+ final JsonObject coordinateJson = new JsonObject();
+ allTicketsJson.add(coordinateJson);
+
+ coordinateJson.addProperty("chunkX", Long.valueOf(CoordinateUtils.getChunkX(coordinate)));
+ coordinateJson.addProperty("chunkZ", Long.valueOf(CoordinateUtils.getChunkZ(coordinate)));
+
+ final JsonArray ticketsSerialized = new JsonArray();
+ coordinateJson.add("tickets", ticketsSerialized);
+
+ for (final Ticket<?> ticket : tickets) {
+ final JsonObject ticketSerialized = new JsonObject();
+ ticketsSerialized.add(ticketSerialized);
+
+ ticketSerialized.addProperty("type", ticket.getType().toString());
+ ticketSerialized.addProperty("level", Integer.valueOf(ticket.getTicketLevel()));
+ ticketSerialized.addProperty("identifier", Objects.toString(ticket.key));
+ ticketSerialized.addProperty("remove_tick", Long.valueOf(ticket.removalTick));
+ }
+ }
+
+ return ret;
+ }
+
+ public JsonObject getDebugJson() {
+ final List<ChunkProgressionTask> scheduleList = new ArrayList<>();
+ try {
+ final JsonObject ret;
+ this.ticketLock.lock();
+ try {
+ this.taskScheduler.schedulingLock.lock();
+ try {
+ this.processTicketUpdates(false, false, scheduleList);
+ ret = this.getDebugJsonNoLock();
+ } finally {
+ this.taskScheduler.schedulingLock.unlock();
+ }
+ } finally {
+ this.ticketLock.unlock();
+ }
+ return ret;
+ } finally {
+ // schedule tasks, we can't let processTicketUpdates do this because we call it holding the schedule lock
+ for (int i = 0, len = scheduleList.size(); i < len; ++i) {
+ scheduleList.get(i).schedule();
+ }
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLightTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLightTask.java
new file mode 100644
index 0000000000000000000000000000000000000000..53ddd7e9ac05e6a9eb809f329796e6d4f6bb2ab1
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLightTask.java
@@ -0,0 +1,181 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.starlight.common.light.StarLightEngine;
+import ca.spottedleaf.starlight.common.light.StarLightInterface;
+import io.papermc.paper.chunk.system.light.LightQueue;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.ProtoChunk;
+import org.apache.logging.log4j.LogManager;
+import org.apache.logging.log4j.Logger;
+import java.util.function.BooleanSupplier;
+
+public final class ChunkLightTask extends ChunkProgressionTask {
+
+ private static final Logger LOGGER = LogManager.getLogger();
+
+ protected final ChunkAccess fromChunk;
+
+ private final LightTaskPriorityHolder priorityHolder;
+
+ public ChunkLightTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX, final int chunkZ,
+ final ChunkAccess chunk, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ);
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.priorityHolder = new LightTaskPriorityHolder(priority, this);
+ this.fromChunk = chunk;
+ }
+
+ @Override
+ public boolean isScheduled() {
+ return this.priorityHolder.isScheduled();
+ }
+
+ @Override
+ public ChunkStatus getTargetStatus() {
+ return ChunkStatus.LIGHT;
+ }
+
+ @Override
+ public void schedule() {
+ this.priorityHolder.schedule();
+ }
+
+ @Override
+ public void cancel() {
+ this.priorityHolder.cancel();
+ }
+
+ @Override
+ public PrioritisedExecutor.Priority getPriority() {
+ return this.priorityHolder.getPriority();
+ }
+
+ @Override
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ this.priorityHolder.raisePriority(priority);
+ }
+
+ @Override
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ this.priorityHolder.setPriority(priority);
+ }
+
+ @Override
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ this.priorityHolder.raisePriority(priority);
+ }
+
+ private static final class LightTaskPriorityHolder extends PriorityHolder {
+
+ protected final ChunkLightTask task;
+
+ protected LightTaskPriorityHolder(final PrioritisedExecutor.Priority priority, final ChunkLightTask task) {
+ super(priority);
+ this.task = task;
+ }
+
+ @Override
+ protected void cancelScheduled() {
+ final ChunkLightTask task = this.task;
+ task.complete(null, null);
+ }
+
+ @Override
+ protected PrioritisedExecutor.Priority getScheduledPriority() {
+ final ChunkLightTask task = this.task;
+ return task.world.getChunkSource().getLightEngine().theLightEngine.lightQueue.getPriority(task.chunkX, task.chunkZ);
+ }
+
+ @Override
+ protected void scheduleTask(final PrioritisedExecutor.Priority priority) {
+ final ChunkLightTask task = this.task;
+ final StarLightInterface starLightInterface = task.world.getChunkSource().getLightEngine().theLightEngine;
+ final LightQueue lightQueue = starLightInterface.lightQueue;
+ lightQueue.queueChunkLightTask(new ChunkPos(task.chunkX, task.chunkZ), new LightTask(starLightInterface, task), priority);
+ lightQueue.setPriority(task.chunkX, task.chunkZ, priority);
+ }
+
+ @Override
+ protected void lowerPriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final ChunkLightTask task = this.task;
+ final StarLightInterface starLightInterface = task.world.getChunkSource().getLightEngine().theLightEngine;
+ final LightQueue lightQueue = starLightInterface.lightQueue;
+ lightQueue.lowerPriority(task.chunkX, task.chunkZ, priority);
+ }
+
+ @Override
+ protected void setPriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final ChunkLightTask task = this.task;
+ final StarLightInterface starLightInterface = task.world.getChunkSource().getLightEngine().theLightEngine;
+ final LightQueue lightQueue = starLightInterface.lightQueue;
+ lightQueue.setPriority(task.chunkX, task.chunkZ, priority);
+ }
+
+ @Override
+ protected void raisePriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final ChunkLightTask task = this.task;
+ final StarLightInterface starLightInterface = task.world.getChunkSource().getLightEngine().theLightEngine;
+ final LightQueue lightQueue = starLightInterface.lightQueue;
+ lightQueue.raisePriority(task.chunkX, task.chunkZ, priority);
+ }
+ }
+
+ private static final class LightTask implements BooleanSupplier {
+
+ protected final StarLightInterface lightEngine;
+ protected final ChunkLightTask task;
+
+ public LightTask(final StarLightInterface lightEngine, final ChunkLightTask task) {
+ this.lightEngine = lightEngine;
+ this.task = task;
+ }
+
+ @Override
+ public boolean getAsBoolean() {
+ final ChunkLightTask task = this.task;
+ // executed on light thread
+ if (!task.priorityHolder.markExecuting()) {
+ // cancelled
+ return false;
+ }
+
+ try {
+ final Boolean[] emptySections = StarLightEngine.getEmptySectionsForChunk(task.fromChunk);
+
+ if (task.fromChunk.isLightCorrect() && task.fromChunk.getStatus().isOrAfter(ChunkStatus.LIGHT)) {
+ this.lightEngine.forceLoadInChunk(task.fromChunk, emptySections);
+ this.lightEngine.checkChunkEdges(task.chunkX, task.chunkZ);
+ } else {
+ task.fromChunk.setLightCorrect(false);
+ this.lightEngine.lightChunk(task.fromChunk, emptySections);
+ task.fromChunk.setLightCorrect(true);
+ }
+ // we need to advance status
+ if (task.fromChunk instanceof ProtoChunk chunk && chunk.getStatus() == ChunkStatus.LIGHT.getParent()) {
+ chunk.setStatus(ChunkStatus.LIGHT);
+ }
+ } catch (final Throwable thr) {
+ if (!(thr instanceof ThreadDeath)) {
+ LOGGER.fatal("Failed to light chunk " + task.fromChunk.getPos().toString() + " in world '" + this.lightEngine.getWorld().getWorld().getName() + "'", thr);
+ }
+
+ task.complete(null, thr);
+
+ if (thr instanceof ThreadDeath) {
+ throw (ThreadDeath)thr;
+ }
+
+ return true;
+ }
+
+ task.complete(task.fromChunk, null);
+ return true;
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLoadTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLoadTask.java
new file mode 100644
2023-03-15 20:17:42 +01:00
index 0000000000000000000000000000000000000000..be6f3f6a57668a9bd50d0ea5f2dd2335355b69d6
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkLoadTask.java
@@ -0,0 +1,499 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.collection.MultiThreadedQueue;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import ca.spottedleaf.dataconverter.minecraft.MCDataConverter;
+import ca.spottedleaf.dataconverter.minecraft.datatypes.MCTypeRegistry;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.chunk.system.io.RegionFileIOThread;
+import io.papermc.paper.chunk.system.poi.PoiChunk;
+import net.minecraft.SharedConstants;
+import net.minecraft.core.registries.Registries;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.ProtoChunk;
+import net.minecraft.world.level.chunk.UpgradeData;
+import net.minecraft.world.level.chunk.storage.ChunkSerializer;
+import net.minecraft.world.level.chunk.storage.EntityStorage;
+import net.minecraft.world.level.levelgen.blending.BlendingData;
+import org.slf4j.Logger;
+import java.lang.invoke.VarHandle;
+import java.util.Map;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.Consumer;
+
+public final class ChunkLoadTask extends ChunkProgressionTask {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ private final NewChunkHolder chunkHolder;
+ private final ChunkDataLoadTask loadTask;
+
+ private boolean cancelled;
+ private NewChunkHolder.GenericDataLoadTaskCallback entityLoadTask;
+ private NewChunkHolder.GenericDataLoadTaskCallback poiLoadTask;
+
+ protected ChunkLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX, final int chunkZ,
+ final NewChunkHolder chunkHolder, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ);
+ this.chunkHolder = chunkHolder;
+ this.loadTask = new ChunkDataLoadTask(scheduler, world, chunkX, chunkZ, priority);
+ this.loadTask.addCallback((final GenericDataLoadTask.TaskResult<ChunkAccess, Throwable> result) -> {
+ ChunkLoadTask.this.complete(result == null ? null : result.left(), result == null ? null : result.right());
+ });
+ }
+
+ @Override
+ public ChunkStatus getTargetStatus() {
+ return ChunkStatus.EMPTY;
+ }
+
+ private boolean scheduled;
+
+ @Override
+ public boolean isScheduled() {
+ return this.scheduled;
+ }
+
+ @Override
+ public void schedule() {
+ final NewChunkHolder.GenericDataLoadTaskCallback entityLoadTask;
+ final NewChunkHolder.GenericDataLoadTaskCallback poiLoadTask;
+
+ final AtomicInteger count = new AtomicInteger();
+ final Consumer<GenericDataLoadTask.TaskResult<?, ?>> scheduleLoadTask = (final GenericDataLoadTask.TaskResult<?, ?> result) -> {
+ if (count.decrementAndGet() == 0) {
+ ChunkLoadTask.this.loadTask.schedule(false);
+ }
+ };
+
+ // NOTE: it is IMPOSSIBLE for getOrLoadEntityData/getOrLoadPoiData to complete synchronously, because
+ // they must schedule a task to off main or to on main to complete
+ this.scheduler.schedulingLock.lock();
+ try {
+ if (this.scheduled) {
+ throw new IllegalStateException("schedule() called twice");
+ }
+ this.scheduled = true;
+ if (this.cancelled) {
+ return;
+ }
+ if (!this.chunkHolder.isEntityChunkNBTLoaded()) {
+ entityLoadTask = this.chunkHolder.getOrLoadEntityData((Consumer)scheduleLoadTask);
+ count.setPlain(count.getPlain() + 1);
+ } else {
+ entityLoadTask = null;
+ }
+
+ if (!this.chunkHolder.isPoiChunkLoaded()) {
+ poiLoadTask = this.chunkHolder.getOrLoadPoiData((Consumer)scheduleLoadTask);
+ count.setPlain(count.getPlain() + 1);
+ } else {
+ poiLoadTask = null;
+ }
+
+ this.entityLoadTask = entityLoadTask;
+ this.poiLoadTask = poiLoadTask;
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+
+ if (entityLoadTask != null) {
+ entityLoadTask.schedule();
+ }
+
+ if (poiLoadTask != null) {
+ poiLoadTask.schedule();
+ }
+
+ if (entityLoadTask == null && poiLoadTask == null) {
+ // no need to wait on those, we can schedule now
+ this.loadTask.schedule(false);
+ }
+ }
+
+ @Override
+ public void cancel() {
+ // must be before load task access, so we can synchronise with the writes to the fields
+ this.scheduler.schedulingLock.lock();
+ try {
+ this.cancelled = true;
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+
+ /*
+ Note: The entityLoadTask/poiLoadTask do not complete when cancelled,
+ but this is fine because if they are successfully cancelled then
+ we will successfully cancel the load task, which will complete when cancelled
+ */
+
+ if (this.entityLoadTask != null) {
+ this.entityLoadTask.cancel();
+ }
+ if (this.poiLoadTask != null) {
+ this.poiLoadTask.cancel();
+ }
+ this.loadTask.cancel();
+ }
+
+ @Override
+ public PrioritisedExecutor.Priority getPriority() {
+ return this.loadTask.getPriority();
+ }
+
+ @Override
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ final EntityDataLoadTask entityLoad = this.chunkHolder.getEntityDataLoadTask();
+ if (entityLoad != null) {
+ entityLoad.lowerPriority(priority);
+ }
+
+ final PoiDataLoadTask poiLoad = this.chunkHolder.getPoiDataLoadTask();
+
+ if (poiLoad != null) {
+ poiLoad.lowerPriority(priority);
+ }
+
+ this.loadTask.lowerPriority(priority);
+ }
+
+ @Override
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ final EntityDataLoadTask entityLoad = this.chunkHolder.getEntityDataLoadTask();
+ if (entityLoad != null) {
+ entityLoad.setPriority(priority);
+ }
+
+ final PoiDataLoadTask poiLoad = this.chunkHolder.getPoiDataLoadTask();
+
+ if (poiLoad != null) {
+ poiLoad.setPriority(priority);
+ }
+
+ this.loadTask.setPriority(priority);
+ }
+
+ @Override
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ final EntityDataLoadTask entityLoad = this.chunkHolder.getEntityDataLoadTask();
+ if (entityLoad != null) {
+ entityLoad.raisePriority(priority);
+ }
+
+ final PoiDataLoadTask poiLoad = this.chunkHolder.getPoiDataLoadTask();
+
+ if (poiLoad != null) {
+ poiLoad.raisePriority(priority);
+ }
+
+ this.loadTask.raisePriority(priority);
+ }
+
+ protected static abstract class CallbackDataLoadTask<OnMain,FinalCompletion> extends GenericDataLoadTask<OnMain,FinalCompletion> {
+
+ private TaskResult<FinalCompletion, Throwable> result;
+ private final MultiThreadedQueue<Consumer<TaskResult<FinalCompletion, Throwable>>> waiters = new MultiThreadedQueue<>();
+
+ protected volatile boolean completed;
+ protected static final VarHandle COMPLETED_HANDLE = ConcurrentUtil.getVarHandle(CallbackDataLoadTask.class, "completed", boolean.class);
+
+ protected CallbackDataLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final RegionFileIOThread.RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ, type, priority);
+ }
+
+ public void addCallback(final Consumer<TaskResult<FinalCompletion, Throwable>> consumer) {
+ if (!this.waiters.add(consumer)) {
+ try {
+ consumer.accept(this.result);
+ } catch (final Throwable throwable) {
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Consumer", ChunkTaskScheduler.stringIfNull(consumer),
+ "Completed throwable", ChunkTaskScheduler.stringIfNull(this.result.right())
+ ), throwable);
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ }
+ }
+ }
+
+ @Override
+ protected void onComplete(final TaskResult<FinalCompletion, Throwable> result) {
+ if ((boolean)COMPLETED_HANDLE.getAndSet((CallbackDataLoadTask)this, (boolean)true)) {
+ throw new IllegalStateException("Already completed");
+ }
+ this.result = result;
+ Consumer<TaskResult<FinalCompletion, Throwable>> consumer;
+ while ((consumer = this.waiters.pollOrBlockAdds()) != null) {
+ try {
+ consumer.accept(result);
+ } catch (final Throwable throwable) {
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Consumer", ChunkTaskScheduler.stringIfNull(consumer),
+ "Completed throwable", ChunkTaskScheduler.stringIfNull(result.right())
+ ), throwable);
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ return;
+ }
+ }
+ }
+ }
+
+ public final class ChunkDataLoadTask extends CallbackDataLoadTask<ChunkSerializer.InProgressChunkHolder, ChunkAccess> {
+ protected ChunkDataLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ, RegionFileIOThread.RegionFileType.CHUNK_DATA, priority);
+ }
+
+ @Override
+ protected boolean hasOffMain() {
+ return true;
+ }
+
+ @Override
+ protected boolean hasOnMain() {
+ return true;
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOffMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ return this.scheduler.loadExecutor.createTask(run, priority);
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOnMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ return this.scheduler.createChunkTask(this.chunkX, this.chunkZ, run, priority);
+ }
+
+ @Override
+ protected TaskResult<ChunkAccess, Throwable> completeOnMainOffMain(final ChunkSerializer.InProgressChunkHolder data, final Throwable throwable) {
+ if (data != null) {
+ return null;
+ }
+
+ final PoiChunk poiChunk = ChunkLoadTask.this.chunkHolder.getPoiChunk();
+ if (poiChunk == null) {
+ LOGGER.error("Expected poi chunk to be loaded with chunk for task " + this.toString());
+ } else if (!poiChunk.isLoaded()) {
+ // need to call poiChunk.load() on main
+ return null;
+ }
+
+ return new TaskResult<>(this.getEmptyChunk(), null);
+ }
+
+ @Override
+ protected TaskResult<ChunkSerializer.InProgressChunkHolder, Throwable> runOffMain(final CompoundTag data, final Throwable throwable) {
+ if (throwable != null) {
+ LOGGER.error("Failed to load chunk data for task: " + this.toString() + ", chunk data will be lost", throwable);
+ return new TaskResult<>(null, null);
+ }
+
+ if (data == null) {
+ return new TaskResult<>(null, null);
+ }
+
+ // need to convert data, and then deserialize it
+
+ try {
+ final ChunkPos chunkPos = new ChunkPos(this.chunkX, this.chunkZ);
+ final ChunkMap chunkMap = this.world.getChunkSource().chunkMap;
+ // run converters
+ // note: upgradeChunkTag copies the data already
+ final CompoundTag converted = chunkMap.upgradeChunkTag(
+ this.world.getTypeKey(), chunkMap.overworldDataStorage, data, chunkMap.generator.getTypeNameForDataFixer(),
+ chunkPos, this.world
+ );
+ // deserialize
+ final ChunkSerializer.InProgressChunkHolder chunkHolder = ChunkSerializer.loadChunk(
+ this.world, chunkMap.getPoiManager(), chunkPos, converted, true
+ );
+
+ return new TaskResult<>(chunkHolder, null);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr2) {
+ LOGGER.error("Failed to parse chunk data for task: " + this.toString() + ", chunk data will be lost", thr2);
+ return new TaskResult<>(null, thr2);
+ }
+ }
+
+ private ProtoChunk getEmptyChunk() {
+ return new ProtoChunk(
+ new ChunkPos(this.chunkX, this.chunkZ), UpgradeData.EMPTY, this.world,
+ this.world.registryAccess().registryOrThrow(Registries.BIOME), (BlendingData)null
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ );
+ }
+
+ @Override
+ protected TaskResult<ChunkAccess, Throwable> runOnMain(final ChunkSerializer.InProgressChunkHolder data, final Throwable throwable) {
+ final PoiChunk poiChunk = ChunkLoadTask.this.chunkHolder.getPoiChunk();
+ if (poiChunk == null) {
+ LOGGER.error("Expected poi chunk to be loaded with chunk for task " + this.toString());
+ } else {
+ poiChunk.load();
+ }
+
+ if (data == null || data.protoChunk == null) {
+ // throwable could be non-null, but the off-main task will print its exceptions - so we don't need to care,
+ // it's handled already
+
+ return new TaskResult<>(this.getEmptyChunk(), null);
+ }
+
+ // have tasks to run (at this point, it's just the POI consistency checking)
+ try {
+ if (data.tasks != null) {
+ for (int i = 0, len = data.tasks.size(); i < len; ++i) {
+ data.tasks.poll().run();
+ }
+ }
+
+ return new TaskResult<>(data.protoChunk, null);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr2) {
+ LOGGER.error("Failed to parse main tasks for task " + this.toString() + ", chunk data will be lost", thr2);
+ return new TaskResult<>(this.getEmptyChunk(), null);
+ }
+ }
+ }
+
+ public static final class PoiDataLoadTask extends CallbackDataLoadTask<PoiChunk, PoiChunk> {
+ public PoiDataLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ, RegionFileIOThread.RegionFileType.POI_DATA, priority);
+ }
+
+ @Override
+ protected boolean hasOffMain() {
+ return true;
+ }
+
+ @Override
+ protected boolean hasOnMain() {
+ return false;
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOffMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ return this.scheduler.loadExecutor.createTask(run, priority);
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOnMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ throw new UnsupportedOperationException();
+ }
+
+ @Override
+ protected TaskResult<PoiChunk, Throwable> completeOnMainOffMain(final PoiChunk data, final Throwable throwable) {
+ throw new UnsupportedOperationException();
+ }
+
+ @Override
+ protected TaskResult<PoiChunk, Throwable> runOffMain(CompoundTag data, final Throwable throwable) {
+ if (throwable != null) {
+ LOGGER.error("Failed to load poi data for task: " + this.toString() + ", poi data will be lost", throwable);
+ return new TaskResult<>(PoiChunk.empty(this.world, this.chunkX, this.chunkZ), null);
+ }
+
+ if (data == null || data.isEmpty()) {
+ // nothing to do
+ return new TaskResult<>(PoiChunk.empty(this.world, this.chunkX, this.chunkZ), null);
+ }
+
+ try {
+ data = data.copy(); // coming from the I/O thread, so we need to copy
+ // run converters
+ final int dataVersion = !data.contains(SharedConstants.DATA_VERSION_TAG, 99) ? 1945 : data.getInt(SharedConstants.DATA_VERSION_TAG);
+ final CompoundTag converted = MCDataConverter.convertTag(
2023-03-15 20:17:42 +01:00
+ MCTypeRegistry.POI_CHUNK, data, dataVersion, SharedConstants.getCurrentVersion().getDataVersion().getVersion()
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ );
+
+ // now we need to parse it
+ return new TaskResult<>(PoiChunk.parse(this.world, this.chunkX, this.chunkZ, converted), null);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr2) {
+ LOGGER.error("Failed to run parse poi data for task: " + this.toString() + ", poi data will be lost", thr2);
+ return new TaskResult<>(PoiChunk.empty(this.world, this.chunkX, this.chunkZ), null);
+ }
+ }
+
+ @Override
+ protected TaskResult<PoiChunk, Throwable> runOnMain(final PoiChunk data, final Throwable throwable) {
+ throw new UnsupportedOperationException();
+ }
+ }
+
+ public static final class EntityDataLoadTask extends CallbackDataLoadTask<CompoundTag, CompoundTag> {
+
+ public EntityDataLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ, RegionFileIOThread.RegionFileType.ENTITY_DATA, priority);
+ }
+
+ @Override
+ protected boolean hasOffMain() {
+ return true;
+ }
+
+ @Override
+ protected boolean hasOnMain() {
+ return false;
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOffMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ return this.scheduler.loadExecutor.createTask(run, priority);
+ }
+
+ @Override
+ protected PrioritisedExecutor.PrioritisedTask createOnMain(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ throw new UnsupportedOperationException();
+ }
+
+ @Override
+ protected TaskResult<CompoundTag, Throwable> completeOnMainOffMain(final CompoundTag data, final Throwable throwable) {
+ throw new UnsupportedOperationException();
+ }
+
+ @Override
+ protected TaskResult<CompoundTag, Throwable> runOffMain(final CompoundTag data, final Throwable throwable) {
+ if (throwable != null) {
+ LOGGER.error("Failed to load entity data for task: " + this.toString() + ", entity data will be lost", throwable);
+ return new TaskResult<>(null, null);
+ }
+
+ if (data == null || data.isEmpty()) {
+ // nothing to do
+ return new TaskResult<>(null, null);
+ }
+
+ try {
+ // note: data comes from the I/O thread, so we need to copy it
+ return new TaskResult<>(EntityStorage.upgradeChunkTag(data.copy()), null);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr2) {
+ LOGGER.error("Failed to run converters for entity data for task: " + this.toString() + ", entity data will be lost", thr2);
+ return new TaskResult<>(null, thr2);
+ }
+ }
+
+ @Override
+ protected TaskResult<CompoundTag, Throwable> runOnMain(final CompoundTag data, final Throwable throwable) {
+ throw new UnsupportedOperationException();
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkProgressionTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkProgressionTask.java
new file mode 100644
index 0000000000000000000000000000000000000000..322675a470eacbf0e5452f4009c643f2d0b4ce24
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkProgressionTask.java
@@ -0,0 +1,105 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.collection.MultiThreadedQueue;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import java.lang.invoke.VarHandle;
+import java.util.Map;
+import java.util.function.BiConsumer;
+
+public abstract class ChunkProgressionTask {
+
+ private final MultiThreadedQueue<BiConsumer<ChunkAccess, Throwable>> waiters = new MultiThreadedQueue<>();
+ private ChunkAccess completedChunk;
+ private Throwable completedThrowable;
+
+ protected final ChunkTaskScheduler scheduler;
+ protected final ServerLevel world;
+ protected final int chunkX;
+ protected final int chunkZ;
+
+ protected volatile boolean completed;
+ protected static final VarHandle COMPLETED_HANDLE = ConcurrentUtil.getVarHandle(ChunkProgressionTask.class, "completed", boolean.class);
+
+ protected ChunkProgressionTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX, final int chunkZ) {
+ this.scheduler = scheduler;
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ }
+
+ // Used only for debug json
+ public abstract boolean isScheduled();
+
+ // Note: It is the responsibility of the task to set the chunk's status once it has completed
+ public abstract ChunkStatus getTargetStatus();
+
+ /* Only executed once */
+ /* Implementations must be prepared to handle cases where cancel() is called before schedule() */
+ public abstract void schedule();
+
+ /* May be called multiple times */
+ public abstract void cancel();
+
+ public abstract PrioritisedExecutor.Priority getPriority();
+
+ /* Schedule lock is always held for the priority update calls */
+
+ public abstract void lowerPriority(final PrioritisedExecutor.Priority priority);
+
+ public abstract void setPriority(final PrioritisedExecutor.Priority priority);
+
+ public abstract void raisePriority(final PrioritisedExecutor.Priority priority);
+
+ public final void onComplete(final BiConsumer<ChunkAccess, Throwable> onComplete) {
+ if (!this.waiters.add(onComplete)) {
+ try {
+ onComplete.accept(this.completedChunk, this.completedThrowable);
+ } catch (final Throwable throwable) {
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Consumer", ChunkTaskScheduler.stringIfNull(onComplete),
+ "Completed throwable", ChunkTaskScheduler.stringIfNull(this.completedThrowable)
+ ), throwable);
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ }
+ }
+ }
+
+ protected final void complete(final ChunkAccess chunk, final Throwable throwable) {
+ try {
+ this.complete0(chunk, throwable);
+ } catch (final Throwable thr2) {
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Completed throwable", ChunkTaskScheduler.stringIfNull(throwable)
+ ), thr2);
+ if (thr2 instanceof ThreadDeath) {
+ throw (ThreadDeath)thr2;
+ }
+ }
+ }
+
+ private void complete0(final ChunkAccess chunk, final Throwable throwable) {
+ if ((boolean)COMPLETED_HANDLE.getAndSet((ChunkProgressionTask)this, (boolean)true)) {
+ throw new IllegalStateException("Already completed");
+ }
+ this.completedChunk = chunk;
+ this.completedThrowable = throwable;
+
+ BiConsumer<ChunkAccess, Throwable> consumer;
+ while ((consumer = this.waiters.pollOrBlockAdds()) != null) {
+ consumer.accept(chunk, throwable);
+ }
+ }
+
+ @Override
+ public String toString() {
+ return "ChunkProgressionTask{class: " + this.getClass().getName() + ", for world: " + this.world.getWorld().getName() +
+ ", chunk: (" + this.chunkX + "," + this.chunkZ + "), hashcode: " + System.identityHashCode(this) + ", priority: " + this.getPriority() +
+ ", status: " + this.getTargetStatus().toString() + ", scheduled: " + this.isScheduled() + "}";
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkTaskScheduler.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkTaskScheduler.java
new file mode 100644
index 0000000000000000000000000000000000000000..84cc9397237fa0c17aa1012dfb5683c90eb6d3b8
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkTaskScheduler.java
@@ -0,0 +1,780 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedThreadPool;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedThreadedTaskQueue;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.configuration.GlobalConfiguration;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import net.minecraft.CrashReport;
+import net.minecraft.CrashReportCategory;
+import net.minecraft.ReportedException;
+import io.papermc.paper.util.MCUtil;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.server.MinecraftServer;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.server.level.TicketType;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.LevelChunk;
+import org.bukkit.Bukkit;
+import org.slf4j.Logger;
+import java.io.File;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Collections;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicLong;
+import java.util.concurrent.locks.ReentrantLock;
+import java.util.function.BooleanSupplier;
+import java.util.function.Consumer;
+
+public final class ChunkTaskScheduler {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ static int newChunkSystemIOThreads;
+ static int newChunkSystemWorkerThreads;
+ static int newChunkSystemGenParallelism;
+ static int newChunkSystemLoadParallelism;
+
+ public static ca.spottedleaf.concurrentutil.executor.standard.PrioritisedThreadPool workerThreads;
+
+ private static boolean initialised = false;
+
+ public static void init(final GlobalConfiguration.ChunkSystem config) {
+ if (initialised) {
+ return;
+ }
+ initialised = true;
+ newChunkSystemIOThreads = config.ioThreads;
+ newChunkSystemWorkerThreads = config.workerThreads;
+ if (newChunkSystemIOThreads < 0) {
+ newChunkSystemIOThreads = 1;
+ } else {
+ newChunkSystemIOThreads = Math.max(1, newChunkSystemIOThreads);
+ }
+ int defaultWorkerThreads = Runtime.getRuntime().availableProcessors() / 2;
+ if (defaultWorkerThreads <= 4) {
+ defaultWorkerThreads = defaultWorkerThreads <= 3 ? 1 : 2;
+ } else {
+ defaultWorkerThreads = defaultWorkerThreads / 2;
+ }
+ defaultWorkerThreads = Integer.getInteger("Paper.WorkerThreadCount", Integer.valueOf(defaultWorkerThreads));
+
+ if (newChunkSystemWorkerThreads < 0) {
+ newChunkSystemWorkerThreads = defaultWorkerThreads;
+ } else {
+ newChunkSystemWorkerThreads = Math.max(1, newChunkSystemWorkerThreads);
+ }
+
+ String newChunkSystemGenParallelism = config.genParallelism;
+ if (newChunkSystemGenParallelism.equalsIgnoreCase("default")) {
+ newChunkSystemGenParallelism = "true";
+ }
+ boolean useParallelGen;
+ if (newChunkSystemGenParallelism.equalsIgnoreCase("on") || newChunkSystemGenParallelism.equalsIgnoreCase("enabled")
+ || newChunkSystemGenParallelism.equalsIgnoreCase("true")) {
+ useParallelGen = true;
+ } else if (newChunkSystemGenParallelism.equalsIgnoreCase("off") || newChunkSystemGenParallelism.equalsIgnoreCase("disabled")
+ || newChunkSystemGenParallelism.equalsIgnoreCase("false")) {
+ useParallelGen = false;
+ } else {
+ throw new IllegalStateException("Invalid option for gen-parallelism: must be one of [on, off, enabled, disabled, true, false, default]");
+ }
+
+ ChunkTaskScheduler.newChunkSystemGenParallelism = useParallelGen ? newChunkSystemWorkerThreads : 1;
+ ChunkTaskScheduler.newChunkSystemLoadParallelism = newChunkSystemWorkerThreads;
+
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.init(newChunkSystemIOThreads);
+ workerThreads = new ca.spottedleaf.concurrentutil.executor.standard.PrioritisedThreadPool(
+ "Paper Chunk System Worker Pool", newChunkSystemWorkerThreads,
+ (final Thread thread, final Integer id) -> {
+ thread.setPriority(Thread.NORM_PRIORITY - 2);
+ thread.setName("Tuinity Chunk System Worker #" + id.intValue());
+ thread.setUncaughtExceptionHandler(io.papermc.paper.chunk.system.scheduling.NewChunkHolder.CHUNKSYSTEM_UNCAUGHT_EXCEPTION_HANDLER);
+ }, (long)(20.0e6)); // 20ms
+
+ LOGGER.info("Chunk system is using " + newChunkSystemIOThreads + " I/O threads, " + newChunkSystemWorkerThreads + " worker threads, and gen parallelism of " + ChunkTaskScheduler.newChunkSystemGenParallelism + " threads");
+ }
+
+ public final ServerLevel world;
+ public final PrioritisedThreadPool workers;
+ public final PrioritisedThreadPool.PrioritisedPoolExecutor lightExecutor;
+ public final PrioritisedThreadPool.PrioritisedPoolExecutor genExecutor;
+ public final PrioritisedThreadPool.PrioritisedPoolExecutor parallelGenExecutor;
+ public final PrioritisedThreadPool.PrioritisedPoolExecutor loadExecutor;
+
+ private final PrioritisedThreadedTaskQueue mainThreadExecutor = new PrioritisedThreadedTaskQueue();
+
+ final ReentrantLock schedulingLock = new ReentrantLock();
+ public final ChunkHolderManager chunkHolderManager;
+
+ static {
+ ChunkStatus.EMPTY.writeRadius = 0;
+ ChunkStatus.STRUCTURE_STARTS.writeRadius = 0;
+ ChunkStatus.STRUCTURE_REFERENCES.writeRadius = 0;
+ ChunkStatus.BIOMES.writeRadius = 0;
+ ChunkStatus.NOISE.writeRadius = 0;
+ ChunkStatus.SURFACE.writeRadius = 0;
+ ChunkStatus.CARVERS.writeRadius = 0;
+ ChunkStatus.LIQUID_CARVERS.writeRadius = 0;
+ ChunkStatus.FEATURES.writeRadius = 1;
+ ChunkStatus.LIGHT.writeRadius = 1;
+ ChunkStatus.SPAWN.writeRadius = 0;
+ ChunkStatus.HEIGHTMAPS.writeRadius = 0;
+ ChunkStatus.FULL.writeRadius = 0;
+
+ /*
+ It's important that the neighbour read radius is taken into account. If _any_ later status is using some chunk as
+ a neighbour, it must be also safe if that neighbour is being generated. i.e for any status later than FEATURES,
+ for a status to be parallel safe it must not read the block data from its neighbours.
+ */
+ final List<ChunkStatus> parallelCapableStatus = Arrays.asList(
+ // No-op executor.
+ ChunkStatus.EMPTY,
+
+ // This is parallel capable, as CB has fixed the concurrency issue with stronghold generations.
+ // Does not touch neighbour chunks.
+ // TODO On another note, what the fuck is StructureFeatureManager.StructureCheck and why is it used? it's leaking
+ ChunkStatus.STRUCTURE_STARTS,
+
+ // Surprisingly this is parallel capable. It is simply reading the already-created structure starts
+ // into the structure references for the chunk. So while it reads from it neighbours, its neighbours
+ // will not change, even if executed in parallel.
+ ChunkStatus.STRUCTURE_REFERENCES,
+
+ // Safe. Mojang runs it in parallel as well.
+ ChunkStatus.BIOMES,
+
+ // Safe. Mojang runs it in parallel as well.
+ ChunkStatus.NOISE,
+
+ // Parallel safe. Only touches the target chunk. Biome retrieval is now noise based, which is
+ // completely thread-safe.
+ ChunkStatus.SURFACE,
+
+ // No global state is modified in the carvers. It only touches the specified chunk. So it is parallel safe.
+ ChunkStatus.CARVERS,
+
+ // No-op executor. Was replaced in 1.18 with carvers, I think.
+ ChunkStatus.LIQUID_CARVERS,
+
+ // FEATURES is not parallel safe. It writes to neighbours.
+
+ // LIGHT is not parallel safe. It also doesn't run on the generation executor, so no point.
+
+ // Only writes to the specified chunk. State is not read by later statuses. Parallel safe.
+ // Note: it may look unsafe because it writes to a worldgenregion, but the region size is always 0 -
+ // see the task margin.
+ // However, if the neighbouring FEATURES chunk is unloaded, but then fails to load in again (for whatever
+ // reason), then it would write to this chunk - and since this status reads blocks from itself, it's not
+ // safe to execute this in parallel.
+ // SPAWN
+
+ // No-op executor.
+ ChunkStatus.HEIGHTMAPS
+
+ // FULL is executed on main.
+ );
+
+ for (final ChunkStatus status : parallelCapableStatus) {
+ status.isParallelCapable = true;
+ }
+ }
+
+ public ChunkTaskScheduler(final ServerLevel world, final PrioritisedThreadPool workers) {
+ this.world = world;
+ this.workers = workers;
+
+ final String worldName = world.getWorld().getName();
+ this.genExecutor = workers.createExecutor("Chunk single-threaded generation executor for world '" + worldName + "'", 1);
+ // same as genExecutor, as there are race conditions between updating blocks in FEATURE status while lighting chunks
+ this.lightExecutor = this.genExecutor;
+ this.parallelGenExecutor = newChunkSystemGenParallelism <= 1 ? this.genExecutor
+ : workers.createExecutor("Chunk parallel generation executor for world '" + worldName + "'", newChunkSystemGenParallelism);
+ this.loadExecutor = workers.createExecutor("Chunk load executor for world '" + worldName + "'", newChunkSystemLoadParallelism);
+ this.chunkHolderManager = new ChunkHolderManager(world, this);
+ }
+
+ private final AtomicBoolean failedChunkSystem = new AtomicBoolean();
+
+ public static Object stringIfNull(final Object obj) {
+ return obj == null ? "null" : obj;
+ }
+
+ public void unrecoverableChunkSystemFailure(final int chunkX, final int chunkZ, final Map<String, Object> objectsOfInterest, final Throwable thr) {
+ final NewChunkHolder holder = this.chunkHolderManager.getChunkHolder(chunkX, chunkZ);
+ LOGGER.error("Chunk system error at chunk (" + chunkX + "," + chunkZ + "), holder: " + holder + ", exception:", new Throwable(thr));
+
+ if (this.failedChunkSystem.getAndSet(true)) {
+ return;
+ }
+
+ final ReportedException reportedException = thr instanceof ReportedException ? (ReportedException)thr : new ReportedException(new CrashReport("Chunk system error", thr));
+
+ CrashReportCategory crashReportCategory = reportedException.getReport().addCategory("Chunk system details");
+ crashReportCategory.setDetail("Chunk coordinate", new ChunkPos(chunkX, chunkZ).toString());
+ crashReportCategory.setDetail("ChunkHolder", Objects.toString(holder));
+ crashReportCategory.setDetail("unrecoverableChunkSystemFailure caller thread", Thread.currentThread().getName());
+
+ crashReportCategory = reportedException.getReport().addCategory("Chunk System Objects of Interest");
+ for (final Map.Entry<String, Object> entry : objectsOfInterest.entrySet()) {
+ if (entry.getValue() instanceof Throwable thrObject) {
+ crashReportCategory.setDetailError(Objects.toString(entry.getKey()), thrObject);
+ } else {
+ crashReportCategory.setDetail(Objects.toString(entry.getKey()), Objects.toString(entry.getValue()));
+ }
+ }
+
+ final Runnable crash = () -> {
+ throw new RuntimeException("Chunk system crash propagated from unrecoverableChunkSystemFailure", reportedException);
+ };
+
+ // this may not be good enough, specifically thanks to stupid ass plugins swallowing exceptions
+ this.scheduleChunkTask(chunkX, chunkZ, crash, PrioritisedExecutor.Priority.BLOCKING);
+ // so, make the main thread pick it up
+ MinecraftServer.chunkSystemCrash = new RuntimeException("Chunk system crash propagated from unrecoverableChunkSystemFailure", reportedException);
+ }
+
+ public boolean executeMainThreadTask() {
+ TickThread.ensureTickThread("Cannot execute main thread task off-main");
+ return this.mainThreadExecutor.executeTask();
+ }
+
+ public void raisePriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ this.chunkHolderManager.raisePriority(x, z, priority);
+ }
+
+ public void setPriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ this.chunkHolderManager.setPriority(x, z, priority);
+ }
+
+ public void lowerPriority(final int x, final int z, final PrioritisedExecutor.Priority priority) {
+ this.chunkHolderManager.lowerPriority(x, z, priority);
+ }
+
+ private final AtomicLong chunkLoadCounter = new AtomicLong();
+
+ public void scheduleTickingState(final int chunkX, final int chunkZ, final ChunkHolder.FullChunkStatus toStatus,
+ final boolean addTicket, final PrioritisedExecutor.Priority priority,
+ final Consumer<LevelChunk> onComplete) {
+ if (!TickThread.isTickThread()) {
+ this.scheduleChunkTask(chunkX, chunkZ, () -> {
+ ChunkTaskScheduler.this.scheduleTickingState(chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
+ }, priority);
+ return;
+ }
+ if (this.chunkHolderManager.ticketLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot schedule chunk load during ticket level update");
+ }
+ if (this.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot schedule chunk loading recursively");
+ }
+
+ if (toStatus == ChunkHolder.FullChunkStatus.INACCESSIBLE) {
+ throw new IllegalArgumentException("Cannot wait for INACCESSIBLE status");
+ }
+
+ final int minLevel = 33 - (toStatus.ordinal() - 1);
+ final Long chunkReference = addTicket ? Long.valueOf(this.chunkLoadCounter.getAndIncrement()) : null;
+ final long chunkKey = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+
+ if (addTicket) {
+ this.chunkHolderManager.addTicketAtLevel(TicketType.CHUNK_LOAD, chunkKey, minLevel, chunkReference);
+ this.chunkHolderManager.processTicketUpdates();
+ }
+
+ final Consumer<LevelChunk> loadCallback = (final LevelChunk chunk) -> {
+ try {
+ if (onComplete != null) {
+ onComplete.accept(chunk);
+ }
+ } finally {
+ if (addTicket) {
+ ChunkTaskScheduler.this.chunkHolderManager.addAndRemoveTickets(chunkKey,
+ TicketType.UNKNOWN, minLevel, new ChunkPos(chunkKey),
+ TicketType.CHUNK_LOAD, minLevel, chunkReference
+ );
+ }
+ }
+ };
+
+ final boolean scheduled;
+ final LevelChunk chunk;
+ this.chunkHolderManager.ticketLock.lock();
+ try {
+ this.schedulingLock.lock();
+ try {
+ final NewChunkHolder chunkHolder = this.chunkHolderManager.getChunkHolder(chunkKey);
+ if (chunkHolder == null || chunkHolder.getTicketLevel() > minLevel) {
+ scheduled = false;
+ chunk = null;
+ } else {
+ final ChunkHolder.FullChunkStatus currStatus = chunkHolder.getChunkStatus();
+ if (currStatus.isOrAfter(toStatus)) {
+ scheduled = false;
+ chunk = (LevelChunk)chunkHolder.getCurrentChunk();
+ } else {
+ scheduled = true;
+ chunk = null;
+
+ final int radius = toStatus.ordinal() - 1; // 0 -> BORDER, 1 -> TICKING, 2 -> ENTITY_TICKING
+ for (int dz = -radius; dz <= radius; ++dz) {
+ for (int dx = -radius; dx <= radius; ++dx) {
+ final NewChunkHolder neighbour =
+ (dx | dz) == 0 ? chunkHolder : this.chunkHolderManager.getChunkHolder(dx + chunkX, dz + chunkZ);
+ if (neighbour != null) {
+ neighbour.raisePriority(priority);
+ }
+ }
+ }
+
+ // ticket level should schedule for us
+ chunkHolder.addFullStatusConsumer(toStatus, loadCallback);
+ }
+ }
+ } finally {
+ this.schedulingLock.unlock();
+ }
+ } finally {
+ this.chunkHolderManager.ticketLock.unlock();
+ }
+
+ if (!scheduled) {
+ // couldn't schedule
+ try {
+ loadCallback.accept(chunk);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to process chunk full status callback", thr);
+ }
+ }
+ }
+
+ public void scheduleChunkLoad(final int chunkX, final int chunkZ, final boolean gen, final ChunkStatus toStatus, final boolean addTicket,
+ final PrioritisedExecutor.Priority priority, final Consumer<ChunkAccess> onComplete) {
+ if (gen) {
+ this.scheduleChunkLoad(chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
+ return;
+ }
+ this.scheduleChunkLoad(chunkX, chunkZ, ChunkStatus.EMPTY, addTicket, priority, (final ChunkAccess chunk) -> {
+ if (chunk == null) {
+ onComplete.accept(null);
+ } else {
+ if (chunk.getStatus().isOrAfter(toStatus)) {
+ this.scheduleChunkLoad(chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
+ } else {
+ onComplete.accept(null);
+ }
+ }
+ });
+ }
+
+ public void scheduleChunkLoad(final int chunkX, final int chunkZ, final ChunkStatus toStatus, final boolean addTicket,
+ final PrioritisedExecutor.Priority priority, final Consumer<ChunkAccess> onComplete) {
+ if (!TickThread.isTickThread()) {
+ this.scheduleChunkTask(chunkX, chunkZ, () -> {
+ ChunkTaskScheduler.this.scheduleChunkLoad(chunkX, chunkZ, toStatus, addTicket, priority, onComplete);
+ }, priority);
+ return;
+ }
+ if (this.chunkHolderManager.ticketLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot schedule chunk load during ticket level update");
+ }
+ if (this.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Cannot schedule chunk loading recursively");
+ }
+
+ if (toStatus == ChunkStatus.FULL) {
+ this.scheduleTickingState(chunkX, chunkZ, ChunkHolder.FullChunkStatus.BORDER, addTicket, priority, (Consumer)onComplete);
+ return;
+ }
+
+ final int minLevel = 33 + ChunkStatus.getDistance(toStatus);
+ final Long chunkReference = addTicket ? Long.valueOf(this.chunkLoadCounter.getAndIncrement()) : null;
+ final long chunkKey = CoordinateUtils.getChunkKey(chunkX, chunkZ);
+
+ if (addTicket) {
+ this.chunkHolderManager.addTicketAtLevel(TicketType.CHUNK_LOAD, chunkKey, minLevel, chunkReference);
+ this.chunkHolderManager.processTicketUpdates();
+ }
+
+ final Consumer<ChunkAccess> loadCallback = (final ChunkAccess chunk) -> {
+ try {
+ if (onComplete != null) {
+ onComplete.accept(chunk);
+ }
+ } finally {
+ if (addTicket) {
+ ChunkTaskScheduler.this.chunkHolderManager.addAndRemoveTickets(chunkKey,
+ TicketType.UNKNOWN, minLevel, new ChunkPos(chunkKey),
+ TicketType.CHUNK_LOAD, minLevel, chunkReference
+ );
+ }
+ }
+ };
+
+ final List<ChunkProgressionTask> tasks = new ArrayList<>();
+
+ final boolean scheduled;
+ final ChunkAccess chunk;
+ this.chunkHolderManager.ticketLock.lock();
+ try {
+ this.schedulingLock.lock();
+ try {
+ final NewChunkHolder chunkHolder = this.chunkHolderManager.getChunkHolder(chunkKey);
+ if (chunkHolder == null || chunkHolder.getTicketLevel() > minLevel) {
+ scheduled = false;
+ chunk = null;
+ } else {
+ final ChunkStatus genStatus = chunkHolder.getCurrentGenStatus();
+ if (genStatus != null && genStatus.isOrAfter(toStatus)) {
+ scheduled = false;
+ chunk = chunkHolder.getCurrentChunk();
+ } else {
+ scheduled = true;
+ chunk = null;
+ chunkHolder.raisePriority(priority);
+
+ if (!chunkHolder.upgradeGenTarget(toStatus)) {
+ this.schedule(chunkX, chunkZ, toStatus, chunkHolder, tasks);
+ }
+ chunkHolder.addStatusConsumer(toStatus, loadCallback);
+ }
+ }
+ } finally {
+ this.schedulingLock.unlock();
+ }
+ } finally {
+ this.chunkHolderManager.ticketLock.unlock();
+ }
+
+ for (int i = 0, len = tasks.size(); i < len; ++i) {
+ tasks.get(i).schedule();
+ }
+
+ if (!scheduled) {
+ // couldn't schedule
+ try {
+ loadCallback.accept(chunk);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to process chunk status callback", thr);
+ }
+ }
+ }
+
+ private ChunkProgressionTask createTask(final int chunkX, final int chunkZ, final ChunkAccess chunk,
+ final NewChunkHolder chunkHolder, final List<ChunkAccess> neighbours,
+ final ChunkStatus toStatus, final PrioritisedExecutor.Priority initialPriority) {
+ if (toStatus == ChunkStatus.EMPTY) {
+ return new ChunkLoadTask(this, this.world, chunkX, chunkZ, chunkHolder, initialPriority);
+ }
+ if (toStatus == ChunkStatus.LIGHT) {
+ return new ChunkLightTask(this, this.world, chunkX, chunkZ, chunk, initialPriority);
+ }
+ if (toStatus == ChunkStatus.FULL) {
+ return new ChunkFullTask(this, this.world, chunkX, chunkZ, chunkHolder, chunk, initialPriority);
+ }
+
+ return new ChunkUpgradeGenericStatusTask(this, this.world, chunkX, chunkZ, chunk, neighbours, toStatus, initialPriority);
+ }
+
+ ChunkProgressionTask schedule(final int chunkX, final int chunkZ, final ChunkStatus targetStatus, final NewChunkHolder chunkHolder,
+ final List<ChunkProgressionTask> allTasks) {
+ return this.schedule(chunkX, chunkZ, targetStatus, chunkHolder, allTasks, chunkHolder.getEffectivePriority());
+ }
+
+ // rets new task scheduled for the _specified_ chunk
+ // note: this must hold the scheduling lock
+ // minPriority is only used to pass the priority through to neighbours, as priority calculation has not yet been done
+ // schedule will ignore the generation target, so it should be checked by the caller to ensure the target is not regressed!
+ private ChunkProgressionTask schedule(final int chunkX, final int chunkZ, final ChunkStatus targetStatus,
+ final NewChunkHolder chunkHolder, final List<ChunkProgressionTask> allTasks,
+ final PrioritisedExecutor.Priority minPriority) {
+ if (!this.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Not holding scheduling lock");
+ }
+
+ if (chunkHolder.hasGenerationTask()) {
+ chunkHolder.upgradeGenTarget(targetStatus);
+ return null;
+ }
+
+ final PrioritisedExecutor.Priority requestedPriority = PrioritisedExecutor.Priority.max(minPriority, chunkHolder.getEffectivePriority());
+ final ChunkStatus currentGenStatus = chunkHolder.getCurrentGenStatus();
+ final ChunkAccess chunk = chunkHolder.getCurrentChunk();
+
+ if (currentGenStatus == null) {
+ // not yet loaded
+ final ChunkProgressionTask task = this.createTask(
+ chunkX, chunkZ, chunk, chunkHolder, Collections.emptyList(), ChunkStatus.EMPTY, requestedPriority
+ );
+
+ allTasks.add(task);
+
+ final List<NewChunkHolder> chunkHolderNeighbours = new ArrayList<>(1);
+ chunkHolderNeighbours.add(chunkHolder);
+
+ chunkHolder.setGenerationTarget(targetStatus);
+ chunkHolder.setGenerationTask(task, ChunkStatus.EMPTY, chunkHolderNeighbours);
+
+ return task;
+ }
+
+ if (currentGenStatus.isOrAfter(targetStatus)) {
+ // nothing to do
+ return null;
+ }
+
+ // we know for sure now that we want to schedule _something_, so set the target
+ chunkHolder.setGenerationTarget(targetStatus);
+
+ final ChunkStatus chunkRealStatus = chunk.getStatus();
+ final ChunkStatus toStatus = currentGenStatus.getNextStatus();
+
+ // if this chunk has already generated up to or past the specified status, then we don't
+ // need the neighbours AT ALL.
+ final int neighbourReadRadius = chunkRealStatus.isOrAfter(toStatus) ? toStatus.loadRange : toStatus.getRange();
+
+ boolean unGeneratedNeighbours = false;
+
+ // copied from MCUtil.getSpiralOutChunks
+ for (int r = 1; r <= neighbourReadRadius; r++) {
+ int x = -r;
+ int z = r;
+
+ // Iterates the edge of half of the box; then negates for other half.
+ while (x <= r && z > -r) {
+ final int radius = Math.max(Math.abs(x), Math.abs(z));
+ final ChunkStatus requiredNeighbourStatus = ChunkMap.getDependencyStatus(toStatus, radius);
+
+ unGeneratedNeighbours |= this.checkNeighbour(
+ chunkX + x, chunkZ + z, requiredNeighbourStatus, chunkHolder, allTasks, requestedPriority
+ );
+ unGeneratedNeighbours |= this.checkNeighbour(
+ chunkX - x, chunkZ - z, requiredNeighbourStatus, chunkHolder, allTasks, requestedPriority
+ );
+
+ if (x < r) {
+ x++;
+ } else {
+ z--;
+ }
+ }
+ }
+
+ if (unGeneratedNeighbours) {
+ // can't schedule, but neighbour completion will schedule for us when they're ALL done
+
+ // propagate our priority to neighbours
+ chunkHolder.recalculateNeighbourPriorities();
+ return null;
+ }
+
+ // need to gather neighbours
+
+ final List<ChunkAccess> neighbours;
+ final List<NewChunkHolder> chunkHolderNeighbours;
+ if (neighbourReadRadius <= 0) {
+ neighbours = new ArrayList<>(1);
+ chunkHolderNeighbours = new ArrayList<>(1);
+ neighbours.add(chunk);
+ chunkHolderNeighbours.add(chunkHolder);
+ } else {
+ // the iteration order is _very_ important, as all generation statuses expect a certain order such that:
+ // chunkAtRelative = neighbours.get(relX + relZ * (2 * radius + 1))
+ neighbours = new ArrayList<>((2 * neighbourReadRadius + 1) * (2 * neighbourReadRadius + 1));
+ chunkHolderNeighbours = new ArrayList<>((2 * neighbourReadRadius + 1) * (2 * neighbourReadRadius + 1));
+ for (int dz = -neighbourReadRadius; dz <= neighbourReadRadius; ++dz) {
+ for (int dx = -neighbourReadRadius; dx <= neighbourReadRadius; ++dx) {
+ final NewChunkHolder holder = (dx | dz) == 0 ? chunkHolder : this.chunkHolderManager.getChunkHolder(dx + chunkX, dz + chunkZ);
+ neighbours.add(holder.getChunkForNeighbourAccess());
+ chunkHolderNeighbours.add(holder);
+ }
+ }
+ }
+
+ final ChunkProgressionTask task = this.createTask(chunkX, chunkZ, chunk, chunkHolder, neighbours, toStatus, chunkHolder.getEffectivePriority());
+ allTasks.add(task);
+
+ chunkHolder.setGenerationTask(task, toStatus, chunkHolderNeighbours);
+
+ return task;
+ }
+
+ // rets true if the neighbour is not at the required status, false otherwise
+ private boolean checkNeighbour(final int chunkX, final int chunkZ, final ChunkStatus requiredStatus, final NewChunkHolder center,
+ final List<ChunkProgressionTask> tasks, final PrioritisedExecutor.Priority minPriority) {
+ final NewChunkHolder chunkHolder = this.chunkHolderManager.getChunkHolder(chunkX, chunkZ);
+
+ if (chunkHolder == null) {
+ throw new IllegalStateException("Missing chunkholder when required");
+ }
+
+ final ChunkStatus holderStatus = chunkHolder.getCurrentGenStatus();
+ if (holderStatus != null && holderStatus.isOrAfter(requiredStatus)) {
+ return false;
+ }
+
+ if (chunkHolder.hasFailedGeneration()) {
+ return true;
+ }
+
+ center.addGenerationBlockingNeighbour(chunkHolder);
+ chunkHolder.addWaitingNeighbour(center, requiredStatus);
+
+ if (chunkHolder.upgradeGenTarget(requiredStatus)) {
+ return true;
+ }
+
+ // not at status required, so we need to schedule its generation
+ this.schedule(
+ chunkX, chunkZ, requiredStatus, chunkHolder, tasks, minPriority
+ );
+
+ return true;
+ }
+
+ /**
+ * @deprecated Chunk tasks must be tied to coordinates in the future
+ */
+ @Deprecated
+ public PrioritisedExecutor.PrioritisedTask scheduleChunkTask(final Runnable run) {
+ return this.scheduleChunkTask(run, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ /**
+ * @deprecated Chunk tasks must be tied to coordinates in the future
+ */
+ @Deprecated
+ public PrioritisedExecutor.PrioritisedTask scheduleChunkTask(final Runnable run, final PrioritisedExecutor.Priority priority) {
+ return this.mainThreadExecutor.queueRunnable(run, priority);
+ }
+
+ public PrioritisedExecutor.PrioritisedTask createChunkTask(final int chunkX, final int chunkZ, final Runnable run) {
+ return this.createChunkTask(chunkX, chunkZ, run, PrioritisedExecutor.Priority.NORMAL);
+ }
+
+ public PrioritisedExecutor.PrioritisedTask createChunkTask(final int chunkX, final int chunkZ, final Runnable run,
+ final PrioritisedExecutor.Priority priority) {
+ return this.mainThreadExecutor.createTask(run, priority);
+ }
+
+ public PrioritisedExecutor.PrioritisedTask scheduleChunkTask(final int chunkX, final int chunkZ, final Runnable run) {
+ return this.mainThreadExecutor.queueRunnable(run);
+ }
+
+ public PrioritisedExecutor.PrioritisedTask scheduleChunkTask(final int chunkX, final int chunkZ, final Runnable run,
+ final PrioritisedExecutor.Priority priority) {
+ return this.mainThreadExecutor.queueRunnable(run, priority);
+ }
+
+ public void executeTasksUntil(final BooleanSupplier exit) {
+ if (Bukkit.isPrimaryThread()) {
+ this.mainThreadExecutor.executeConditionally(exit);
+ } else {
+ long counter = 1L;
+ while (!exit.getAsBoolean()) {
+ counter = ConcurrentUtil.linearLongBackoff(counter, 100_000L, 5_000_000L); // 100us, 5ms
+ }
+ }
+ }
+
+ public boolean halt(final boolean sync, final long maxWaitNS) {
+ this.lightExecutor.halt();
+ this.genExecutor.halt();
+ this.parallelGenExecutor.halt();
+ this.loadExecutor.halt();
+ final long time = System.nanoTime();
+ if (sync) {
+ for (long failures = 9L;; failures = ConcurrentUtil.linearLongBackoff(failures, 500_000L, 50_000_000L)) {
+ if (
+ !this.lightExecutor.isActive() &&
+ !this.genExecutor.isActive() &&
+ !this.parallelGenExecutor.isActive() &&
+ !this.loadExecutor.isActive()
+ ) {
+ return true;
+ }
+ if ((System.nanoTime() - time) >= maxWaitNS) {
+ return false;
+ }
+ }
+ }
+
+ return true;
+ }
+
+ public static final ArrayDeque<ChunkInfo> WAITING_CHUNKS = new ArrayDeque<>(); // stack
+
+ public static final class ChunkInfo {
+
+ public final int chunkX;
+ public final int chunkZ;
+ public final ServerLevel world;
+
+ public ChunkInfo(final int chunkX, final int chunkZ, final ServerLevel world) {
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.world = world;
+ }
+
+ @Override
+ public String toString() {
+ return "[( " + this.chunkX + "," + this.chunkZ + ") in '" + this.world.getWorld().getName() + "']";
+ }
+ }
+
+ public static void pushChunkWait(final ServerLevel world, final int chunkX, final int chunkZ) {
+ synchronized (WAITING_CHUNKS) {
+ WAITING_CHUNKS.push(new ChunkInfo(chunkX, chunkZ, world));
+ }
+ }
+
+ public static void popChunkWait() {
+ synchronized (WAITING_CHUNKS) {
+ WAITING_CHUNKS.pop();
+ }
+ }
+
+ public static ChunkInfo[] getChunkInfos() {
+ synchronized (WAITING_CHUNKS) {
+ return WAITING_CHUNKS.toArray(new ChunkInfo[0]);
+ }
+ }
+
+ public static void dumpAllChunkLoadInfo(final boolean longPrint) {
+ final ChunkInfo[] chunkInfos = getChunkInfos();
+ if (chunkInfos.length > 0) {
+ LOGGER.error("Chunk wait task info below: ");
+ for (final ChunkInfo chunkInfo : chunkInfos) {
+ final NewChunkHolder holder = chunkInfo.world.chunkTaskScheduler.chunkHolderManager.getChunkHolder(chunkInfo.chunkX, chunkInfo.chunkZ);
+ LOGGER.error("Chunk wait: " + chunkInfo);
+ LOGGER.error("Chunk holder: " + holder);
+ }
+
+ if (longPrint) {
+ final File file = new File(new File(new File("."), "debug"), "chunks-watchdog.txt");
+ LOGGER.error("Writing chunk information dump to " + file);
+ try {
+ MCUtil.dumpChunks(file, true);
+ LOGGER.error("Successfully written chunk information!");
+ } catch (final Throwable thr) {
+ MinecraftServer.LOGGER.warn("Failed to dump chunk information to file " + file.toString(), thr);
+ }
+ }
+ }
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkUpgradeGenericStatusTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkUpgradeGenericStatusTask.java
new file mode 100644
index 0000000000000000000000000000000000000000..73ce0909bd89244835a0d0f2030a25871461f1e0
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/ChunkUpgradeGenericStatusTask.java
@@ -0,0 +1,209 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import com.mojang.datafixers.util.Either;
+import com.mojang.logging.LogUtils;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerChunkCache;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.ProtoChunk;
+import org.slf4j.Logger;
+import java.lang.invoke.VarHandle;
+import java.util.List;
+import java.util.Map;
+import java.util.concurrent.CompletableFuture;
+
+public final class ChunkUpgradeGenericStatusTask extends ChunkProgressionTask implements Runnable {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ protected final ChunkAccess fromChunk;
+ protected final ChunkStatus fromStatus;
+ protected final ChunkStatus toStatus;
+ protected final List<ChunkAccess> neighbours;
+
+ protected final PrioritisedExecutor.PrioritisedTask generateTask;
+
+ public ChunkUpgradeGenericStatusTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final ChunkAccess chunk, final List<ChunkAccess> neighbours,
+ final ChunkStatus toStatus, final PrioritisedExecutor.Priority priority) {
+ super(scheduler, world, chunkX, chunkZ);
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.fromChunk = chunk;
+ this.fromStatus = chunk.getStatus();
+ this.toStatus = toStatus;
+ this.neighbours = neighbours;
+ this.generateTask = (this.toStatus.isParallelCapable ? this.scheduler.parallelGenExecutor : this.scheduler.genExecutor)
+ .createTask(this, priority);
+ }
+
+ @Override
+ public ChunkStatus getTargetStatus() {
+ return this.toStatus;
+ }
+
+ private boolean isEmptyTask() {
+ // must use fromStatus here to avoid any race condition with run() overwriting the status
+ final boolean generation = !this.fromStatus.isOrAfter(this.toStatus);
+ return (generation && this.toStatus.isEmptyGenStatus()) || (!generation && this.toStatus.isEmptyLoadStatus());
+ }
+
+ @Override
+ public void run() {
+ final ChunkAccess chunk = this.fromChunk;
+
+ final ServerChunkCache serverChunkCache = this.world.chunkSource;
+ final ChunkMap chunkMap = serverChunkCache.chunkMap;
+
+ final CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completeFuture;
+
+ final boolean generation;
+ boolean completing = false;
+
+ // note: should optimise the case where the chunk does not need to execute the status, because
+ // schedule() calls this synchronously if it will run through that path
+
+ try {
+ generation = !chunk.getStatus().isOrAfter(this.toStatus);
+ if (generation) {
+ if (this.toStatus.isEmptyGenStatus()) {
+ if (chunk instanceof ProtoChunk) {
+ ((ProtoChunk)chunk).setStatus(this.toStatus);
+ }
+ completing = true;
+ this.complete(chunk, null);
+ return;
+ }
+ completeFuture = this.toStatus.generate(Runnable::run, this.world, chunkMap.generator, chunkMap.structureTemplateManager,
+ serverChunkCache.getLightEngine(), null, this.neighbours, false)
+ .whenComplete((final Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either, final Throwable throwable) -> {
+ final ChunkAccess newChunk = (either == null) ? null : either.left().orElse(null);
+ if (newChunk instanceof ProtoChunk) {
+ ((ProtoChunk)newChunk).setStatus(ChunkUpgradeGenericStatusTask.this.toStatus);
+ }
+ }
+ );
+ } else {
+ if (this.toStatus.isEmptyLoadStatus()) {
+ completing = true;
+ this.complete(chunk, null);
+ return;
+ }
+ completeFuture = this.toStatus.load(this.world, chunkMap.structureTemplateManager, serverChunkCache.getLightEngine(), null, chunk);
+ }
+ } catch (final Throwable throwable) {
+ if (!completing) {
+ this.complete(null, throwable);
+
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ return;
+ }
+
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Target status", ChunkTaskScheduler.stringIfNull(this.toStatus),
+ "From status", ChunkTaskScheduler.stringIfNull(this.fromStatus),
+ "Generation task", this
+ ), throwable);
+
+ if (!(throwable instanceof ThreadDeath)) {
+ LOGGER.error("Failed to complete status for chunk: status:" + this.toStatus + ", chunk: (" + this.chunkX + "," + this.chunkZ + "), world: " + this.world.getWorld().getName(), throwable);
+ } else {
+ // ensure the chunk system can respond, then die
+ throw (ThreadDeath)throwable;
+ }
+ return;
+ }
+
+ if (!completeFuture.isDone() && !this.toStatus.warnedAboutNoImmediateComplete.getAndSet(true)) {
+ LOGGER.warn("Future status not complete after scheduling: " + this.toStatus.toString() + ", generate: " + generation);
+ }
+
+ final Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either;
+ final ChunkAccess newChunk;
+
+ try {
+ either = completeFuture.join();
+ newChunk = (either == null) ? null : either.left().orElse(null);
+ } catch (final Throwable throwable) {
+ this.complete(null, throwable);
+ // ensure the chunk system can respond, then die
+ if (throwable instanceof ThreadDeath) {
+ throw (ThreadDeath)throwable;
+ }
+ return;
+ }
+
+ if (newChunk == null) {
+ this.complete(null, new IllegalStateException("Chunk for status: " + ChunkUpgradeGenericStatusTask.this.toStatus.toString() + ", generation: " + generation + " should not be null! Either: " + either).fillInStackTrace());
+ return;
+ }
+
+ this.complete(newChunk, null);
+ }
+
+ protected volatile boolean scheduled;
+ protected static final VarHandle SCHEDULED_HANDLE = ConcurrentUtil.getVarHandle(ChunkUpgradeGenericStatusTask.class, "scheduled", boolean.class);
+
+ @Override
+ public boolean isScheduled() {
+ return this.scheduled;
+ }
+
+ @Override
+ public void schedule() {
+ if ((boolean)SCHEDULED_HANDLE.getAndSet((ChunkUpgradeGenericStatusTask)this, true)) {
+ throw new IllegalStateException("Cannot double call schedule()");
+ }
+ if (this.isEmptyTask()) {
+ if (this.generateTask.cancel()) {
+ this.run();
+ }
+ } else {
+ this.generateTask.queue();
+ }
+ }
+
+ @Override
+ public void cancel() {
+ if (this.generateTask.cancel()) {
+ this.complete(null, null);
+ }
+ }
+
+ @Override
+ public PrioritisedExecutor.Priority getPriority() {
+ return this.generateTask.getPriority();
+ }
+
+ @Override
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.generateTask.lowerPriority(priority);
+ }
+
+ @Override
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.generateTask.setPriority(priority);
+ }
+
+ @Override
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.generateTask.raisePriority(priority);
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/GenericDataLoadTask.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/GenericDataLoadTask.java
new file mode 100644
index 0000000000000000000000000000000000000000..396d72c00e47cf1669ae20dc839c1c961b1f262a
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/GenericDataLoadTask.java
@@ -0,0 +1,746 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.completable.Completable;
+import ca.spottedleaf.concurrentutil.executor.Cancellable;
+import ca.spottedleaf.concurrentutil.executor.standard.DelayedPrioritisedTask;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.chunk.system.io.RegionFileIOThread;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ServerLevel;
+import org.slf4j.Logger;
+import java.lang.invoke.VarHandle;
+import java.util.Map;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicLong;
+import java.util.function.BiConsumer;
+
+public abstract class GenericDataLoadTask<OnMain,FinalCompletion> {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ protected static final CompoundTag CANCELLED_DATA = new CompoundTag();
+
+ // reference count is the upper 32 bits
+ protected final AtomicLong stageAndReferenceCount = new AtomicLong(STAGE_NOT_STARTED);
+
+ protected static final long STAGE_MASK = 0xFFFFFFFFL;
+ protected static final long STAGE_CANCELLED = 0xFFFFFFFFL;
+ protected static final long STAGE_NOT_STARTED = 0L;
+ protected static final long STAGE_LOADING = 1L;
+ protected static final long STAGE_PROCESSING = 2L;
+ protected static final long STAGE_COMPLETED = 3L;
+
+ // for loading data off disk
+ protected final LoadDataFromDiskTask loadDataFromDiskTask;
+ // processing off-main
+ protected final PrioritisedExecutor.PrioritisedTask processOffMain;
+ // processing on-main
+ protected final PrioritisedExecutor.PrioritisedTask processOnMain;
+
+ protected final ChunkTaskScheduler scheduler;
+ protected final ServerLevel world;
+ protected final int chunkX;
+ protected final int chunkZ;
+ protected final RegionFileIOThread.RegionFileType type;
+
+ public GenericDataLoadTask(final ChunkTaskScheduler scheduler, final ServerLevel world, final int chunkX,
+ final int chunkZ, final RegionFileIOThread.RegionFileType type,
+ final PrioritisedExecutor.Priority priority) {
+ this.scheduler = scheduler;
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.type = type;
+
+ final ProcessOnMainTask mainTask;
+ if (this.hasOnMain()) {
+ mainTask = new ProcessOnMainTask();
+ this.processOnMain = this.createOnMain(mainTask, priority);
+ } else {
+ mainTask = null;
+ this.processOnMain = null;
+ }
+
+ final ProcessOffMainTask offMainTask;
+ if (this.hasOffMain()) {
+ offMainTask = new ProcessOffMainTask(mainTask);
+ this.processOffMain = this.createOffMain(offMainTask, priority);
+ } else {
+ offMainTask = null;
+ this.processOffMain = null;
+ }
+
+ if (this.processOffMain == null && this.processOnMain == null) {
+ throw new IllegalStateException("Illegal class implementation: " + this.getClass().getName() + ", should be able to schedule at least one task!");
+ }
+
+ this.loadDataFromDiskTask = new LoadDataFromDiskTask(world, chunkX, chunkZ, type, new DataLoadCallback(offMainTask, mainTask), priority);
+ }
+
+ public static final record TaskResult<L, R>(L left, R right) {}
+
+ protected abstract boolean hasOffMain();
+
+ protected abstract boolean hasOnMain();
+
+ protected abstract PrioritisedExecutor.PrioritisedTask createOffMain(final Runnable run, final PrioritisedExecutor.Priority priority);
+
+ protected abstract PrioritisedExecutor.PrioritisedTask createOnMain(final Runnable run, final PrioritisedExecutor.Priority priority);
+
+ protected abstract TaskResult<OnMain, Throwable> runOffMain(final CompoundTag data, final Throwable throwable);
+
+ protected abstract TaskResult<FinalCompletion, Throwable> runOnMain(final OnMain data, final Throwable throwable);
+
+ protected abstract void onComplete(final TaskResult<FinalCompletion,Throwable> result);
+
+ protected abstract TaskResult<FinalCompletion, Throwable> completeOnMainOffMain(final OnMain data, final Throwable throwable);
+
+ @Override
+ public String toString() {
+ return "GenericDataLoadTask{class: " + this.getClass().getName() + ", world: " + this.world.getWorld().getName() +
+ ", chunk: (" + this.chunkX + "," + this.chunkZ + "), hashcode: " + System.identityHashCode(this) + ", priority: " + this.getPriority() +
+ ", type: " + this.type.toString() + "}";
+ }
+
+ public PrioritisedExecutor.Priority getPriority() {
+ if (this.processOnMain != null) {
+ return this.processOnMain.getPriority();
+ } else {
+ return this.processOffMain.getPriority();
+ }
+ }
+
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ // can't lower I/O tasks, we don't know what they affect
+ if (this.processOffMain != null) {
+ this.processOffMain.lowerPriority(priority);
+ }
+ if (this.processOnMain != null) {
+ this.processOnMain.lowerPriority(priority);
+ }
+ }
+
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ // can't lower I/O tasks, we don't know what they affect
+ this.loadDataFromDiskTask.raisePriority(priority);
+ if (this.processOffMain != null) {
+ this.processOffMain.setPriority(priority);
+ }
+ if (this.processOnMain != null) {
+ this.processOnMain.setPriority(priority);
+ }
+ }
+
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ // can't lower I/O tasks, we don't know what they affect
+ this.loadDataFromDiskTask.raisePriority(priority);
+ if (this.processOffMain != null) {
+ this.processOffMain.raisePriority(priority);
+ }
+ if (this.processOnMain != null) {
+ this.processOnMain.raisePriority(priority);
+ }
+ }
+
+ // returns whether scheduleNow() needs to be called
+ public boolean schedule(final boolean delay) {
+ if (this.stageAndReferenceCount.get() != STAGE_NOT_STARTED ||
+ !this.stageAndReferenceCount.compareAndSet(STAGE_NOT_STARTED, (1L << 32) | STAGE_LOADING)) {
+ // try and increment reference count
+ int failures = 0;
+ for (long curr = this.stageAndReferenceCount.get();;) {
+ if ((curr & STAGE_MASK) == STAGE_CANCELLED || (curr & STAGE_MASK) == STAGE_COMPLETED) {
+ // cancelled or completed, nothing to do here
+ return false;
+ }
+
+ if (curr == (curr = this.stageAndReferenceCount.compareAndExchange(curr, curr + (1L << 32)))) {
+ // successful
+ return false;
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ if (!delay) {
+ this.scheduleNow();
+ return false;
+ }
+ return true;
+ }
+
+ public void scheduleNow() {
+ this.loadDataFromDiskTask.schedule(); // will schedule the rest
+ }
+
+ // assumes the current stage cannot be completed
+ // returns false if cancelled, returns true if can proceed
+ private boolean advanceStage(final long expect, final long to) {
+ int failures = 0;
+ for (long curr = this.stageAndReferenceCount.get();;) {
+ if ((curr & STAGE_MASK) != expect) {
+ // must be cancelled
+ return false;
+ }
+
+ final long newVal = (curr & ~STAGE_MASK) | to;
+ if (curr == (curr = this.stageAndReferenceCount.compareAndExchange(curr, newVal))) {
+ return true;
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public boolean cancel() {
+ int failures = 0;
+ for (long curr = this.stageAndReferenceCount.get();;) {
+ if ((curr & STAGE_MASK) == STAGE_COMPLETED || (curr & STAGE_MASK) == STAGE_CANCELLED) {
+ return false;
+ }
+
+ if ((curr & STAGE_MASK) == STAGE_NOT_STARTED || (curr & ~STAGE_MASK) == (1L << 32)) {
+ // no other references, so we can cancel
+ final long newVal = STAGE_CANCELLED;
+ if (curr == (curr = this.stageAndReferenceCount.compareAndExchange(curr, newVal))) {
+ this.loadDataFromDiskTask.cancel();
+ if (this.processOffMain != null) {
+ this.processOffMain.cancel();
+ }
+ if (this.processOnMain != null) {
+ this.processOnMain.cancel();
+ }
+ this.onComplete(null);
+ return true;
+ }
+ } else {
+ if ((curr & ~STAGE_MASK) == (0L << 32)) {
+ throw new IllegalStateException("Reference count cannot be zero here");
+ }
+ // just decrease the reference count
+ final long newVal = curr - (1L << 32);
+ if (curr == (curr = this.stageAndReferenceCount.compareAndExchange(curr, newVal))) {
+ return false;
+ }
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ protected final class DataLoadCallback implements BiConsumer<CompoundTag, Throwable> {
+
+ protected final ProcessOffMainTask offMainTask;
+ protected final ProcessOnMainTask onMainTask;
+
+ public DataLoadCallback(final ProcessOffMainTask offMainTask, final ProcessOnMainTask onMainTask) {
+ this.offMainTask = offMainTask;
+ this.onMainTask = onMainTask;
+ }
+
+ @Override
+ public void accept(final CompoundTag compoundTag, final Throwable throwable) {
+ if (GenericDataLoadTask.this.stageAndReferenceCount.get() == STAGE_CANCELLED) {
+ // don't try to schedule further
+ return;
+ }
+
+ try {
+ if (compoundTag == CANCELLED_DATA) {
+ // cancelled, except this isn't possible
+ LOGGER.error("Data callback says cancelled, but stage does not?");
+ return;
+ }
+
+ // get off of the regionfile callback ASAP, no clue what locks are held right now...
+ if (GenericDataLoadTask.this.processOffMain != null) {
+ this.offMainTask.data = compoundTag;
+ this.offMainTask.throwable = throwable;
+ GenericDataLoadTask.this.processOffMain.queue();
+ return;
+ } else {
+ // no off-main task, so go straight to main
+ this.onMainTask.data = (OnMain)compoundTag;
+ this.onMainTask.throwable = throwable;
+ GenericDataLoadTask.this.processOnMain.queue();
+ }
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr2) {
+ LOGGER.error("Failed I/O callback for task: " + GenericDataLoadTask.this.toString(), thr2);
+ GenericDataLoadTask.this.scheduler.unrecoverableChunkSystemFailure(
+ GenericDataLoadTask.this.chunkX, GenericDataLoadTask.this.chunkZ, Map.of(
+ "Callback throwable", ChunkTaskScheduler.stringIfNull(throwable)
+ ), thr2);
+ }
+ }
+ }
+
+ protected final class ProcessOffMainTask implements Runnable {
+
+ protected CompoundTag data;
+ protected Throwable throwable;
+ protected final ProcessOnMainTask schedule;
+
+ public ProcessOffMainTask(final ProcessOnMainTask schedule) {
+ this.schedule = schedule;
+ }
+
+ @Override
+ public void run() {
+ if (!GenericDataLoadTask.this.advanceStage(STAGE_LOADING, this.schedule == null ? STAGE_COMPLETED : STAGE_PROCESSING)) {
+ // cancelled
+ return;
+ }
+ final TaskResult<OnMain, Throwable> newData = GenericDataLoadTask.this.runOffMain(this.data, this.throwable);
+
+ if (GenericDataLoadTask.this.stageAndReferenceCount.get() == STAGE_CANCELLED) {
+ // don't try to schedule further
+ return;
+ }
+
+ if (this.schedule != null) {
+ final TaskResult<FinalCompletion, Throwable> syncComplete = GenericDataLoadTask.this.completeOnMainOffMain(newData.left, newData.right);
+
+ if (syncComplete != null) {
+ if (GenericDataLoadTask.this.advanceStage(STAGE_PROCESSING, STAGE_COMPLETED)) {
+ GenericDataLoadTask.this.onComplete(syncComplete);
+ } // else: cancelled
+ return;
+ }
+
+ this.schedule.data = newData.left;
+ this.schedule.throwable = newData.right;
+
+ GenericDataLoadTask.this.processOnMain.queue();
+ } else {
+ GenericDataLoadTask.this.onComplete((TaskResult<FinalCompletion, Throwable>)newData);
+ }
+ }
+ }
+
+ protected final class ProcessOnMainTask implements Runnable {
+
+ protected OnMain data;
+ protected Throwable throwable;
+
+ @Override
+ public void run() {
+ if (!GenericDataLoadTask.this.advanceStage(STAGE_PROCESSING, STAGE_COMPLETED)) {
+ // cancelled
+ return;
+ }
+ final TaskResult<FinalCompletion, Throwable> result = GenericDataLoadTask.this.runOnMain(this.data, this.throwable);
+
+ GenericDataLoadTask.this.onComplete(result);
+ }
+ }
+
+ public static final class LoadDataFromDiskTask {
+
+ protected volatile int priority;
+ protected static final VarHandle PRIORITY_HANDLE = ConcurrentUtil.getVarHandle(LoadDataFromDiskTask.class, "priority", int.class);
+
+ protected static final int PRIORITY_EXECUTED = Integer.MIN_VALUE >>> 0;
+ protected static final int PRIORITY_LOAD_SCHEDULED = Integer.MIN_VALUE >>> 1;
+ protected static final int PRIORITY_UNLOAD_SCHEDULED = Integer.MIN_VALUE >>> 2;
+
+ protected static final int PRIORITY_FLAGS = ~Character.MAX_VALUE;
+
+ protected final int getPriorityVolatile() {
+ return (int)PRIORITY_HANDLE.getVolatile((LoadDataFromDiskTask)this);
+ }
+
+ protected final int compareAndExchangePriorityVolatile(final int expect, final int update) {
+ return (int)PRIORITY_HANDLE.compareAndExchange((LoadDataFromDiskTask)this, (int)expect, (int)update);
+ }
+
+ protected final int getAndOrPriorityVolatile(final int val) {
+ return (int)PRIORITY_HANDLE.getAndBitwiseOr((LoadDataFromDiskTask)this, (int)val);
+ }
+
+ protected final void setPriorityPlain(final int val) {
+ PRIORITY_HANDLE.set((LoadDataFromDiskTask)this, (int)val);
+ }
+
+ private final ServerLevel world;
+ private final int chunkX;
+ private final int chunkZ;
+
+ private final RegionFileIOThread.RegionFileType type;
+ private Cancellable dataLoadTask;
+ private Cancellable dataUnloadCancellable;
+ private DelayedPrioritisedTask dataUnloadTask;
+
+ private final BiConsumer<CompoundTag, Throwable> onComplete;
+
+ // onComplete should be caller sensitive, it may complete synchronously with schedule() - which does
+ // hold a priority lock.
+ public LoadDataFromDiskTask(final ServerLevel world, final int chunkX, final int chunkZ,
+ final RegionFileIOThread.RegionFileType type,
+ final BiConsumer<CompoundTag, Throwable> onComplete,
+ final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.type = type;
+ this.onComplete = onComplete;
+ this.setPriorityPlain(priority.priority);
+ }
+
+ private void complete(final CompoundTag data, final Throwable throwable) {
+ try {
+ this.onComplete.accept(data, throwable);
+ } catch (final Throwable thr2) {
+ this.world.chunkTaskScheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Completed throwable", ChunkTaskScheduler.stringIfNull(throwable),
+ "Regionfile type", ChunkTaskScheduler.stringIfNull(this.type)
+ ), thr2);
+ if (thr2 instanceof ThreadDeath) {
+ throw (ThreadDeath)thr2;
+ }
+ }
+ }
+
+ protected boolean markExecuting() {
+ return (this.getAndOrPriorityVolatile(PRIORITY_EXECUTED) & PRIORITY_EXECUTED) == 0;
+ }
+
+ protected boolean isMarkedExecuted() {
+ return (this.getPriorityVolatile() & PRIORITY_EXECUTED) != 0;
+ }
+
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed
+ return;
+ }
+
+ if ((curr & PRIORITY_LOAD_SCHEDULED) != 0) {
+ RegionFileIOThread.lowerPriority(this.world, this.chunkX, this.chunkZ, this.type, priority);
+ return;
+ }
+
+ if ((curr & PRIORITY_UNLOAD_SCHEDULED) != 0) {
+ if (this.dataUnloadTask != null) {
+ this.dataUnloadTask.lowerPriority(priority);
+ }
+ // no return - we need to propagate priority
+ }
+
+ if (!priority.isHigherPriority(curr & ~PRIORITY_FLAGS)) {
+ return;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority | (curr & PRIORITY_FLAGS)))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed
+ return;
+ }
+
+ if ((curr & PRIORITY_LOAD_SCHEDULED) != 0) {
+ RegionFileIOThread.setPriority(this.world, this.chunkX, this.chunkZ, this.type, priority);
+ return;
+ }
+
+ if ((curr & PRIORITY_UNLOAD_SCHEDULED) != 0) {
+ if (this.dataUnloadTask != null) {
+ this.dataUnloadTask.setPriority(priority);
+ }
+ // no return - we need to propagate priority
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority | (curr & PRIORITY_FLAGS)))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed
+ return;
+ }
+
+ if ((curr & PRIORITY_LOAD_SCHEDULED) != 0) {
+ RegionFileIOThread.raisePriority(this.world, this.chunkX, this.chunkZ, this.type, priority);
+ return;
+ }
+
+ if ((curr & PRIORITY_UNLOAD_SCHEDULED) != 0) {
+ if (this.dataUnloadTask != null) {
+ this.dataUnloadTask.raisePriority(priority);
+ }
+ // no return - we need to propagate priority
+ }
+
+ if (!priority.isLowerPriority(curr & ~PRIORITY_FLAGS)) {
+ return;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority | (curr & PRIORITY_FLAGS)))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public void cancel() {
+ if ((this.getAndOrPriorityVolatile(PRIORITY_EXECUTED) & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed already
+ return;
+ }
+
+ // OK if we miss the field read, the task cannot complete if the cancelled bit is set and
+ // the write to dataLoadTask will check for the cancelled bit
+ if (this.dataUnloadCancellable != null) {
+ this.dataUnloadCancellable.cancel();
+ }
+
+ if (this.dataLoadTask != null) {
+ this.dataLoadTask.cancel();
+ }
+
+ this.complete(CANCELLED_DATA, null);
+ }
+
+ private final AtomicBoolean scheduled = new AtomicBoolean();
+
+ public void schedule() {
+ if (this.scheduled.getAndSet(true)) {
+ throw new IllegalStateException("schedule() called twice");
+ }
+ int priority = this.getPriorityVolatile();
+
+ if ((priority & PRIORITY_EXECUTED) != 0) {
+ // cancelled
+ return;
+ }
+
+ final BiConsumer<CompoundTag, Throwable> consumer = (final CompoundTag data, final Throwable thr) -> {
+ // because cancelScheduled() cannot actually stop this task from executing in every case, we need
+ // to mark complete here to ensure we do not double complete
+ if (LoadDataFromDiskTask.this.markExecuting()) {
+ LoadDataFromDiskTask.this.complete(data, thr);
+ } // else: cancelled
+ };
+
+ final PrioritisedExecutor.Priority initialPriority = PrioritisedExecutor.Priority.getPriority(priority);
+ boolean scheduledUnload = false;
+
+ final NewChunkHolder holder = this.world.chunkTaskScheduler.chunkHolderManager.getChunkHolder(this.chunkX, this.chunkZ);
+ if (holder != null) {
+ final BiConsumer<CompoundTag, Throwable> unloadConsumer = (final CompoundTag data, final Throwable thr) -> {
+ if (data != null) {
+ consumer.accept(data, null);
+ } else {
+ // need to schedule task
+ LoadDataFromDiskTask.this.schedule(false, consumer, PrioritisedExecutor.Priority.getPriority(LoadDataFromDiskTask.this.getPriorityVolatile() & ~PRIORITY_FLAGS));
+ }
+ };
+ Cancellable unloadCancellable = null;
+ CompoundTag syncComplete = null;
+ final NewChunkHolder.UnloadTask unloadTask = holder.getUnloadTask(this.type); // can be null if no task exists
+ final Completable<CompoundTag> unloadCompletable = unloadTask == null ? null : unloadTask.completable();
+ if (unloadCompletable != null) {
+ unloadCancellable = unloadCompletable.addAsynchronousWaiter(unloadConsumer);
+ if (unloadCancellable == null) {
+ syncComplete = unloadCompletable.getResult();
+ }
+ }
+
+ if (syncComplete != null) {
+ consumer.accept(syncComplete, null);
+ return;
+ }
+
+ if (unloadCancellable != null) {
+ scheduledUnload = true;
+ this.dataUnloadCancellable = unloadCancellable;
+ this.dataUnloadTask = unloadTask.task();
+ }
+ }
+
+ this.schedule(scheduledUnload, consumer, initialPriority);
+ }
+
+ private void schedule(final boolean scheduledUnload, final BiConsumer<CompoundTag, Throwable> consumer, final PrioritisedExecutor.Priority initialPriority) {
+ int priority = this.getPriorityVolatile();
+
+ if ((priority & PRIORITY_EXECUTED) != 0) {
+ // cancelled
+ return;
+ }
+
+ if (!scheduledUnload) {
+ this.dataLoadTask = RegionFileIOThread.loadDataAsync(
+ this.world, this.chunkX, this.chunkZ, this.type, consumer,
+ initialPriority.isHigherPriority(PrioritisedExecutor.Priority.NORMAL), initialPriority
+ );
+ }
+
+ int failures = 0;
+ for (;;) {
+ if (priority == (priority = this.compareAndExchangePriorityVolatile(priority, priority | (scheduledUnload ? PRIORITY_UNLOAD_SCHEDULED : PRIORITY_LOAD_SCHEDULED)))) {
+ return;
+ }
+
+ if ((priority & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed
+ if (this.dataUnloadCancellable != null) {
+ this.dataUnloadCancellable.cancel();
+ }
+
+ if (this.dataLoadTask != null) {
+ this.dataLoadTask.cancel();
+ }
+ return;
+ }
+
+ if (scheduledUnload) {
+ if (this.dataUnloadTask != null) {
+ this.dataUnloadTask.setPriority(PrioritisedExecutor.Priority.getPriority(priority & ~PRIORITY_FLAGS));
+ }
+ } else {
+ RegionFileIOThread.setPriority(this.world, this.chunkX, this.chunkZ, this.type, PrioritisedExecutor.Priority.getPriority(priority & ~PRIORITY_FLAGS));
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ /*
+ private static final class LoadDataPriorityHolder extends PriorityHolder {
+
+ protected final LoadDataFromDiskTask task;
+
+ protected LoadDataPriorityHolder(final PrioritisedExecutor.Priority priority, final LoadDataFromDiskTask task) {
+ super(priority);
+ this.task = task;
+ }
+
+ @Override
+ protected void cancelScheduled() {
+ final Cancellable dataLoadTask = this.task.dataLoadTask;
+ if (dataLoadTask != null) {
+ // OK if we miss the field read, the task cannot complete if the cancelled bit is set and
+ // the write to dataLoadTask will check for the cancelled bit
+ this.task.dataLoadTask.cancel();
+ }
+ this.task.complete(CANCELLED_DATA, null);
+ }
+
+ @Override
+ protected PrioritisedExecutor.Priority getScheduledPriority() {
+ final LoadDataFromDiskTask task = this.task;
+ return RegionFileIOThread.getPriority(task.world, task.chunkX, task.chunkZ, task.type);
+ }
+
+ @Override
+ protected void scheduleTask(final PrioritisedExecutor.Priority priority) {
+ final LoadDataFromDiskTask task = this.task;
+ final BiConsumer<CompoundTag, Throwable> consumer = (final CompoundTag data, final Throwable thr) -> {
+ // because cancelScheduled() cannot actually stop this task from executing in every case, we need
+ // to mark complete here to ensure we do not double complete
+ if (LoadDataPriorityHolder.this.markExecuting()) {
+ LoadDataPriorityHolder.this.task.complete(data, thr);
+ } // else: cancelled
+ };
+ task.dataLoadTask = RegionFileIOThread.loadDataAsync(
+ task.world, task.chunkX, task.chunkZ, task.type, consumer,
+ priority.isHigherPriority(PrioritisedExecutor.Priority.NORMAL), priority
+ );
+ if (this.isMarkedExecuted()) {
+ // if we are marked as completed, it could be:
+ // 1. we were cancelled
+ // 2. the consumer was completed
+ // in the 2nd case, cancel() does nothing
+ // in the 1st case, we ensure cancel() is called as it is possible for the cancelling thread
+ // to miss the field write here
+ task.dataLoadTask.cancel();
+ }
+ }
+
+ @Override
+ protected void lowerPriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final LoadDataFromDiskTask task = this.task;
+ RegionFileIOThread.lowerPriority(task.world, task.chunkX, task.chunkZ, task.type, priority);
+ }
+
+ @Override
+ protected void setPriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final LoadDataFromDiskTask task = this.task;
+ RegionFileIOThread.setPriority(task.world, task.chunkX, task.chunkZ, task.type, priority);
+ }
+
+ @Override
+ protected void raisePriorityScheduled(final PrioritisedExecutor.Priority priority) {
+ final LoadDataFromDiskTask task = this.task;
+ RegionFileIOThread.raisePriority(task.world, task.chunkX, task.chunkZ, task.type, priority);
+ }
+ }
+ */
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/NewChunkHolder.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/NewChunkHolder.java
new file mode 100644
index 0000000000000000000000000000000000000000..8013dd333e27aa5fd0beb431fa32491eec9f5246
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/NewChunkHolder.java
@@ -0,0 +1,2077 @@
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.completable.Completable;
+import ca.spottedleaf.concurrentutil.executor.Cancellable;
+import ca.spottedleaf.concurrentutil.executor.standard.DelayedPrioritisedTask;
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import com.google.gson.JsonArray;
+import com.google.gson.JsonElement;
+import com.google.gson.JsonObject;
+import com.google.gson.JsonPrimitive;
+import com.mojang.logging.LogUtils;
+import io.papermc.paper.chunk.system.io.RegionFileIOThread;
+import io.papermc.paper.chunk.system.poi.PoiChunk;
+import io.papermc.paper.util.CoordinateUtils;
+import io.papermc.paper.util.TickThread;
+import io.papermc.paper.util.WorldUtil;
+import io.papermc.paper.world.ChunkEntitySlices;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectLinkedOpenHashMap;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectMap;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectOpenHashMap;
+import it.unimi.dsi.fastutil.objects.ReferenceLinkedOpenHashSet;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.server.level.TicketType;
+import net.minecraft.world.entity.Entity;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
+import net.minecraft.world.level.chunk.ImposterProtoChunk;
+import net.minecraft.world.level.chunk.LevelChunk;
+import net.minecraft.world.level.chunk.storage.ChunkSerializer;
+import net.minecraft.world.level.chunk.storage.EntityStorage;
+import org.slf4j.Logger;
+import java.lang.invoke.VarHandle;
+import java.util.ArrayList;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.function.Consumer;
+
+public final class NewChunkHolder {
+
+ private static final Logger LOGGER = LogUtils.getClassLogger();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ public static final Thread.UncaughtExceptionHandler CHUNKSYSTEM_UNCAUGHT_EXCEPTION_HANDLER = new Thread.UncaughtExceptionHandler() {
+ @Override
+ public void uncaughtException(final Thread thread, final Throwable throwable) {
+ if (!(throwable instanceof ThreadDeath)) {
+ LOGGER.error("Uncaught exception in thread " + thread.getName(), throwable);
+ }
+ }
+ };
+
+ public final ServerLevel world;
+ public final int chunkX;
+ public final int chunkZ;
+
+ public final ChunkTaskScheduler scheduler;
+
+ // load/unload state
+
+ // chunk data state
+
+ private ChunkEntitySlices entityChunk;
+ // entity chunk that is loaded, but not yet deserialized
+ private CompoundTag pendingEntityChunk;
+
+ ChunkEntitySlices loadInEntityChunk(final boolean transientChunk) {
+ TickThread.ensureTickThread(this.world, this.chunkX, this.chunkZ, "Cannot sync load entity data off-main");
+ final CompoundTag entityChunk;
+ final ChunkEntitySlices ret;
+ this.scheduler.schedulingLock.lock();
+ try {
+ if (this.entityChunk != null && (transientChunk || !this.entityChunk.isTransient())) {
+ return this.entityChunk;
+ }
+ final CompoundTag pendingEntityChunk = this.pendingEntityChunk;
+ if (!transientChunk && pendingEntityChunk == null) {
+ throw new IllegalStateException("Must load entity data from disk before loading in the entity chunk!");
+ }
+
+ if (this.entityChunk == null) {
+ ret = this.entityChunk = new ChunkEntitySlices(
+ this.world, this.chunkX, this.chunkZ, this.getChunkStatus(),
+ WorldUtil.getMinSection(this.world), WorldUtil.getMaxSection(this.world)
+ );
+
+ ret.setTransient(transientChunk);
+
+ this.world.getEntityLookup().entitySectionLoad(this.chunkX, this.chunkZ, ret);
+ } else {
+ // transientChunk = false here
+ ret = this.entityChunk;
+ this.entityChunk.setTransient(false);
+ }
+
+ if (!transientChunk) {
+ this.pendingEntityChunk = null;
+ entityChunk = pendingEntityChunk == EMPTY_ENTITY_CHUNK ? null : pendingEntityChunk;
+ } else {
+ entityChunk = null;
+ }
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+
+ if (!transientChunk) {
+ if (entityChunk != null) {
+ final List<Entity> entities = EntityStorage.readEntities(this.world, entityChunk);
+
+ this.world.getEntityLookup().addEntityChunkEntities(entities);
+ }
+ }
+
+ return ret;
+ }
+
+ // needed to distinguish whether the entity chunk has been read from disk but is empty or whether it has _not_
+ // been read from disk
+ private static final CompoundTag EMPTY_ENTITY_CHUNK = new CompoundTag();
+
+ private ChunkLoadTask.EntityDataLoadTask entityDataLoadTask;
+ // note: if entityDataLoadTask is cancelled, but on its completion entityDataLoadTaskWaiters.size() != 0,
+ // then the task is rescheduled
+ private List<GenericDataLoadTaskCallback> entityDataLoadTaskWaiters;
+
+ public ChunkLoadTask.EntityDataLoadTask getEntityDataLoadTask() {
+ return this.entityDataLoadTask;
+ }
+
+ // must hold schedule lock for the two below functions
+
+ // returns only if the data has been loaded from disk, DOES NOT relate to whether it has been deserialized
+ // or added into the world (or even into entityChunk)
+ public boolean isEntityChunkNBTLoaded() {
+ return (this.entityChunk != null && !this.entityChunk.isTransient()) || this.pendingEntityChunk != null;
+ }
+
+ private void completeEntityLoad(final GenericDataLoadTask.TaskResult<CompoundTag, Throwable> result) {
+ final List<GenericDataLoadTaskCallback> completeWaiters;
+ ChunkLoadTask.EntityDataLoadTask entityDataLoadTask = null;
+ boolean scheduleEntityTask = false;
+ this.scheduler.schedulingLock.lock();
+ try {
+ final List<GenericDataLoadTaskCallback> waiters = this.entityDataLoadTaskWaiters;
+ this.entityDataLoadTask = null;
+ if (result != null) {
+ this.entityDataLoadTaskWaiters = null;
+ this.pendingEntityChunk = result.left() == null ? EMPTY_ENTITY_CHUNK : result.left();
+ if (result.right() != null) {
+ LOGGER.error("Unhandled entity data load exception, data data will be lost: ", result.right());
+ }
+
+ completeWaiters = waiters;
+ } else {
+ // cancelled
+ completeWaiters = null;
+
+ // need to re-schedule?
+ if (waiters.isEmpty()) {
+ this.entityDataLoadTaskWaiters = null;
+ // no tasks to schedule _for_
+ } else {
+ entityDataLoadTask = this.entityDataLoadTask = new ChunkLoadTask.EntityDataLoadTask(
+ this.scheduler, this.world, this.chunkX, this.chunkZ, this.getEffectivePriority()
+ );
+ entityDataLoadTask.addCallback(this::completeEntityLoad);
+ // need one schedule() per waiter
+ for (final GenericDataLoadTaskCallback callback : waiters) {
+ scheduleEntityTask |= entityDataLoadTask.schedule(true);
+ }
+ }
+ }
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+
+ if (scheduleEntityTask) {
+ entityDataLoadTask.scheduleNow();
+ }
+
+ // avoid holding the scheduling lock while completing
+ if (completeWaiters != null) {
+ for (final GenericDataLoadTaskCallback callback : completeWaiters) {
+ callback.accept(result);
+ }
+ }
+
+ this.scheduler.schedulingLock.lock();
+ try {
+ this.checkUnload();
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ }
+
+ // note: it is guaranteed that the consumer cannot be called for the entirety that the schedule lock is held
+ // however, when the consumer is invoked, it will hold the schedule lock
+ public GenericDataLoadTaskCallback getOrLoadEntityData(final Consumer<GenericDataLoadTask.TaskResult<CompoundTag, Throwable>> consumer) {
+ if (this.isEntityChunkNBTLoaded()) {
+ throw new IllegalStateException("Cannot load entity data, it is already loaded");
+ }
+ // why not just acquire the lock? because the caller NEEDS to call isEntityChunkNBTLoaded before this!
+ if (!this.scheduler.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Must hold scheduling lock");
+ }
+
+ final GenericDataLoadTaskCallback ret = new EntityDataLoadTaskCallback((Consumer)consumer, this);
+
+ if (this.entityDataLoadTask == null) {
+ this.entityDataLoadTask = new ChunkLoadTask.EntityDataLoadTask(
+ this.scheduler, this.world, this.chunkX, this.chunkZ, this.getEffectivePriority()
+ );
+ this.entityDataLoadTask.addCallback(this::completeEntityLoad);
+ this.entityDataLoadTaskWaiters = new ArrayList<>();
+ }
+ this.entityDataLoadTaskWaiters.add(ret);
+ if (this.entityDataLoadTask.schedule(true)) {
+ ret.schedule = this.entityDataLoadTask;
+ }
+ this.checkUnload();
+
+ return ret;
+ }
+
+ private static final class EntityDataLoadTaskCallback extends GenericDataLoadTaskCallback {
+
+ public EntityDataLoadTaskCallback(final Consumer<GenericDataLoadTask.TaskResult<?, Throwable>> consumer, final NewChunkHolder chunkHolder) {
+ super(consumer, chunkHolder);
+ }
+
+ @Override
+ void internalCancel() {
+ this.chunkHolder.entityDataLoadTaskWaiters.remove(this);
+ this.chunkHolder.entityDataLoadTask.cancel();
+ }
+ }
+
+ private PoiChunk poiChunk;
+
+ private ChunkLoadTask.PoiDataLoadTask poiDataLoadTask;
+ // note: if entityDataLoadTask is cancelled, but on its completion entityDataLoadTaskWaiters.size() != 0,
+ // then the task is rescheduled
+ private List<GenericDataLoadTaskCallback> poiDataLoadTaskWaiters;
+
+ public ChunkLoadTask.PoiDataLoadTask getPoiDataLoadTask() {
+ return this.poiDataLoadTask;
+ }
+
+ // must hold schedule lock for the two below functions
+
+ public boolean isPoiChunkLoaded() {
+ return this.poiChunk != null;
+ }
+
+ private void completePoiLoad(final GenericDataLoadTask.TaskResult<PoiChunk, Throwable> result) {
+ final List<GenericDataLoadTaskCallback> completeWaiters;
+ ChunkLoadTask.PoiDataLoadTask poiDataLoadTask = null;
+ boolean schedulePoiTask = false;
+ this.scheduler.schedulingLock.lock();
+ try {
+ final List<GenericDataLoadTaskCallback> waiters = this.poiDataLoadTaskWaiters;
+ this.poiDataLoadTask = null;
+ if (result != null) {
+ this.poiDataLoadTaskWaiters = null;
+ this.poiChunk = result.left();
+ if (result.right() != null) {
+ LOGGER.error("Unhandled poi load exception, poi data will be lost: ", result.right());
+ }
+
+ completeWaiters = waiters;
+ } else {
+ // cancelled
+ completeWaiters = null;
+
+ // need to re-schedule?
+ if (waiters.isEmpty()) {
+ this.poiDataLoadTaskWaiters = null;
+ // no tasks to schedule _for_
+ } else {
+ poiDataLoadTask = this.poiDataLoadTask = new ChunkLoadTask.PoiDataLoadTask(
+ this.scheduler, this.world, this.chunkX, this.chunkZ, this.getEffectivePriority()
+ );
+ poiDataLoadTask.addCallback(this::completePoiLoad);
+ // need one schedule() per waiter
+ for (final GenericDataLoadTaskCallback callback : waiters) {
+ schedulePoiTask |= poiDataLoadTask.schedule(true);
+ }
+ }
+ }
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+
+ if (schedulePoiTask) {
+ poiDataLoadTask.scheduleNow();
+ }
+
+ // avoid holding the scheduling lock while completing
+ if (completeWaiters != null) {
+ for (final GenericDataLoadTaskCallback callback : completeWaiters) {
+ callback.accept(result);
+ }
+ }
+ this.scheduler.schedulingLock.lock();
+ try {
+ this.checkUnload();
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ }
+
+ // note: it is guaranteed that the consumer cannot be called for the entirety that the schedule lock is held
+ // however, when the consumer is invoked, it will hold the schedule lock
+ public GenericDataLoadTaskCallback getOrLoadPoiData(final Consumer<GenericDataLoadTask.TaskResult<PoiChunk, Throwable>> consumer) {
+ if (this.isPoiChunkLoaded()) {
+ throw new IllegalStateException("Cannot load poi data, it is already loaded");
+ }
+ // why not just acquire the lock? because the caller NEEDS to call isPoiChunkLoaded before this!
+ if (!this.scheduler.schedulingLock.isHeldByCurrentThread()) {
+ throw new IllegalStateException("Must hold scheduling lock");
+ }
+
+ final GenericDataLoadTaskCallback ret = new PoiDataLoadTaskCallback((Consumer)consumer, this);
+
+ if (this.poiDataLoadTask == null) {
+ this.poiDataLoadTask = new ChunkLoadTask.PoiDataLoadTask(
+ this.scheduler, this.world, this.chunkX, this.chunkZ, this.getEffectivePriority()
+ );
+ this.poiDataLoadTask.addCallback(this::completePoiLoad);
+ this.poiDataLoadTaskWaiters = new ArrayList<>();
+ }
+ this.poiDataLoadTaskWaiters.add(ret);
+ if (this.poiDataLoadTask.schedule(true)) {
+ ret.schedule = this.poiDataLoadTask;
+ }
+ this.checkUnload();
+
+ return ret;
+ }
+
+ private static final class PoiDataLoadTaskCallback extends GenericDataLoadTaskCallback {
+
+ public PoiDataLoadTaskCallback(final Consumer<GenericDataLoadTask.TaskResult<?, Throwable>> consumer, final NewChunkHolder chunkHolder) {
+ super(consumer, chunkHolder);
+ }
+
+ @Override
+ void internalCancel() {
+ this.chunkHolder.poiDataLoadTaskWaiters.remove(this);
+ this.chunkHolder.poiDataLoadTask.cancel();
+ }
+ }
+
+ public static abstract class GenericDataLoadTaskCallback implements Cancellable, Consumer<GenericDataLoadTask.TaskResult<?, Throwable>> {
+
+ protected final Consumer<GenericDataLoadTask.TaskResult<?, Throwable>> consumer;
+ protected final NewChunkHolder chunkHolder;
+ protected boolean completed;
+ protected GenericDataLoadTask<?, ?> schedule;
+ protected final AtomicBoolean scheduled = new AtomicBoolean();
+
+ public GenericDataLoadTaskCallback(final Consumer<GenericDataLoadTask.TaskResult<?, Throwable>> consumer,
+ final NewChunkHolder chunkHolder) {
+ this.consumer = consumer;
+ this.chunkHolder = chunkHolder;
+ }
+
+ public void schedule() {
+ if (this.scheduled.getAndSet(true)) {
+ throw new IllegalStateException("Double calling schedule()");
+ }
+ if (this.schedule != null) {
+ this.schedule.scheduleNow();
+ this.schedule = null;
+ }
+ }
+
+ boolean isCompleted() {
+ return this.completed;
+ }
+
+ // must hold scheduling lock
+ private boolean setCompleted() {
+ if (this.completed) {
+ return false;
+ }
+ return this.completed = true;
+ }
+
+ @Override
+ public void accept(final GenericDataLoadTask.TaskResult<?, Throwable> result) {
+ if (result != null) {
+ if (this.setCompleted()) {
+ this.consumer.accept(result);
+ } else {
+ throw new IllegalStateException("Cannot be cancelled at this point");
+ }
+ } else {
+ throw new NullPointerException("Result cannot be null (cancelled)");
+ }
+ }
+
+ // holds scheduling lock
+ abstract void internalCancel();
+
+ @Override
+ public boolean cancel() {
+ this.chunkHolder.scheduler.schedulingLock.lock();
+ try {
+ if (!this.completed) {
+ this.completed = true;
+ this.internalCancel();
+ return true;
+ }
+ return false;
+ } finally {
+ this.chunkHolder.scheduler.schedulingLock.unlock();
+ }
+ }
+ }
+
+ private ChunkAccess currentChunk;
+
+ // generation status state
+
+ /**
+ * Current status the chunk has been brought up to by the chunk system. null indicates no work at all
+ */
+ private ChunkStatus currentGenStatus;
+
+ // This allows unsynchronised access to the chunk and last gen status
+ private volatile ChunkCompletion lastChunkCompletion;
+
+ public ChunkCompletion getLastChunkCompletion() {
+ return this.lastChunkCompletion;
+ }
+
+ public static final record ChunkCompletion(ChunkAccess chunk, ChunkStatus genStatus) {};
+
+ /**
+ * The target final chunk status the chunk system will bring the chunk to.
+ */
+ private ChunkStatus requestedGenStatus;
+
+ private ChunkProgressionTask generationTask;
+ private ChunkStatus generationTaskStatus;
+
+ /**
+ * contains the neighbours that this chunk generation is blocking on
+ */
+ protected final ReferenceLinkedOpenHashSet<NewChunkHolder> neighboursBlockingGenTask = new ReferenceLinkedOpenHashSet<>(4);
+
+ /**
+ * map of ChunkHolder -> Required Status for this chunk
+ */
+ protected final Reference2ObjectLinkedOpenHashMap<NewChunkHolder, ChunkStatus> neighboursWaitingForUs = new Reference2ObjectLinkedOpenHashMap<>();
+
+ public void addGenerationBlockingNeighbour(final NewChunkHolder neighbour) {
+ this.neighboursBlockingGenTask.add(neighbour);
+ }
+
+ public void addWaitingNeighbour(final NewChunkHolder neighbour, final ChunkStatus requiredStatus) {
+ final boolean wasEmpty = this.neighboursWaitingForUs.isEmpty();
+ this.neighboursWaitingForUs.put(neighbour, requiredStatus);
+ if (wasEmpty) {
+ this.checkUnload();
+ }
+ }
+
+ // priority state
+
+ // the target priority for this chunk to generate at
+ // TODO this will screw over scheduling at lower priorities to neighbours, fix
+ private PrioritisedExecutor.Priority priority = PrioritisedExecutor.Priority.NORMAL;
+ private boolean priorityLocked;
+
+ // the priority neighbouring chunks have requested this chunk generate at
+ private PrioritisedExecutor.Priority neighbourRequestedPriority = PrioritisedExecutor.Priority.IDLE;
+
+ public PrioritisedExecutor.Priority getEffectivePriority() {
+ return PrioritisedExecutor.Priority.max(this.priority, this.neighbourRequestedPriority);
+ }
+
+ protected void recalculateNeighbourRequestedPriority() {
+ if (this.neighboursWaitingForUs.isEmpty()) {
+ this.neighbourRequestedPriority = PrioritisedExecutor.Priority.IDLE;
+ return;
+ }
+
+ PrioritisedExecutor.Priority max = PrioritisedExecutor.Priority.IDLE;
+
+ for (final NewChunkHolder holder : this.neighboursWaitingForUs.keySet()) {
+ final PrioritisedExecutor.Priority neighbourPriority = holder.getEffectivePriority();
+ if (neighbourPriority.isHigherPriority(max)) {
+ max = neighbourPriority;
+ }
+ }
+
+ final PrioritisedExecutor.Priority current = this.getEffectivePriority();
+ this.neighbourRequestedPriority = max;
+ final PrioritisedExecutor.Priority next = this.getEffectivePriority();
+
+ if (current == next) {
+ return;
+ }
+
+ // our effective priority has changed, so change our task
+ if (this.generationTask != null) {
+ this.generationTask.setPriority(next);
+ }
+
+ // now propagate this to our neighbours
+ this.recalculateNeighbourPriorities();
+ }
+
+ public void recalculateNeighbourPriorities() {
+ for (final NewChunkHolder holder : this.neighboursBlockingGenTask) {
+ holder.recalculateNeighbourRequestedPriority();
+ }
+ }
+
+ // must hold scheduling lock
+ public void raisePriority(final PrioritisedExecutor.Priority priority) {
+ if (this.priority != null && this.priority.isHigherOrEqualPriority(priority)) {
+ return;
+ }
+ this.setPriority(priority);
+ }
+
+ private void lockPriority() {
+ this.priority = PrioritisedExecutor.Priority.NORMAL;
+ this.priorityLocked = true;
+ }
+
+ // must hold scheduling lock
+ public void setPriority(final PrioritisedExecutor.Priority priority) {
+ if (this.priorityLocked) {
+ return;
+ }
+ final PrioritisedExecutor.Priority old = this.getEffectivePriority();
+ this.priority = priority;
+ final PrioritisedExecutor.Priority newPriority = this.getEffectivePriority();
+
+ if (old != newPriority) {
+ if (this.generationTask != null) {
+ this.generationTask.setPriority(newPriority);
+ }
+ }
+
+ this.recalculateNeighbourPriorities();
+ }
+
+ // must hold scheduling lock
+ public void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ if (this.priority != null && this.priority.isLowerOrEqualPriority(priority)) {
+ return;
+ }
+ this.setPriority(priority);
+ }
+
+ // error handling state
+ private ChunkStatus failedGenStatus;
+ private Throwable genTaskException;
+ private Thread genTaskFailedThread;
+
+ private boolean failedLightUpdate;
+
+ public void failedLightUpdate() {
+ this.failedLightUpdate = true;
+ }
+
+ public boolean hasFailedGeneration() {
+ return this.genTaskException != null;
+ }
+
+ // ticket level state
+ private int oldTicketLevel = ChunkMap.MAX_CHUNK_DISTANCE + 1;
+ private int currentTicketLevel = ChunkMap.MAX_CHUNK_DISTANCE + 1;
+
+ public int getTicketLevel() {
+ return this.currentTicketLevel;
+ }
+
+ public final ChunkHolder vanillaChunkHolder;
+
+ public NewChunkHolder(final ServerLevel world, final int chunkX, final int chunkZ, final ChunkTaskScheduler scheduler) {
+ this.world = world;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.scheduler = scheduler;
+ this.vanillaChunkHolder = new ChunkHolder(new ChunkPos(chunkX, chunkZ), world, world.getLightEngine(), world.chunkSource.chunkMap, this);
+ }
+
+ protected ImposterProtoChunk wrappedChunkForNeighbour;
+
+ // holds scheduling lock
+ public ChunkAccess getChunkForNeighbourAccess() {
+ // Vanilla overrides the status futures with an imposter chunk to prevent writes to full chunks
+ // But we don't store per-status futures, so we need this hack
+ if (this.wrappedChunkForNeighbour != null) {
+ return this.wrappedChunkForNeighbour;
+ }
+ final ChunkAccess ret = this.currentChunk;
+ return ret instanceof LevelChunk fullChunk ? this.wrappedChunkForNeighbour = new ImposterProtoChunk(fullChunk, false) : ret;
+ }
+
+ public ChunkAccess getCurrentChunk() {
+ return this.currentChunk;
+ }
+
+ int getCurrentTicketLevel() {
+ return this.currentTicketLevel;
+ }
+
+ void updateTicketLevel(final int toLevel) {
+ this.currentTicketLevel = toLevel;
+ }
+
+ private int totalNeighboursUsingThisChunk = 0;
+
+ // holds schedule lock
+ public void addNeighbourUsingChunk() {
+ final int now = ++this.totalNeighboursUsingThisChunk;
+
+ if (now == 1) {
+ this.checkUnload();
+ }
+ }
+
+ // holds schedule lock
+ public void removeNeighbourUsingChunk() {
+ final int now = --this.totalNeighboursUsingThisChunk;
+
+ if (now == 0) {
+ this.checkUnload();
+ }
+
+ if (now < 0) {
+ throw new IllegalStateException("Neighbours using this chunk cannot be negative");
+ }
+ }
+
+ // must hold scheduling lock
+ // returns string reason for why chunk should remain loaded, null otherwise
+ public final String isSafeToUnload() {
+ // is ticket level below threshold?
+ if (this.oldTicketLevel <= ChunkHolderManager.MAX_TICKET_LEVEL) {
+ return "ticket_level";
+ }
+
+ // are we being used by another chunk for generation?
+ if (this.totalNeighboursUsingThisChunk != 0) {
+ return "neighbours_generating";
+ }
+
+ // are we going to be used by another chunk for generation?
+ if (!this.neighboursWaitingForUs.isEmpty()) {
+ return "neighbours_waiting";
+ }
+
+ // chunk must be marked inaccessible (i.e unloaded to plugins)
+ if (this.getChunkStatus() != ChunkHolder.FullChunkStatus.INACCESSIBLE) {
+ return "fullchunkstatus";
+ }
+
+ // are we currently generating anything, or have requested generation?
+ if (this.generationTask != null) {
+ return "generating";
+ }
+ if (this.requestedGenStatus != null) {
+ return "requested_generation";
+ }
+
+ // entity data requested?
+ if (this.entityDataLoadTask != null) {
+ return "entity_data_requested";
+ }
+
+ // poi data requested?
+ if (this.poiDataLoadTask != null) {
+ return "poi_data_requested";
+ }
+
+ // are we pending serialization?
+ if (this.entityDataUnload != null) {
+ return "entity_serialization";
+ }
+ if (this.poiDataUnload != null) {
+ return "poi_serialization";
+ }
+ if (this.chunkDataUnload != null) {
+ return "chunk_serialization";
+ }
+
+ // Note: light tasks do not need a check, as they add a ticket.
+
+ // nothing is using this chunk, so it should be unloaded
+ return null;
+ }
+
+ /** Unloaded from chunk map */
+ boolean killed;
+
+ // must hold scheduling lock
+ private void checkUnload() {
+ if (this.killed) {
+ return;
+ }
+ if (this.isSafeToUnload() == null) {
+ // ensure in unload queue
+ this.scheduler.chunkHolderManager.unloadQueue.add(this);
+ } else {
+ // ensure not in unload queue
+ this.scheduler.chunkHolderManager.unloadQueue.remove(this);
+ }
+ }
+
+ static final record UnloadState(NewChunkHolder holder, ChunkAccess chunk, ChunkEntitySlices entityChunk, PoiChunk poiChunk) {};
+
+ // note: these are completed with null to indicate that no write occurred
+ // they are also completed with null to indicate a null write occurred
+ private UnloadTask chunkDataUnload;
+ private UnloadTask entityDataUnload;
+ private UnloadTask poiDataUnload;
+
+ public static final record UnloadTask(Completable<CompoundTag> completable, DelayedPrioritisedTask task) {}
+
+ public UnloadTask getUnloadTask(final RegionFileIOThread.RegionFileType type) {
+ switch (type) {
+ case CHUNK_DATA:
+ return this.chunkDataUnload;
+ case ENTITY_DATA:
+ return this.entityDataUnload;
+ case POI_DATA:
+ return this.poiDataUnload;
+ default:
+ throw new IllegalStateException("Unknown regionfile type " + type);
+ }
+ }
+
+ private UnloadState unloadState;
+
+ // holds schedule lock
+ UnloadState unloadStage1() {
+ // because we hold the scheduling lock, we cannot actually unload anything
+ // so we need to null this chunk's state
+ ChunkAccess chunk = this.currentChunk;
+ ChunkEntitySlices entityChunk = this.entityChunk;
+ PoiChunk poiChunk = this.poiChunk;
+ // chunk state
+ this.currentChunk = null;
+ this.currentGenStatus = null;
+ this.wrappedChunkForNeighbour = null;
+ this.lastChunkCompletion = null;
+ // entity chunk state
+ this.entityChunk = null;
+ this.pendingEntityChunk = null;
+
+ // poi chunk state
+ this.poiChunk = null;
+
+ // priority state
+ this.priorityLocked = false;
+
+ if (chunk != null) {
+ this.chunkDataUnload = new UnloadTask(new Completable<>(), new DelayedPrioritisedTask(PrioritisedExecutor.Priority.NORMAL));
+ }
+ if (poiChunk != null) {
+ this.poiDataUnload = new UnloadTask(new Completable<>(), null);
+ }
+ if (entityChunk != null) {
+ this.entityDataUnload = new UnloadTask(new Completable<>(), null);
+ }
+
+ return this.unloadState = (chunk != null || entityChunk != null || poiChunk != null) ? new UnloadState(this, chunk, entityChunk, poiChunk) : null;
+ }
+
+ // data is null if failed or does not need to be saved
+ void completeAsyncChunkDataSave(final CompoundTag data) {
+ if (data != null) {
+ RegionFileIOThread.scheduleSave(this.world, this.chunkX, this.chunkZ, data, RegionFileIOThread.RegionFileType.CHUNK_DATA);
+ }
+ this.chunkDataUnload.completable().complete(data);
+ this.scheduler.schedulingLock.lock();
+ try {
+ // can only write to these fields while holding the schedule lock
+ this.chunkDataUnload = null;
+ this.checkUnload();
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ }
+
+ void unloadStage2(final UnloadState state) {
+ this.unloadState = null;
+ final ChunkAccess chunk = state.chunk();
+ final ChunkEntitySlices entityChunk = state.entityChunk();
+ final PoiChunk poiChunk = state.poiChunk();
+
+ final boolean shouldLevelChunkNotSave = (chunk instanceof LevelChunk levelChunk && levelChunk.mustNotSave);
+
+ // unload chunk data
+ if (chunk != null) {
+ if (chunk instanceof LevelChunk levelChunk) {
+ levelChunk.setLoaded(false);
+ }
+
+ if (!shouldLevelChunkNotSave) {
+ this.saveChunk(chunk, true);
+ } else {
+ this.completeAsyncChunkDataSave(null);
+ }
+
+ if (chunk instanceof LevelChunk levelChunk) {
+ this.world.unload(levelChunk);
+ }
+ }
+
+ // unload entity data
+ if (entityChunk != null) {
+ this.saveEntities(entityChunk, true);
+ // yes this is a hack to pass the compound tag through...
+ final CompoundTag lastEntityUnload = this.lastEntityUnload;
+ this.lastEntityUnload = null;
+
+ if (entityChunk.unload()) {
+ this.scheduler.schedulingLock.lock();
+ try {
+ entityChunk.setTransient(true);
+ this.entityChunk = entityChunk;
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ } else {
+ this.world.getEntityLookup().entitySectionUnload(this.chunkX, this.chunkZ);
+ }
+ // we need to delay the callback until after determining transience, otherwise a potential loader could
+ // set entityChunk before we do
+ this.entityDataUnload.completable().complete(lastEntityUnload);
+ }
+
+ // unload poi data
+ if (poiChunk != null) {
+ if (poiChunk.isDirty() && !shouldLevelChunkNotSave) {
+ this.savePOI(poiChunk, true);
+ } else {
+ this.poiDataUnload.completable().complete(null);
+ }
+
+ if (poiChunk.isLoaded()) {
+ this.world.getPoiManager().onUnload(CoordinateUtils.getChunkKey(this.chunkX, this.chunkZ));
+ }
+ }
+ }
+
+ boolean unloadStage3() {
+ // can only write to these while holding the schedule lock, and we instantly complete them in stage2
+ this.poiDataUnload = null;
+ this.entityDataUnload = null;
+
+ // we need to check if anything has been loaded in the meantime (or if we have transient entities)
+ if (this.entityChunk != null || this.poiChunk != null || this.currentChunk != null) {
+ return false;
+ }
+
+ return this.isSafeToUnload() == null;
+ }
+
+ private void cancelGenTask() {
+ if (this.generationTask != null) {
+ this.generationTask.cancel();
+ } else {
+ // otherwise, we are blocking on neighbours, so remove them
+ if (!this.neighboursBlockingGenTask.isEmpty()) {
+ for (final NewChunkHolder neighbour : this.neighboursBlockingGenTask) {
+ if (neighbour.neighboursWaitingForUs.remove(this) == null) {
+ throw new IllegalStateException("Corrupt state");
+ }
+ if (neighbour.neighboursWaitingForUs.isEmpty()) {
+ neighbour.checkUnload();
+ }
+ }
+ this.neighboursBlockingGenTask.clear();
+ this.checkUnload();
+ }
+ }
+ }
+
+ // holds: ticket level update lock
+ // holds: schedule lock
+ public void processTicketLevelUpdate(final List<ChunkProgressionTask> scheduledTasks, final List<NewChunkHolder> changedLoadStatus) {
+ final int oldLevel = this.oldTicketLevel;
+ final int newLevel = this.currentTicketLevel;
+
+ if (oldLevel == newLevel) {
+ return;
+ }
+
+ this.oldTicketLevel = newLevel;
+
+ final ChunkHolder.FullChunkStatus oldState = ChunkHolder.getFullChunkStatus(oldLevel);
+ final ChunkHolder.FullChunkStatus newState = ChunkHolder.getFullChunkStatus(newLevel);
+ final boolean oldUnloaded = oldLevel > ChunkHolderManager.MAX_TICKET_LEVEL;
+ final boolean newUnloaded = newLevel > ChunkHolderManager.MAX_TICKET_LEVEL;
+
+ final ChunkStatus maxGenerationStatusOld = ChunkHolder.getStatus(oldLevel);
+ final ChunkStatus maxGenerationStatusNew = ChunkHolder.getStatus(newLevel);
+
+ // check for cancellations from downgrading ticket level
+ if (this.requestedGenStatus != null && !newState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && newLevel > oldLevel) {
+ // note: cancel() may invoke onChunkGenComplete synchronously here
+ if (newUnloaded) {
+ // need to cancel all tasks
+ // note: requested status must be set to null here before cancellation, to indicate to the
+ // completion logic that we do not want rescheduling to occur
+ this.requestedGenStatus = null;
+ this.cancelGenTask();
+ } else {
+ final ChunkStatus toCancel = maxGenerationStatusNew.getNextStatus();
+ final ChunkStatus currentRequestedStatus = this.requestedGenStatus;
+
+ if (currentRequestedStatus.isOrAfter(toCancel)) {
+ // we do have to cancel something here
+ // clamp requested status to the maximum
+ if (this.currentGenStatus != null && this.currentGenStatus.isOrAfter(maxGenerationStatusNew)) {
+ // already generated to status, so we must cancel
+ this.requestedGenStatus = null;
+ this.cancelGenTask();
+ } else {
+ // not generated to status, so we may have to cancel
+ // note: gen task is always 1 status above current gen status if not null
+ this.requestedGenStatus = maxGenerationStatusNew;
+ if (this.generationTaskStatus != null && this.generationTaskStatus.isOrAfter(toCancel)) {
+ // TOOD is this even possible? i don't think so
+ throw new IllegalStateException("?????");
+ }
+ }
+ }
+ }
+ }
+
+ if (newState != oldState) {
+ if (newState.isOrAfter(oldState)) {
+ // status upgrade
+ if (!oldState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && newState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ // may need to schedule full load
+ if (this.currentGenStatus != ChunkStatus.FULL) {
+ if (this.requestedGenStatus != null) {
+ this.requestedGenStatus = ChunkStatus.FULL;
+ } else {
+ this.scheduler.schedule(
+ this.chunkX, this.chunkZ, ChunkStatus.FULL, this, scheduledTasks
+ );
+ }
+ } else {
+ // now we are fully loaded
+ this.queueBorderFullStatus(true, changedLoadStatus);
+ }
+ }
+ } else {
+ // status downgrade
+ if (!newState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING) && oldState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING)) {
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.ENTITY_TICKING, null);
+ }
+
+ if (!newState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING) && oldState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING)) {
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.TICKING, null);
+ }
+
+ if (!newState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && oldState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.BORDER, null);
+ }
+ }
+ }
+
+ if (oldState != newState) {
+ if (this.onTicketUpdate(oldState, newState)) {
+ changedLoadStatus.add(this);
+ }
+ }
+
+ if (oldUnloaded != newUnloaded) {
+ this.checkUnload();
+ }
+ }
+
+ /*
+ For full chunks, vanilla just loads chunks around it up to FEATURES, 1 radius
+
+ For ticking chunks, it updates the persistent entity manager (soon to be completely nuked by EntitySliceManager, which
+ will also need to be updated but with far less implications)
+ It also shoves the scheduled block ticks into the tick scheduler
+
+ For entity ticking chunks, updates the entity manager (see above)
+ */
+
+ static final int NEIGHBOUR_RADIUS = 2;
+ private long fullNeighbourChunksLoadedBitset;
+
+ private static int getFullNeighbourIndex(final int relativeX, final int relativeZ) {
+ // index = (relativeX + NEIGHBOUR_CACHE_RADIUS) + (relativeZ + NEIGHBOUR_CACHE_RADIUS) * (NEIGHBOUR_CACHE_RADIUS * 2 + 1)
+ // optimised variant of the above by moving some of the ops to compile time
+ return relativeX + (relativeZ * (NEIGHBOUR_RADIUS * 2 + 1)) + (NEIGHBOUR_RADIUS + NEIGHBOUR_RADIUS * ((NEIGHBOUR_RADIUS * 2 + 1)));
+ }
+ public final boolean isNeighbourFullLoaded(final int relativeX, final int relativeZ) {
+ return (this.fullNeighbourChunksLoadedBitset & (1L << getFullNeighbourIndex(relativeX, relativeZ))) != 0;
+ }
+
+ // returns true if this chunk changed full status
+ public final boolean setNeighbourFullLoaded(final int relativeX, final int relativeZ) {
+ final long before = this.fullNeighbourChunksLoadedBitset;
+ final int index = getFullNeighbourIndex(relativeX, relativeZ);
+ this.fullNeighbourChunksLoadedBitset |= (1L << index);
+ return this.onNeighbourChange(before, this.fullNeighbourChunksLoadedBitset);
+ }
+
+ // returns true if this chunk changed full status
+ public final boolean setNeighbourFullUnloaded(final int relativeX, final int relativeZ) {
+ final long before = this.fullNeighbourChunksLoadedBitset;
+ final int index = getFullNeighbourIndex(relativeX, relativeZ);
+ this.fullNeighbourChunksLoadedBitset &= ~(1L << index);
+ return this.onNeighbourChange(before, this.fullNeighbourChunksLoadedBitset);
+ }
+
+ public static boolean areNeighboursFullLoaded(final long bitset, final int radius) {
+ // index = relativeX + (relativeZ * (NEIGHBOUR_CACHE_RADIUS * 2 + 1)) + (NEIGHBOUR_CACHE_RADIUS + NEIGHBOUR_CACHE_RADIUS * ((NEIGHBOUR_CACHE_RADIUS * 2 + 1)))
+ switch (radius) {
+ case 0: {
+ return (bitset & (1L << getFullNeighbourIndex(0, 0))) != 0L;
+ }
+ case 1: {
+ long mask = 0L;
+ for (int dx = -1; dx <= 1; ++dx) {
+ for (int dz = -1; dz <= 1; ++dz) {
+ mask |= (1L << getFullNeighbourIndex(dx, dz));
+ }
+ }
+ return (bitset & mask) == mask;
+ }
+ case 2: {
+ long mask = 0L;
+ for (int dx = -2; dx <= 2; ++dx) {
+ for (int dz = -2; dz <= 2; ++dz) {
+ mask |= (1L << getFullNeighbourIndex(dx, dz));
+ }
+ }
+ return (bitset & mask) == mask;
+ }
+
+ default: {
+ throw new IllegalArgumentException("Radius not recognized: " + radius);
+ }
+ }
+ }
+
+ // upper 16 bits are pending status, lower 16 bits are current status
+ private volatile long chunkStatus;
+ private static final long PENDING_STATUS_MASK = Long.MIN_VALUE >> 31;
+ private static final ChunkHolder.FullChunkStatus[] CHUNK_STATUS_BY_ID = ChunkHolder.FullChunkStatus.values();
+ private static final VarHandle CHUNK_STATUS_HANDLE = ConcurrentUtil.getVarHandle(NewChunkHolder.class, "chunkStatus", long.class);
+
+ public static ChunkHolder.FullChunkStatus getCurrentChunkStatus(final long encoded) {
+ return CHUNK_STATUS_BY_ID[(int)encoded];
+ }
+
+ public static ChunkHolder.FullChunkStatus getPendingChunkStatus(final long encoded) {
+ return CHUNK_STATUS_BY_ID[(int)(encoded >>> 32)];
+ }
+
+ public ChunkHolder.FullChunkStatus getChunkStatus() {
+ return getCurrentChunkStatus(((long)CHUNK_STATUS_HANDLE.getVolatile((NewChunkHolder)this)));
+ }
+
+ public boolean isEntityTickingReady() {
+ return this.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING);
+ }
+
+ public boolean isTickingReady() {
+ return this.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.TICKING);
+ }
+
+ public boolean isFullChunkReady() {
+ return this.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.BORDER);
+ }
+
+ private static ChunkHolder.FullChunkStatus getStatusForBitset(final long bitset) {
+ if (areNeighboursFullLoaded(bitset, 2)) {
+ return ChunkHolder.FullChunkStatus.ENTITY_TICKING;
+ } else if (areNeighboursFullLoaded(bitset, 1)) {
+ return ChunkHolder.FullChunkStatus.TICKING;
+ } else if (areNeighboursFullLoaded(bitset, 0)) {
+ return ChunkHolder.FullChunkStatus.BORDER;
+ } else {
+ return ChunkHolder.FullChunkStatus.INACCESSIBLE;
+ }
+ }
+
+ // note: only while updating ticket level, so holds ticket update lock + scheduling lock
+ protected final boolean onTicketUpdate(final ChunkHolder.FullChunkStatus oldState, final ChunkHolder.FullChunkStatus newState) {
+ if (oldState == newState) {
+ return false;
+ }
+
+ // preserve border request after full status complete, as it does not set anything in the bitset
+ ChunkHolder.FullChunkStatus byNeighbours = getStatusForBitset(this.fullNeighbourChunksLoadedBitset);
+ if (byNeighbours == ChunkHolder.FullChunkStatus.INACCESSIBLE && newState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && this.currentGenStatus == ChunkStatus.FULL) {
+ byNeighbours = ChunkHolder.FullChunkStatus.BORDER;
+ }
+
+ final ChunkHolder.FullChunkStatus toSet;
+
+ if (newState.isOrAfter(byNeighbours)) {
+ // must clamp to neighbours level, even though we have the ticket level
+ toSet = byNeighbours;
+ } else {
+ // must clamp to ticket level, even though we have the neighbours
+ toSet = newState;
+ }
+
+ long curr = (long)CHUNK_STATUS_HANDLE.getVolatile((NewChunkHolder)this);
+
+ if (curr == ((long)toSet.ordinal() | ((long)toSet.ordinal() << 32))) {
+ // nothing to do
+ return false;
+ }
+
+ int failures = 0;
+ for (;;) {
+ final long update = (curr & ~PENDING_STATUS_MASK) | ((long)toSet.ordinal() << 32);
+ if (curr == (curr = (long)CHUNK_STATUS_HANDLE.compareAndExchange((NewChunkHolder)this, curr, update))) {
+ return true;
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ protected final boolean onNeighbourChange(final long bitsetBefore, final long bitsetAfter) {
+ ChunkHolder.FullChunkStatus oldState = getStatusForBitset(bitsetBefore);
+ ChunkHolder.FullChunkStatus newState = getStatusForBitset(bitsetAfter);
+ final ChunkHolder.FullChunkStatus currStateTicketLevel = ChunkHolder.getFullChunkStatus(this.oldTicketLevel);
+ if (oldState.isOrAfter(currStateTicketLevel)) {
+ oldState = currStateTicketLevel;
+ }
+ if (newState.isOrAfter(currStateTicketLevel)) {
+ newState = currStateTicketLevel;
+ }
+ // preserve border request after full status complete, as it does not set anything in the bitset
+ if (newState == ChunkHolder.FullChunkStatus.INACCESSIBLE && currStateTicketLevel.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && this.currentGenStatus == ChunkStatus.FULL) {
+ newState = ChunkHolder.FullChunkStatus.BORDER;
+ }
+
+ if (oldState == newState) {
+ return false;
+ }
+
+ int failures = 0;
+ for (long curr = (long)CHUNK_STATUS_HANDLE.getVolatile((NewChunkHolder)this);;) {
+ final long update = (curr & ~PENDING_STATUS_MASK) | ((long)newState.ordinal() << 32);
+ if (curr == (curr = (long)CHUNK_STATUS_HANDLE.compareAndExchange((NewChunkHolder)this, curr, update))) {
+ return true;
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ private boolean queueBorderFullStatus(final boolean loaded, final List<NewChunkHolder> changedFullStatus) {
+ final ChunkHolder.FullChunkStatus toStatus = loaded ? ChunkHolder.FullChunkStatus.BORDER : ChunkHolder.FullChunkStatus.INACCESSIBLE;
+
+ int failures = 0;
+ for (long curr = (long)CHUNK_STATUS_HANDLE.getVolatile((NewChunkHolder)this);;) {
+ final ChunkHolder.FullChunkStatus currPending = getPendingChunkStatus(curr);
+ if (loaded && currPending != ChunkHolder.FullChunkStatus.INACCESSIBLE) {
+ throw new IllegalStateException("Expected " + ChunkHolder.FullChunkStatus.INACCESSIBLE + " for pending, but got " + currPending);
+ }
+
+ final long update = (curr & ~PENDING_STATUS_MASK) | ((long)toStatus.ordinal() << 32);
+ if (curr == (curr = (long)CHUNK_STATUS_HANDLE.compareAndExchange((NewChunkHolder)this, curr, update))) {
+ if ((int)(update) != (int)(update >>> 32)) {
+ changedFullStatus.add(this);
+ return true;
+ }
+ return false;
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ // only call on main thread, must hold ticket level and scheduling lock
+ private void onFullChunkLoadChange(final boolean loaded, final List<NewChunkHolder> changedFullStatus) {
+ for (int dz = -NEIGHBOUR_RADIUS; dz <= NEIGHBOUR_RADIUS; ++dz) {
+ for (int dx = -NEIGHBOUR_RADIUS; dx <= NEIGHBOUR_RADIUS; ++dx) {
+ final NewChunkHolder holder = (dx | dz) == 0 ? this : this.scheduler.chunkHolderManager.getChunkHolder(dx + this.chunkX, dz + this.chunkZ);
+ if (loaded) {
+ if (holder.setNeighbourFullLoaded(-dx, -dz)) {
+ changedFullStatus.add(holder);
+ }
+ } else {
+ if (holder != null && holder.setNeighbourFullUnloaded(-dx, -dz)) {
+ changedFullStatus.add(holder);
+ }
+ }
+ }
+ }
+ }
+
+ private ChunkHolder.FullChunkStatus updateCurrentState(final ChunkHolder.FullChunkStatus to) {
+ int failures = 0;
+ for (long curr = (long)CHUNK_STATUS_HANDLE.getVolatile((NewChunkHolder)this);;) {
+ final long update = (curr & PENDING_STATUS_MASK) | (long)to.ordinal();
+ if (curr == (curr = (long)CHUNK_STATUS_HANDLE.compareAndExchange((NewChunkHolder)this, curr, update))) {
+ return getPendingChunkStatus(curr);
+ }
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ private void changeEntityChunkStatus(final ChunkHolder.FullChunkStatus toStatus) {
+ this.world.getEntityLookup().chunkStatusChange(this.chunkX, this.chunkZ, toStatus);
+ }
+
+ private boolean processingFullStatus = false;
+
+ // only to be called on the main thread, no locks need to be held
+ public boolean handleFullStatusChange(final List<NewChunkHolder> changedFullStatus) {
+ TickThread.ensureTickThread(this.world, this.chunkX, this.chunkZ, "Cannot update full status thread off-main");
+
+ boolean ret = false;
+
+ if (this.processingFullStatus) {
+ // we cannot process updates recursively
+ return ret;
+ }
+
+ // note: use opaque reads for chunk status read since we need it to be atomic
+
+ // test if anything changed
+ final long statusCheck = (long)CHUNK_STATUS_HANDLE.getOpaque((NewChunkHolder)this);
+ if ((int)statusCheck == (int)(statusCheck >>> 32)) {
+ // nothing changed
+ return ret;
+ }
+
+ final ChunkTaskScheduler scheduler = this.scheduler;
+ final ChunkHolderManager holderManager = scheduler.chunkHolderManager;
+ final int ticketKeep;
+ final Long ticketId;
+ holderManager.ticketLock.lock();
+ try {
+ ticketKeep = this.currentTicketLevel;
+ ticketId = Long.valueOf(holderManager.getNextStatusUpgradeId());
+ holderManager.addTicketAtLevel(TicketType.STATUS_UPGRADE, this.chunkX, this.chunkZ, ticketKeep, ticketId);
+ } finally {
+ holderManager.ticketLock.unlock();
+ }
+
+ this.processingFullStatus = true;
+ try {
+ for (;;) {
+ final long currStateEncoded = (long)CHUNK_STATUS_HANDLE.getOpaque((NewChunkHolder)this);
+ final ChunkHolder.FullChunkStatus currState = getCurrentChunkStatus(currStateEncoded);
+ ChunkHolder.FullChunkStatus nextState = getPendingChunkStatus(currStateEncoded);
+ if (currState == nextState) {
+ if (nextState == ChunkHolder.FullChunkStatus.INACCESSIBLE) {
+ this.scheduler.schedulingLock.lock();
+ try {
+ this.checkUnload();
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ }
+ break;
+ }
+
+ // chunks cannot downgrade state while status is pending a change
+ final LevelChunk chunk = (LevelChunk)this.currentChunk;
+
+ // Note: we assume that only load/unload contain plugin logic
+ // plugin logic is anything stupid enough to possibly change the chunk status while it is already
+ // being changed (i.e during load it is possible it will try to set to full ticking)
+ // in order to allow this change, we also need this plugin logic to be contained strictly after all
+ // of the chunk system load callbacks are invoked
+ if (nextState.isOrAfter(currState)) {
+ // state upgrade
+ if (!currState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && nextState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.BORDER);
+ holderManager.ensureInAutosave(this);
+ chunk.pushChunkIntoLoadedMap();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.BORDER);
+ chunk.onChunkLoad(this);
+ this.onFullChunkLoadChange(true, changedFullStatus);
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.BORDER, chunk);
+ }
+
+ if (!currState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING) && nextState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING)) {
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.TICKING);
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.TICKING);
+ chunk.onChunkTicking(this);
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.TICKING, chunk);
+ }
+
+ if (!currState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING) && nextState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING)) {
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.ENTITY_TICKING);
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.ENTITY_TICKING);
+ chunk.onChunkEntityTicking(this);
+ this.completeFullStatusConsumers(ChunkHolder.FullChunkStatus.ENTITY_TICKING, chunk);
+ }
+ } else {
+ if (currState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING) && !nextState.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING)) {
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.TICKING);
+ chunk.onChunkNotEntityTicking(this);
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.TICKING);
+ }
+
+ if (currState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING) && !nextState.isOrAfter(ChunkHolder.FullChunkStatus.TICKING)) {
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.BORDER);
+ chunk.onChunkNotTicking(this);
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.BORDER);
+ }
+
+ if (currState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && !nextState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ this.onFullChunkLoadChange(false, changedFullStatus);
+ this.changeEntityChunkStatus(ChunkHolder.FullChunkStatus.INACCESSIBLE);
+ chunk.onChunkUnload(this);
+ nextState = this.updateCurrentState(ChunkHolder.FullChunkStatus.INACCESSIBLE);
+ }
+ }
+
+ ret = true;
+ }
+ } finally {
+ this.processingFullStatus = false;
+ holderManager.removeTicketAtLevel(TicketType.STATUS_UPGRADE, this.chunkX, this.chunkZ, ticketKeep, ticketId);
+ }
+
+ return ret;
+ }
+
+ // note: must hold scheduling lock
+ // rets true if the current requested gen status is not null (effectively, whether further scheduling is not needed)
+ boolean upgradeGenTarget(final ChunkStatus toStatus) {
+ if (toStatus == null) {
+ throw new NullPointerException("toStatus cannot be null");
+ }
+ if (this.requestedGenStatus == null && this.generationTask == null) {
+ return false;
+ }
+ if (this.requestedGenStatus == null || !this.requestedGenStatus.isOrAfter(toStatus)) {
+ this.requestedGenStatus = toStatus;
+ }
+ return true;
+ }
+
+ public void setGenerationTarget(final ChunkStatus toStatus) {
+ this.requestedGenStatus = toStatus;
+ }
+
+ public boolean hasGenerationTask() {
+ return this.generationTask != null;
+ }
+
+ public ChunkStatus getCurrentGenStatus() {
+ return this.currentGenStatus;
+ }
+
+ public ChunkStatus getRequestedGenStatus() {
+ return this.requestedGenStatus;
+ }
+
+ private final Reference2ObjectOpenHashMap<ChunkStatus, List<Consumer<ChunkAccess>>> statusWaiters = new Reference2ObjectOpenHashMap<>();
+
+ void addStatusConsumer(final ChunkStatus status, final Consumer<ChunkAccess> consumer) {
+ this.statusWaiters.computeIfAbsent(status, (final ChunkStatus keyInMap) -> {
+ return new ArrayList<>(4);
+ }).add(consumer);
+ }
+
+ private void completeStatusConsumers(ChunkStatus status, final ChunkAccess chunk) {
+ // need to tell future statuses to complete if cancelled
+ do {
+ this.completeStatusConsumers0(status, chunk);
+ } while (chunk == null && status != (status = status.getNextStatus()));
+ }
+
+ private void completeStatusConsumers0(final ChunkStatus status, final ChunkAccess chunk) {
+ final List<Consumer<ChunkAccess>> consumers;
+ consumers = this.statusWaiters.remove(status);
+
+ if (consumers == null) {
+ return;
+ }
+
+ // must be scheduled to main, we do not trust the callback to not do anything stupid
+ this.scheduler.scheduleChunkTask(this.chunkX, this.chunkZ, () -> {
+ for (final Consumer<ChunkAccess> consumer : consumers) {
+ try {
+ consumer.accept(chunk);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to process chunk status callback", thr);
+ }
+ }
+ }, PrioritisedExecutor.Priority.HIGHEST);
+ }
+
+ private final Reference2ObjectOpenHashMap<ChunkHolder.FullChunkStatus, List<Consumer<LevelChunk>>> fullStatusWaiters = new Reference2ObjectOpenHashMap<>();
+
+ void addFullStatusConsumer(final ChunkHolder.FullChunkStatus status, final Consumer<LevelChunk> consumer) {
+ this.fullStatusWaiters.computeIfAbsent(status, (final ChunkHolder.FullChunkStatus keyInMap) -> {
+ return new ArrayList<>(4);
+ }).add(consumer);
+ }
+
+ private void completeFullStatusConsumers(ChunkHolder.FullChunkStatus status, final LevelChunk chunk) {
+ // need to tell future statuses to complete if cancelled
+ final ChunkHolder.FullChunkStatus max = CHUNK_STATUS_BY_ID[CHUNK_STATUS_BY_ID.length - 1];
+
+ for (;;) {
+ this.completeFullStatusConsumers0(status, chunk);
+ if (chunk != null || status == max) {
+ break;
+ }
+ status = CHUNK_STATUS_BY_ID[status.ordinal() + 1];
+ }
+ }
+
+ private void completeFullStatusConsumers0(final ChunkHolder.FullChunkStatus status, final LevelChunk chunk) {
+ final List<Consumer<LevelChunk>> consumers;
+ consumers = this.fullStatusWaiters.remove(status);
+
+ if (consumers == null) {
+ return;
+ }
+
+ // must be scheduled to main, we do not trust the callback to not do anything stupid
+ this.scheduler.scheduleChunkTask(this.chunkX, this.chunkZ, () -> {
+ for (final Consumer<LevelChunk> consumer : consumers) {
+ try {
+ consumer.accept(chunk);
+ } catch (final ThreadDeath thr) {
+ throw thr;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to process chunk status callback", thr);
+ }
+ }
+ }, PrioritisedExecutor.Priority.HIGHEST);
+ }
+
+ // note: must hold scheduling lock
+ private void onChunkGenComplete(final ChunkAccess newChunk, final ChunkStatus newStatus,
+ final List<ChunkProgressionTask> scheduleList, final List<NewChunkHolder> changedLoadStatus) {
+ if (!this.neighboursBlockingGenTask.isEmpty()) {
+ throw new IllegalStateException("Cannot have neighbours blocking this gen task");
+ }
+ if (newChunk != null || (this.requestedGenStatus == null || !this.requestedGenStatus.isOrAfter(newStatus))) {
+ this.completeStatusConsumers(newStatus, newChunk);
+ }
+ // done now, clear state (must be done before scheduling new tasks)
+ this.generationTask = null;
+ this.generationTaskStatus = null;
+ if (newChunk == null) {
+ // task was cancelled
+ // should be careful as this could be called while holding the schedule lock and/or inside the
+ // ticket level update
+ // while a task may be cancelled, it is possible for it to be later re-scheduled
+ // however, because generationTask is only set to null on _completion_, the scheduler leaves
+ // the rescheduling logic to us here
+ final ChunkStatus requestedGenStatus = this.requestedGenStatus;
+ this.requestedGenStatus = null;
+ if (requestedGenStatus != null) {
+ // it looks like it has been requested, so we must reschedule
+ if (!this.neighboursWaitingForUs.isEmpty()) {
+ for (final Iterator<Reference2ObjectMap.Entry<NewChunkHolder, ChunkStatus>> iterator = this.neighboursWaitingForUs.reference2ObjectEntrySet().fastIterator(); iterator.hasNext();) {
+ final Reference2ObjectMap.Entry<NewChunkHolder, ChunkStatus> entry = iterator.next();
+
+ final NewChunkHolder chunkHolder = entry.getKey();
+ final ChunkStatus toStatus = entry.getValue();
+
+ if (!requestedGenStatus.isOrAfter(toStatus)) {
+ // if we were cancelled, we are responsible for removing the waiter
+ if (!chunkHolder.neighboursBlockingGenTask.remove(this)) {
+ throw new IllegalStateException("Corrupt state");
+ }
+ if (chunkHolder.neighboursBlockingGenTask.isEmpty()) {
+ chunkHolder.checkUnload();
+ }
+ iterator.remove();
+ continue;
+ }
+ }
+ }
+
+ // note: only after generationTask -> null, generationTaskStatus -> null, and requestedGenStatus -> null
+ this.scheduler.schedule(
+ this.chunkX, this.chunkZ, requestedGenStatus, this, scheduleList
+ );
+
+ // return, can't do anything further
+ return;
+ }
+
+ if (!this.neighboursWaitingForUs.isEmpty()) {
+ for (final NewChunkHolder chunkHolder : this.neighboursWaitingForUs.keySet()) {
+ if (!chunkHolder.neighboursBlockingGenTask.remove(this)) {
+ throw new IllegalStateException("Corrupt state");
+ }
+ if (chunkHolder.neighboursBlockingGenTask.isEmpty()) {
+ chunkHolder.checkUnload();
+ }
+ }
+ this.neighboursWaitingForUs.clear();
+ }
+ // reset priority, we have nothing left to generate to
+ this.setPriority(PrioritisedExecutor.Priority.NORMAL);
+ this.checkUnload();
+ return;
+ }
+
+ this.currentChunk = newChunk;
+ this.currentGenStatus = newStatus;
+ this.lastChunkCompletion = new ChunkCompletion(newChunk, newStatus);
+
+ final ChunkStatus requestedGenStatus = this.requestedGenStatus;
+
+ List<NewChunkHolder> needsScheduling = null;
+ boolean recalculatePriority = false;
+ for (final Iterator<Reference2ObjectMap.Entry<NewChunkHolder, ChunkStatus>> iterator
+ = this.neighboursWaitingForUs.reference2ObjectEntrySet().fastIterator(); iterator.hasNext();) {
+ final Reference2ObjectMap.Entry<NewChunkHolder, ChunkStatus> entry = iterator.next();
+ final NewChunkHolder neighbour = entry.getKey();
+ final ChunkStatus requiredStatus = entry.getValue();
+
+ if (!newStatus.isOrAfter(requiredStatus)) {
+ if (requestedGenStatus == null || !requestedGenStatus.isOrAfter(requiredStatus)) {
+ // if we're cancelled, still need to clear this map
+ if (!neighbour.neighboursBlockingGenTask.remove(this)) {
+ throw new IllegalStateException("Neighbour is not waiting for us?");
+ }
+ if (neighbour.neighboursBlockingGenTask.isEmpty()) {
+ neighbour.checkUnload();
+ }
+
+ iterator.remove();
+ }
+ continue;
+ }
+
+ // doesn't matter what isCancelled is here, we need to schedule if we can
+
+ recalculatePriority = true;
+ if (!neighbour.neighboursBlockingGenTask.remove(this)) {
+ throw new IllegalStateException("Neighbour is not waiting for us?");
+ }
+
+ if (neighbour.neighboursBlockingGenTask.isEmpty()) {
+ if (neighbour.requestedGenStatus != null) {
+ if (needsScheduling == null) {
+ needsScheduling = new ArrayList<>();
+ }
+ needsScheduling.add(neighbour);
+ } else {
+ neighbour.checkUnload();
+ }
+ }
+
+ // remove last; access to entry will throw if removed
+ iterator.remove();
+ }
+
+ if (newStatus == ChunkStatus.FULL) {
+ this.lockPriority();
+ // must use oldTicketLevel, we hold the schedule lock but not the ticket level lock
+ // however, schedule lock needs to be held for ticket level callback, so we're fine here
+ if (ChunkHolder.getFullChunkStatus(this.oldTicketLevel).isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
+ this.queueBorderFullStatus(true, changedLoadStatus);
+ }
+ }
+
+ if (recalculatePriority) {
+ this.recalculateNeighbourRequestedPriority();
+ }
+
+ if (requestedGenStatus != null && !newStatus.isOrAfter(requestedGenStatus)) {
+ this.scheduleNeighbours(needsScheduling, scheduleList);
+
+ // we need to schedule more tasks now
+ this.scheduler.schedule(
+ this.chunkX, this.chunkZ, requestedGenStatus, this, scheduleList
+ );
+ } else {
+ // we're done now
+ if (requestedGenStatus != null) {
+ this.requestedGenStatus = null;
+ }
+ // reached final stage, so stop scheduling now
+ this.setPriority(PrioritisedExecutor.Priority.NORMAL);
+ this.checkUnload();
+
+ this.scheduleNeighbours(needsScheduling, scheduleList);
+ }
+ }
+
+ private void scheduleNeighbours(final List<NewChunkHolder> needsScheduling, final List<ChunkProgressionTask> scheduleList) {
+ if (needsScheduling != null) {
+ for (int i = 0, len = needsScheduling.size(); i < len; ++i) {
+ final NewChunkHolder neighbour = needsScheduling.get(i);
+
+ this.scheduler.schedule(
+ neighbour.chunkX, neighbour.chunkZ, neighbour.requestedGenStatus, neighbour, scheduleList
+ );
+ }
+ }
+ }
+
+ public void setGenerationTask(final ChunkProgressionTask generationTask, final ChunkStatus taskStatus,
+ final List<NewChunkHolder> neighbours) {
+ if (this.generationTask != null || (this.currentGenStatus != null && this.currentGenStatus.isOrAfter(taskStatus))) {
+ throw new IllegalStateException("Currently generating or provided task is trying to generate to a level we are already at!");
+ }
+ if (this.requestedGenStatus == null || !this.requestedGenStatus.isOrAfter(taskStatus)) {
+ throw new IllegalStateException("Cannot schedule generation task when not requested");
+ }
+ this.generationTask = generationTask;
+ this.generationTaskStatus = taskStatus;
+
+ for (int i = 0, len = neighbours.size(); i < len; ++i) {
+ neighbours.get(i).addNeighbourUsingChunk();
+ }
+
+ this.checkUnload();
+
+ generationTask.onComplete((final ChunkAccess access, final Throwable thr) -> {
+ if (generationTask != this.generationTask) {
+ throw new IllegalStateException(
+ "Cannot complete generation task '" + generationTask + "' because we are waiting on '" + this.generationTask + "' instead!"
+ );
+ }
+ if (thr != null) {
+ if (this.genTaskException != null) {
+ // first one is probably the TRUE problem
+ return;
+ }
+ // don't set generation task to null, so that scheduling will not attempt to create another task and it
+ // will automatically block any further scheduling usage of this chunk as it will wait forever for a failed
+ // task to complete
+ this.genTaskException = thr;
+ this.failedGenStatus = taskStatus;
+ this.genTaskFailedThread = Thread.currentThread();
+
+ this.scheduler.unrecoverableChunkSystemFailure(this.chunkX, this.chunkZ, Map.of(
+ "Generation task", ChunkTaskScheduler.stringIfNull(generationTask),
+ "Task to status", ChunkTaskScheduler.stringIfNull(taskStatus)
+ ), thr);
+ return;
+ }
+
+ final boolean scheduleTasks;
+ List<ChunkProgressionTask> tasks = ChunkHolderManager.getCurrentTicketUpdateScheduling();
+ if (tasks == null) {
+ scheduleTasks = true;
+ tasks = new ArrayList<>();
+ } else {
+ scheduleTasks = false;
+ // we are currently updating ticket levels, so we already hold the schedule lock
+ // this means we have to leave the ticket level update to handle the scheduling
+ }
+ final List<NewChunkHolder> changedLoadStatus = new ArrayList<>();
+ this.scheduler.schedulingLock.lock();
+ try {
+ for (int i = 0, len = neighbours.size(); i < len; ++i) {
+ neighbours.get(i).removeNeighbourUsingChunk();
+ }
+ this.onChunkGenComplete(access, taskStatus, tasks, changedLoadStatus);
+ } finally {
+ this.scheduler.schedulingLock.unlock();
+ }
+ this.scheduler.chunkHolderManager.addChangedStatuses(changedLoadStatus);
+
+ if (scheduleTasks) {
+ // can't hold the lock while scheduling, so we have to build the tasks and then schedule after
+ for (int i = 0, len = tasks.size(); i < len; ++i) {
+ tasks.get(i).schedule();
+ }
+ }
+ });
+ }
+
+ public PoiChunk getPoiChunk() {
+ return this.poiChunk;
+ }
+
+ public ChunkEntitySlices getEntityChunk() {
+ return this.entityChunk;
+ }
+
+ public long lastAutoSave;
+
+ public static final record SaveStat(boolean savedChunk, boolean savedEntityChunk, boolean savedPoiChunk) {}
+
+ public SaveStat save(final boolean shutdown, final boolean unloading) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ TickThread.ensureTickThread(this.world, this.chunkX, this.chunkZ, "Cannot save data off-main");
+
+ ChunkAccess chunk = this.getCurrentChunk();
+ PoiChunk poi = this.getPoiChunk();
+ ChunkEntitySlices entities = this.getEntityChunk();
+ boolean executedUnloadTask = false;
+
+ if (shutdown) {
+ // make sure that the async unloads complete
+ if (this.unloadState != null) {
+ // must have errored during unload
+ chunk = this.unloadState.chunk();
+ poi = this.unloadState.poiChunk();
+ entities = this.unloadState.entityChunk();
+ }
+ final UnloadTask chunkUnloadTask = this.chunkDataUnload;
+ final DelayedPrioritisedTask chunkDataUnloadTask = chunkUnloadTask == null ? null : chunkUnloadTask.task();
+ if (chunkDataUnloadTask != null) {
+ final PrioritisedExecutor.PrioritisedTask unloadTask = chunkDataUnloadTask.getTask();
+ if (unloadTask != null) {
+ executedUnloadTask = unloadTask.execute();
+ }
+ }
+ }
+
+ boolean canSaveChunk = !(chunk instanceof LevelChunk levelChunk && levelChunk.mustNotSave) &&
+ (chunk != null && ((shutdown || chunk instanceof LevelChunk) && chunk.isUnsaved()));
+ boolean canSavePOI = !(chunk instanceof LevelChunk levelChunk && levelChunk.mustNotSave) && (poi != null && poi.isDirty());
+ boolean canSaveEntities = entities != null;
+
+ try (co.aikar.timings.Timing ignored = this.world.timings.chunkSave.startTiming()) { // Paper
+ if (canSaveChunk) {
+ canSaveChunk = this.saveChunk(chunk, unloading);
+ }
+ if (canSavePOI) {
+ canSavePOI = this.savePOI(poi, unloading);
+ }
+ if (canSaveEntities) {
+ // on shutdown, we need to force transient entity chunks to save
+ canSaveEntities = this.saveEntities(entities, unloading || shutdown);
+ if (unloading || shutdown) {
+ this.lastEntityUnload = null;
+ }
+ }
+ }
+
+ return executedUnloadTask | canSaveChunk | canSaveEntities | canSavePOI ? new SaveStat(executedUnloadTask || canSaveChunk, canSaveEntities, canSavePOI): null;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ static final class AsyncChunkSerializeTask implements Runnable {
+
+ private final ServerLevel world;
+ private final ChunkAccess chunk;
+ private final ChunkSerializer.AsyncSaveData asyncSaveData;
+ private final NewChunkHolder toComplete;
+
+ public AsyncChunkSerializeTask(final ServerLevel world, final ChunkAccess chunk, final ChunkSerializer.AsyncSaveData asyncSaveData,
+ final NewChunkHolder toComplete) {
+ this.world = world;
+ this.chunk = chunk;
+ this.asyncSaveData = asyncSaveData;
+ this.toComplete = toComplete;
+ }
+
+ @Override
+ public void run() {
+ final CompoundTag toSerialize;
+ try {
+ toSerialize = ChunkSerializer.saveChunk(this.world, this.chunk, this.asyncSaveData);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable throwable) {
+ LOGGER.error("Failed to asynchronously save chunk " + this.chunk.getPos() + " for world '" + this.world.getWorld().getName() + "', falling back to synchronous save", throwable);
+ this.world.chunkTaskScheduler.scheduleChunkTask(this.chunk.locX, this.chunk.locZ, () -> {
+ final CompoundTag synchronousSave;
+ try {
+ synchronousSave = ChunkSerializer.saveChunk(AsyncChunkSerializeTask.this.world, AsyncChunkSerializeTask.this.chunk, AsyncChunkSerializeTask.this.asyncSaveData);
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable throwable2) {
+ LOGGER.error("Failed to synchronously save chunk " + AsyncChunkSerializeTask.this.chunk.getPos() + " for world '" + AsyncChunkSerializeTask.this.world.getWorld().getName() + "', chunk data will be lost", throwable2);
+ AsyncChunkSerializeTask.this.toComplete.completeAsyncChunkDataSave(null);
+ return;
+ }
+
+ AsyncChunkSerializeTask.this.toComplete.completeAsyncChunkDataSave(synchronousSave);
+ LOGGER.info("Successfully serialized chunk " + AsyncChunkSerializeTask.this.chunk.getPos() + " for world '" + AsyncChunkSerializeTask.this.world.getWorld().getName() + "' synchronously");
+
+ }, PrioritisedExecutor.Priority.HIGHEST);
+ return;
+ }
+ this.toComplete.completeAsyncChunkDataSave(toSerialize);
+ }
+
+ @Override
+ public String toString() {
+ return "AsyncChunkSerializeTask{" +
+ "chunk={pos=" + this.chunk.getPos() + ",world=\"" + this.world.getWorld().getName() + "\"}" +
+ "}";
+ }
+ }
+
+ private boolean saveChunk(final ChunkAccess chunk, final boolean unloading) {
+ if (!chunk.isUnsaved()) {
+ if (unloading) {
+ this.completeAsyncChunkDataSave(null);
+ }
+ return false;
+ }
+ boolean completing = false;
+ try {
+ if (unloading) {
+ try {
+ final ChunkSerializer.AsyncSaveData asyncSaveData = ChunkSerializer.getAsyncSaveData(this.world, chunk);
+
+ final PrioritisedExecutor.PrioritisedTask task = this.scheduler.loadExecutor.createTask(new AsyncChunkSerializeTask(this.world, chunk, asyncSaveData, this));
+
+ this.chunkDataUnload.task().setTask(task);
+
+ task.queue();
+
+ chunk.setUnsaved(false);
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return true;
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to prepare async chunk data (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "', falling back to synchronous save", thr);
+ // fall through to synchronous save
+ }
+ }
+
+ final CompoundTag save = ChunkSerializer.saveChunk(this.world, chunk, null);
+
+ if (unloading) {
+ completing = true;
+ this.completeAsyncChunkDataSave(save);
+ LOGGER.info("Successfully serialized chunk data (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "' synchronously");
+ } else {
+ RegionFileIOThread.scheduleSave(this.world, this.chunkX, this.chunkZ, save, RegionFileIOThread.RegionFileType.CHUNK_DATA);
+ }
+ chunk.setUnsaved(false);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to save chunk data (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "'");
+ if (unloading && !completing) {
+ this.completeAsyncChunkDataSave(null);
+ }
+ }
+
+ return true;
+ }
+
+ private boolean lastEntitySaveNull;
+ private CompoundTag lastEntityUnload;
+ private boolean saveEntities(final ChunkEntitySlices entities, final boolean unloading) {
+ try {
+ CompoundTag mergeFrom = null;
+ if (entities.isTransient()) {
+ if (!unloading) {
+ // if we're a transient chunk, we cannot save until unloading because otherwise a double save will
+ // result in double adding the entities
+ return false;
+ }
+ try {
+ mergeFrom = RegionFileIOThread.loadData(this.world, this.chunkX, this.chunkZ, RegionFileIOThread.RegionFileType.ENTITY_DATA, PrioritisedExecutor.Priority.BLOCKING);
+ } catch (final Exception ex) {
+ LOGGER.error("Cannot merge transient entities for chunk (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "', data on disk will be replaced", ex);
+ }
+ }
+
+ final CompoundTag save = entities.save();
+ if (mergeFrom != null) {
+ if (save == null) {
+ // don't override the data on disk with nothing
+ return false;
+ } else {
+ EntityStorage.copyEntities(mergeFrom, save);
+ }
+ }
+ if (save == null && this.lastEntitySaveNull) {
+ return false;
+ }
+
+ RegionFileIOThread.scheduleSave(this.world, this.chunkX, this.chunkZ, save, RegionFileIOThread.RegionFileType.ENTITY_DATA);
+ this.lastEntitySaveNull = save == null;
+ if (unloading) {
+ this.lastEntityUnload = save;
+ }
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to save entity data (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "'");
+ }
+
+ return true;
+ }
+
+ private boolean lastPoiSaveNull;
+ private boolean savePOI(final PoiChunk poi, final boolean unloading) {
+ try {
+ final CompoundTag save = poi.save();
+ poi.setDirty(false);
+ if (save == null && this.lastPoiSaveNull) {
+ if (unloading) {
+ this.poiDataUnload.completable().complete(null);
+ }
+ return false;
+ }
+
+ RegionFileIOThread.scheduleSave(this.world, this.chunkX, this.chunkZ, save, RegionFileIOThread.RegionFileType.POI_DATA);
+ this.lastPoiSaveNull = save == null;
+ if (unloading) {
+ this.poiDataUnload.completable().complete(save);
+ }
+ } catch (final ThreadDeath death) {
+ throw death;
+ } catch (final Throwable thr) {
+ LOGGER.error("Failed to save poi data (" + this.chunkX + "," + this.chunkZ + ") in world '" + this.world.getWorld().getName() + "'");
+ }
+
+ return true;
+ }
+
+ @Override
+ public String toString() {
+ final ChunkCompletion lastCompletion = this.lastChunkCompletion;
+ final ChunkEntitySlices entityChunk = this.entityChunk;
+ final long chunkStatus = this.chunkStatus;
+ final int fullChunkStatus = (int)chunkStatus;
+ final int pendingChunkStatus = (int)(chunkStatus >>> 32);
+ final ChunkHolder.FullChunkStatus currentFullStatus = fullChunkStatus < 0 || fullChunkStatus >= CHUNK_STATUS_BY_ID.length ? null : CHUNK_STATUS_BY_ID[fullChunkStatus];
+ final ChunkHolder.FullChunkStatus pendingFullStatus = pendingChunkStatus < 0 || pendingChunkStatus >= CHUNK_STATUS_BY_ID.length ? null : CHUNK_STATUS_BY_ID[pendingChunkStatus];
+ return "NewChunkHolder{" +
+ "world=" + this.world.getWorld().getName() +
+ ", chunkX=" + this.chunkX +
+ ", chunkZ=" + this.chunkZ +
+ ", entityChunkFromDisk=" + (entityChunk != null && !entityChunk.isTransient()) +
+ ", lastChunkCompletion={chunk_class=" + (lastCompletion == null || lastCompletion.chunk() == null ? "null" : lastCompletion.chunk().getClass().getName()) + ",status=" + (lastCompletion == null ? "null" : lastCompletion.genStatus()) + "}" +
+ ", currentGenStatus=" + this.currentGenStatus +
+ ", requestedGenStatus=" + this.requestedGenStatus +
+ ", generationTask=" + this.generationTask +
+ ", generationTaskStatus=" + this.generationTaskStatus +
+ ", priority=" + this.priority +
+ ", priorityLocked=" + this.priorityLocked +
+ ", neighbourRequestedPriority=" + this.neighbourRequestedPriority +
+ ", effective_priority=" + this.getEffectivePriority() +
+ ", oldTicketLevel=" + this.oldTicketLevel +
+ ", currentTicketLevel=" + this.currentTicketLevel +
+ ", totalNeighboursUsingThisChunk=" + this.totalNeighboursUsingThisChunk +
+ ", fullNeighbourChunksLoadedBitset=" + this.fullNeighbourChunksLoadedBitset +
+ ", chunkStatusRaw=" + chunkStatus +
+ ", currentChunkStatus=" + currentFullStatus +
+ ", pendingChunkStatus=" + pendingFullStatus +
+ ", is_unload_safe=" + this.isSafeToUnload() +
+ ", killed=" + this.killed +
+ '}';
+ }
+
+ private static JsonElement serializeCompletable(final Completable<?> completable) {
+ if (completable == null) {
+ return new JsonPrimitive("null");
+ }
+
+ final JsonObject ret = new JsonObject();
+ final boolean isCompleted = completable.isCompleted();
+ ret.addProperty("completed", Boolean.valueOf(isCompleted));
+
+ if (isCompleted) {
+ ret.addProperty("completed_exceptionally", Boolean.valueOf(completable.getThrowable() != null));
+ }
+
+ return ret;
+ }
+
+ // holds ticket and scheduling lock
+ public JsonObject getDebugJson() {
+ final JsonObject ret = new JsonObject();
+
+ final ChunkCompletion lastCompletion = this.lastChunkCompletion;
+ final ChunkEntitySlices slices = this.entityChunk;
+ final PoiChunk poiChunk = this.poiChunk;
+
+ ret.addProperty("chunkX", Integer.valueOf(this.chunkX));
+ ret.addProperty("chunkZ", Integer.valueOf(this.chunkZ));
+ ret.addProperty("entity_chunk", slices == null ? "null" : "transient=" + slices.isTransient());
+ ret.addProperty("poi_chunk", "null=" + (poiChunk == null));
+ ret.addProperty("completed_chunk_class", lastCompletion == null ? "null" : lastCompletion.chunk().getClass().getName());
+ ret.addProperty("completed_gen_status", lastCompletion == null ? "null" : lastCompletion.genStatus().toString());
+ ret.addProperty("priority", Objects.toString(this.priority));
+ ret.addProperty("neighbour_requested_priority", Objects.toString(this.neighbourRequestedPriority));
+ ret.addProperty("generation_task", Objects.toString(this.generationTask));
+ ret.addProperty("is_safe_unload", Objects.toString(this.isSafeToUnload()));
+ ret.addProperty("old_ticket_level", Integer.valueOf(this.oldTicketLevel));
+ ret.addProperty("current_ticket_level", Integer.valueOf(this.currentTicketLevel));
+ ret.addProperty("neighbours_using_chunk", Integer.valueOf(this.totalNeighboursUsingThisChunk));
+
+ final JsonObject neighbourWaitState = new JsonObject();
+ ret.add("neighbour_state", neighbourWaitState);
+
+ final JsonArray blockingGenNeighbours = new JsonArray();
+ neighbourWaitState.add("blocking_gen_task", blockingGenNeighbours);
+ for (final NewChunkHolder blockingGenNeighbour : this.neighboursBlockingGenTask) {
+ final JsonObject neighbour = new JsonObject();
+ blockingGenNeighbours.add(neighbour);
+
+ neighbour.addProperty("chunkX", Integer.valueOf(blockingGenNeighbour.chunkX));
+ neighbour.addProperty("chunkZ", Integer.valueOf(blockingGenNeighbour.chunkZ));
+ }
+
+ final JsonArray neighboursWaitingForUs = new JsonArray();
+ neighbourWaitState.add("neighbours_waiting_on_us", neighboursWaitingForUs);
+ for (final Reference2ObjectMap.Entry<NewChunkHolder, ChunkStatus> entry : this.neighboursWaitingForUs.reference2ObjectEntrySet()) {
+ final NewChunkHolder holder = entry.getKey();
+ final ChunkStatus status = entry.getValue();
+
+ final JsonObject neighbour = new JsonObject();
+ neighboursWaitingForUs.add(neighbour);
+
+
+ neighbour.addProperty("chunkX", Integer.valueOf(holder.chunkX));
+ neighbour.addProperty("chunkZ", Integer.valueOf(holder.chunkZ));
+ neighbour.addProperty("waiting_for", Objects.toString(status));
+ }
+
+ ret.addProperty("fullchunkstatus", Objects.toString(this.getChunkStatus()));
+ ret.addProperty("fullchunkstatus_raw", Long.valueOf(this.chunkStatus));
+ ret.addProperty("generation_task", Objects.toString(this.generationTask));
+ ret.addProperty("requested_generation", Objects.toString(this.requestedGenStatus));
+ ret.addProperty("has_entity_load_task", Boolean.valueOf(this.entityDataLoadTask != null));
+ ret.addProperty("has_poi_load_task", Boolean.valueOf(this.poiDataLoadTask != null));
+
+ final UnloadTask entityDataUnload = this.entityDataUnload;
+ final UnloadTask poiDataUnload = this.poiDataUnload;
+ final UnloadTask chunkDataUnload = this.chunkDataUnload;
+
+ ret.add("entity_unload_completable", serializeCompletable(entityDataUnload == null ? null : entityDataUnload.completable()));
+ ret.add("poi_unload_completable", serializeCompletable(poiDataUnload == null ? null : poiDataUnload.completable()));
+ ret.add("chunk_unload_completable", serializeCompletable(chunkDataUnload == null ? null : chunkDataUnload.completable()));
+
+ final DelayedPrioritisedTask unloadTask = chunkDataUnload == null ? null : chunkDataUnload.task();
+ if (unloadTask == null) {
+ ret.addProperty("unload_task_priority", "null");
+ ret.addProperty("unload_task_priority_raw", "null");
+ } else {
+ ret.addProperty("unload_task_priority", Objects.toString(unloadTask.getPriority()));
+ ret.addProperty("unload_task_priority_raw", Integer.valueOf(unloadTask.getPriorityInternal()));
+ }
+
+ ret.addProperty("killed", Boolean.valueOf(this.killed));
+
+ return ret;
+ }
+}
diff --git a/src/main/java/io/papermc/paper/chunk/system/scheduling/PriorityHolder.java b/src/main/java/io/papermc/paper/chunk/system/scheduling/PriorityHolder.java
new file mode 100644
index 0000000000000000000000000000000000000000..b4c56bf12dc8dd17452210ece4fd67411cc6b2fd
--- /dev/null
+++ b/src/main/java/io/papermc/paper/chunk/system/scheduling/PriorityHolder.java
@@ -0,0 +1,215 @@
+package io.papermc.paper.chunk.system.scheduling;
+
+import ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor;
+import ca.spottedleaf.concurrentutil.util.ConcurrentUtil;
+import java.lang.invoke.VarHandle;
+
+public abstract class PriorityHolder {
+
+ protected volatile int priority;
+ protected static final VarHandle PRIORITY_HANDLE = ConcurrentUtil.getVarHandle(PriorityHolder.class, "priority", int.class);
+
+ protected static final int PRIORITY_SCHEDULED = Integer.MIN_VALUE >>> 0;
+ protected static final int PRIORITY_EXECUTED = Integer.MIN_VALUE >>> 1;
+
+ protected final int getPriorityVolatile() {
+ return (int)PRIORITY_HANDLE.getVolatile((PriorityHolder)this);
+ }
+
+ protected final int compareAndExchangePriorityVolatile(final int expect, final int update) {
+ return (int)PRIORITY_HANDLE.compareAndExchange((PriorityHolder)this, (int)expect, (int)update);
+ }
+
+ protected final int getAndOrPriorityVolatile(final int val) {
+ return (int)PRIORITY_HANDLE.getAndBitwiseOr((PriorityHolder)this, (int)val);
+ }
+
+ protected final void setPriorityPlain(final int val) {
+ PRIORITY_HANDLE.set((PriorityHolder)this, (int)val);
+ }
+
+ protected PriorityHolder(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+ this.setPriorityPlain(priority.priority);
+ }
+
+ // used only for debug json
+ public boolean isScheduled() {
+ return (this.getPriorityVolatile() & PRIORITY_SCHEDULED) != 0;
+ }
+
+ // returns false if cancelled
+ protected boolean markExecuting() {
+ return (this.getAndOrPriorityVolatile(PRIORITY_EXECUTED) & PRIORITY_EXECUTED) == 0;
+ }
+
+ protected boolean isMarkedExecuted() {
+ return (this.getPriorityVolatile() & PRIORITY_EXECUTED) != 0;
+ }
+
+ public void cancel() {
+ if ((this.getAndOrPriorityVolatile(PRIORITY_EXECUTED) & PRIORITY_EXECUTED) != 0) {
+ // cancelled already
+ return;
+ }
+ this.cancelScheduled();
+ }
+
+ public void schedule() {
+ int priority = this.getPriorityVolatile();
+
+ if ((priority & PRIORITY_SCHEDULED) != 0) {
+ throw new IllegalStateException("schedule() called twice");
+ }
+
+ if ((priority & PRIORITY_EXECUTED) != 0) {
+ // cancelled
+ return;
+ }
+
+ this.scheduleTask(PrioritisedExecutor.Priority.getPriority(priority));
+
+ int failures = 0;
+ for (;;) {
+ if (priority == (priority = this.compareAndExchangePriorityVolatile(priority, priority | PRIORITY_SCHEDULED))) {
+ return;
+ }
+
+ if ((priority & PRIORITY_SCHEDULED) != 0) {
+ throw new IllegalStateException("schedule() called twice");
+ }
+
+ if ((priority & PRIORITY_EXECUTED) != 0) {
+ // cancelled or executed
+ return;
+ }
+
+ this.setPriorityScheduled(PrioritisedExecutor.Priority.getPriority(priority));
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public final PrioritisedExecutor.Priority getPriority() {
+ final int ret = this.getPriorityVolatile();
+ if ((ret & PRIORITY_EXECUTED) != 0) {
+ return PrioritisedExecutor.Priority.COMPLETING;
+ }
+ if ((ret & PRIORITY_SCHEDULED) != 0) {
+ return this.getScheduledPriority();
+ }
+ return PrioritisedExecutor.Priority.getPriority(ret);
+ }
+
+ public final void lowerPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ return;
+ }
+
+ if ((curr & PRIORITY_SCHEDULED) != 0) {
+ this.lowerPriorityScheduled(priority);
+ return;
+ }
+
+ if (!priority.isLowerPriority(curr)) {
+ return;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public final void setPriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ return;
+ }
+
+ if ((curr & PRIORITY_SCHEDULED) != 0) {
+ this.setPriorityScheduled(priority);
+ return;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ public final void raisePriority(final PrioritisedExecutor.Priority priority) {
+ if (!PrioritisedExecutor.Priority.isValidPriority(priority)) {
+ throw new IllegalArgumentException("Invalid priority " + priority);
+ }
+
+ int failures = 0;
+ for (int curr = this.getPriorityVolatile();;) {
+ if ((curr & PRIORITY_EXECUTED) != 0) {
+ return;
+ }
+
+ if ((curr & PRIORITY_SCHEDULED) != 0) {
+ this.raisePriorityScheduled(priority);
+ return;
+ }
+
+ if (!priority.isHigherPriority(curr)) {
+ return;
+ }
+
+ if (curr == (curr = this.compareAndExchangePriorityVolatile(curr, priority.priority))) {
+ return;
+ }
+
+ // failed, retry
+
+ ++failures;
+ for (int i = 0; i < failures; ++i) {
+ ConcurrentUtil.backoff();
+ }
+ }
+ }
+
+ protected abstract void cancelScheduled();
+
+ protected abstract PrioritisedExecutor.Priority getScheduledPriority();
+
+ protected abstract void scheduleTask(final PrioritisedExecutor.Priority priority);
+
+ protected abstract void lowerPriorityScheduled(final PrioritisedExecutor.Priority priority);
+
+ protected abstract void setPriorityScheduled(final PrioritisedExecutor.Priority priority);
+
+ protected abstract void raisePriorityScheduled(final PrioritisedExecutor.Priority priority);
+}
diff --git a/src/main/java/io/papermc/paper/command/PaperCommand.java b/src/main/java/io/papermc/paper/command/PaperCommand.java
2023-03-15 13:19:54 +01:00
index a52e6e8be323863ece6624a8ae7f9455279bdc23..e3467aaf6d0c8d486b84362e3c20b3fe631b50ff 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/io/papermc/paper/command/PaperCommand.java
+++ b/src/main/java/io/papermc/paper/command/PaperCommand.java
2023-03-15 13:19:54 +01:00
@@ -40,6 +40,7 @@ public final class PaperCommand extends Command {
commands.put(Set.of("callback"), new CallbackCommand());
2023-02-19 15:57:10 +01:00
commands.put(Set.of("dumpplugins"), new DumpPluginsCommand());
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
commands.put(Set.of("fixlight"), new FixLightCommand());
+ commands.put(Set.of("debug", "chunkinfo", "holderinfo"), new ChunkDebugCommand());
return commands.entrySet().stream()
.flatMap(entry -> entry.getKey().stream().map(s -> Map.entry(s, entry.getValue())))
diff --git a/src/main/java/io/papermc/paper/command/subcommands/ChunkDebugCommand.java b/src/main/java/io/papermc/paper/command/subcommands/ChunkDebugCommand.java
new file mode 100644
index 0000000000000000000000000000000000000000..a6fb7ae77d7cad2243e28a33718e4631f65697fa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- /dev/null
+++ b/src/main/java/io/papermc/paper/command/subcommands/ChunkDebugCommand.java
@@ -0,0 +1,264 @@
+package io.papermc.paper.command.subcommands;
+
+import io.papermc.paper.command.CommandUtil;
+import io.papermc.paper.command.PaperSubcommand;
+import java.io.File;
+import java.time.LocalDateTime;
+import java.time.format.DateTimeFormatter;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import java.util.Locale;
+import io.papermc.paper.util.MCUtil;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+import net.minecraft.server.MinecraftServer;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ImposterProtoChunk;
+import net.minecraft.world.level.chunk.LevelChunk;
+import net.minecraft.world.level.chunk.ProtoChunk;
+import org.bukkit.Bukkit;
+import org.bukkit.command.CommandSender;
+import org.bukkit.craftbukkit.CraftWorld;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.checkerframework.checker.nullness.qual.Nullable;
+import org.checkerframework.framework.qual.DefaultQualifier;
+
+import static net.kyori.adventure.text.Component.text;
+import static net.kyori.adventure.text.format.NamedTextColor.BLUE;
+import static net.kyori.adventure.text.format.NamedTextColor.DARK_AQUA;
+import static net.kyori.adventure.text.format.NamedTextColor.GREEN;
+import static net.kyori.adventure.text.format.NamedTextColor.RED;
+
+@DefaultQualifier(NonNull.class)
+public final class ChunkDebugCommand implements PaperSubcommand {
+ @Override
+ public boolean execute(final CommandSender sender, final String subCommand, final String[] args) {
+ switch (subCommand) {
+ case "debug" -> this.doDebug(sender, args);
+ case "chunkinfo" -> this.doChunkInfo(sender, args);
+ case "holderinfo" -> this.doHolderInfo(sender, args);
+ }
+ return true;
+ }
+
+ @Override
+ public List<String> tabComplete(final CommandSender sender, final String subCommand, final String[] args) {
+ switch (subCommand) {
+ case "debug" -> {
+ if (args.length == 1) {
+ return CommandUtil.getListMatchingLast(sender, args, "help", "chunks");
+ }
+ }
+ case "holderinfo" -> {
+ List<String> worldNames = new ArrayList<>();
+ worldNames.add("*");
+ for (org.bukkit.World world : Bukkit.getWorlds()) {
+ worldNames.add(world.getName());
+ }
+ if (args.length == 1) {
+ return CommandUtil.getListMatchingLast(sender, args, worldNames);
+ }
+ }
+ case "chunkinfo" -> {
+ List<String> worldNames = new ArrayList<>();
+ worldNames.add("*");
+ for (org.bukkit.World world : Bukkit.getWorlds()) {
+ worldNames.add(world.getName());
+ }
+ if (args.length == 1) {
+ return CommandUtil.getListMatchingLast(sender, args, worldNames);
+ }
+ }
+ }
+ return Collections.emptyList();
+ }
+
+ private void doChunkInfo(final CommandSender sender, final String[] args) {
+ List<org.bukkit.World> worlds;
+ if (args.length < 1 || args[0].equals("*")) {
+ worlds = Bukkit.getWorlds();
+ } else {
+ worlds = new ArrayList<>(args.length);
+ for (final String arg : args) {
+ org.bukkit.@Nullable World world = Bukkit.getWorld(arg);
+ if (world == null) {
+ sender.sendMessage(text("World '" + arg + "' is invalid", RED));
+ return;
+ }
+ worlds.add(world);
+ }
+ }
+
+ int accumulatedTotal = 0;
+ int accumulatedInactive = 0;
+ int accumulatedBorder = 0;
+ int accumulatedTicking = 0;
+ int accumulatedEntityTicking = 0;
+
+ for (final org.bukkit.World bukkitWorld : worlds) {
+ final ServerLevel world = ((CraftWorld) bukkitWorld).getHandle();
+
+ int total = 0;
+ int inactive = 0;
+ int border = 0;
+ int ticking = 0;
+ int entityTicking = 0;
+
+ for (final ChunkHolder chunk : io.papermc.paper.chunk.system.ChunkSystem.getVisibleChunkHolders(world)) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ if (chunk.getFullChunkNowUnchecked() == null) {
+ continue;
+ }
+
+ ++total;
+
+ ChunkHolder.FullChunkStatus state = chunk.getFullStatus();
+
+ switch (state) {
+ case INACCESSIBLE -> ++inactive;
+ case BORDER -> ++border;
+ case TICKING -> ++ticking;
+ case ENTITY_TICKING -> ++entityTicking;
+ }
+ }
+
+ accumulatedTotal += total;
+ accumulatedInactive += inactive;
+ accumulatedBorder += border;
+ accumulatedTicking += ticking;
+ accumulatedEntityTicking += entityTicking;
+
+ sender.sendMessage(text().append(text("Chunks in ", BLUE), text(bukkitWorld.getName(), GREEN), text(":")));
+ sender.sendMessage(text().color(DARK_AQUA).append(
+ text("Total: ", BLUE), text(total),
+ text(" Inactive: ", BLUE), text(inactive),
+ text(" Border: ", BLUE), text(border),
+ text(" Ticking: ", BLUE), text(ticking),
+ text(" Entity: ", BLUE), text(entityTicking)
+ ));
+ }
+ if (worlds.size() > 1) {
+ sender.sendMessage(text().append(text("Chunks in ", BLUE), text("all listed worlds", GREEN), text(":", DARK_AQUA)));
+ sender.sendMessage(text().color(DARK_AQUA).append(
+ text("Total: ", BLUE), text(accumulatedTotal),
+ text(" Inactive: ", BLUE), text(accumulatedInactive),
+ text(" Border: ", BLUE), text(accumulatedBorder),
+ text(" Ticking: ", BLUE), text(accumulatedTicking),
+ text(" Entity: ", BLUE), text(accumulatedEntityTicking)
+ ));
+ }
+ }
+
+ private void doHolderInfo(final CommandSender sender, final String[] args) {
+ List<org.bukkit.World> worlds;
+ if (args.length < 1 || args[0].equals("*")) {
+ worlds = Bukkit.getWorlds();
+ } else {
+ worlds = new ArrayList<>(args.length);
+ for (final String arg : args) {
+ org.bukkit.@Nullable World world = Bukkit.getWorld(arg);
+ if (world == null) {
+ sender.sendMessage(text("World '" + arg + "' is invalid", RED));
+ return;
+ }
+ worlds.add(world);
+ }
+ }
+
+ int accumulatedTotal = 0;
+ int accumulatedCanUnload = 0;
+ int accumulatedNull = 0;
+ int accumulatedReadOnly = 0;
+ int accumulatedProtoChunk = 0;
+ int accumulatedFullChunk = 0;
+
+ for (final org.bukkit.World bukkitWorld : worlds) {
+ final ServerLevel world = ((CraftWorld) bukkitWorld).getHandle();
+
+ int total = 0;
+ int canUnload = 0;
+ int nullChunks = 0;
+ int readOnly = 0;
+ int protoChunk = 0;
+ int fullChunk = 0;
+
+ for (final ChunkHolder chunk : world.chunkTaskScheduler.chunkHolderManager.getOldChunkHolders()) { // Paper - change updating chunks map
+ final ChunkAccess lastChunk = chunk.getAvailableChunkNow();
+
+ ++total;
+
+ if (lastChunk == null) {
+ ++nullChunks;
+ } else if (lastChunk instanceof ImposterProtoChunk) {
+ ++readOnly;
+ } else if (lastChunk instanceof ProtoChunk) {
+ ++protoChunk;
+ } else if (lastChunk instanceof LevelChunk) {
+ ++fullChunk;
+ }
+
+ if (chunk.newChunkHolder.isSafeToUnload() == null) {
+ ++canUnload;
+ }
+ }
+
+ accumulatedTotal += total;
+ accumulatedCanUnload += canUnload;
+ accumulatedNull += nullChunks;
+ accumulatedReadOnly += readOnly;
+ accumulatedProtoChunk += protoChunk;
+ accumulatedFullChunk += fullChunk;
+
+ sender.sendMessage(text().append(text("Chunks in ", BLUE), text(bukkitWorld.getName(), GREEN), text(":")));
+ sender.sendMessage(text().color(DARK_AQUA).append(
+ text("Total: ", BLUE), text(total),
+ text(" Unloadable: ", BLUE), text(canUnload),
+ text(" Null: ", BLUE), text(nullChunks),
+ text(" ReadOnly: ", BLUE), text(readOnly),
+ text(" Proto: ", BLUE), text(protoChunk),
+ text(" Full: ", BLUE), text(fullChunk)
+ ));
+ }
+ if (worlds.size() > 1) {
+ sender.sendMessage(text().append(text("Chunks in ", BLUE), text("all listed worlds", GREEN), text(":", DARK_AQUA)));
+ sender.sendMessage(text().color(DARK_AQUA).append(
+ text("Total: ", BLUE), text(accumulatedTotal),
+ text(" Unloadable: ", BLUE), text(accumulatedCanUnload),
+ text(" Null: ", BLUE), text(accumulatedNull),
+ text(" ReadOnly: ", BLUE), text(accumulatedReadOnly),
+ text(" Proto: ", BLUE), text(accumulatedProtoChunk),
+ text(" Full: ", BLUE), text(accumulatedFullChunk)
+ ));
+ }
+ }
+
+ private void doDebug(final CommandSender sender, final String[] args) {
+ if (args.length < 1) {
+ sender.sendMessage(text("Use /paper debug [chunks] help for more information on a specific command", RED));
+ return;
+ }
+
+ final String debugType = args[0].toLowerCase(Locale.ENGLISH);
+ switch (debugType) {
+ case "chunks" -> {
+ if (args.length >= 2 && args[1].toLowerCase(Locale.ENGLISH).equals("help")) {
+ sender.sendMessage(text("Use /paper debug chunks [world] to dump loaded chunk information to a file", RED));
+ break;
+ }
+ File file = new File(new File(new File("."), "debug"),
+ "chunks-" + DateTimeFormatter.ofPattern("yyyy-MM-dd_HH.mm.ss").format(LocalDateTime.now()) + ".txt");
+ sender.sendMessage(text("Writing chunk information dump to " + file, GREEN));
+ try {
+ MCUtil.dumpChunks(file, false);
+ sender.sendMessage(text("Successfully written chunk information!", GREEN));
+ } catch (Throwable thr) {
+ MinecraftServer.LOGGER.warn("Failed to dump chunk information to file " + file.toString(), thr);
+ sender.sendMessage(text("Failed to dump chunk information, see console", RED));
+ }
+ }
+ // "help" & default
+ default -> sender.sendMessage(text("Use /paper debug [chunks] help for more information on a specific command", RED));
+ }
+ }
+
+}
diff --git a/src/main/java/io/papermc/paper/util/MCUtil.java b/src/main/java/io/papermc/paper/util/MCUtil.java
index 902317d2dc198a1cbfc679810bcb2173644354cb..d58d44faa40be2421f4cb54740a3abdbad72875c 100644
--- a/src/main/java/io/papermc/paper/util/MCUtil.java
+++ b/src/main/java/io/papermc/paper/util/MCUtil.java
@@ -2,16 +2,29 @@ package io.papermc.paper.util;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
import com.google.common.util.concurrent.ThreadFactoryBuilder;
import io.papermc.paper.math.Position;
+import com.google.gson.JsonArray;
+import com.google.gson.JsonObject;
+import com.google.gson.internal.Streams;
+import com.google.gson.stream.JsonWriter;
+import com.mojang.datafixers.util.Either;
import it.unimi.dsi.fastutil.objects.ObjectRBTreeSet;
import java.lang.ref.Cleaner;
+import it.unimi.dsi.fastutil.objects.ReferenceArrayList;
import net.minecraft.core.BlockPos;
import net.minecraft.core.Direction;
import net.minecraft.server.MinecraftServer;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ChunkMap;
+import net.minecraft.server.level.DistanceManager;
import net.minecraft.server.level.ServerLevel;
+import net.minecraft.server.level.ServerPlayer;
+import net.minecraft.server.level.Ticket;
import net.minecraft.world.entity.Entity;
import net.minecraft.world.level.ChunkPos;
import net.minecraft.world.level.ClipContext;
import net.minecraft.world.level.Level;
+import net.minecraft.world.level.chunk.ChunkAccess;
+import net.minecraft.world.level.chunk.ChunkStatus;
import org.apache.commons.lang.exception.ExceptionUtils;
import org.bukkit.Location;
import org.bukkit.block.BlockFace;
@@ -21,8 +34,11 @@ import org.spigotmc.AsyncCatcher;
import javax.annotation.Nonnull;
import javax.annotation.Nullable;
+import java.io.*;
+import java.nio.charset.StandardCharsets;
import java.util.List;
import java.util.Queue;
+import java.util.Set;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.LinkedBlockingQueue;
@@ -511,6 +527,100 @@ public final class MCUtil {
}
}
+ public static ChunkStatus getChunkStatus(ChunkHolder chunk) {
+ return chunk.getChunkHolderStatus();
+ }
+
+ public static void dumpChunks(File file, boolean watchdog) throws IOException {
+ file.getParentFile().mkdirs();
+ file.createNewFile();
+ ReferenceArrayList<org.bukkit.World> worlds = new ReferenceArrayList<>(org.bukkit.Bukkit.getWorlds());
+ ReferenceArrayList<org.bukkit.World> loadedWorlds = new ReferenceArrayList<>(worlds);
+ JsonObject data = new JsonObject();
+
+ data.addProperty("server-version", org.bukkit.Bukkit.getVersion());
+ data.addProperty("data-version", 1);
+
+ {
+ JsonArray players = new JsonArray();
+ data.add("all-players", players);
+ List<ServerPlayer> playerList = MinecraftServer.getServer().getPlayerList().players;
+ for (ServerPlayer player : playerList) {
+ JsonObject playerData = new JsonObject();
+ players.add(playerData);
+
+ Level playerWorld = player.getLevel();
+ org.bukkit.World craftWorld = playerWorld.getWorld();
+ Entity.RemovalReason removalReason = player.getRemovalReason();
+
+ playerData.addProperty("name", player.getScoreboardName());
+ playerData.addProperty("x", player.getX());
+ playerData.addProperty("y", player.getY());
+ playerData.addProperty("z", player.getZ());
+ playerData.addProperty("world", playerWorld == null ? "null world" : craftWorld.getName());
+ playerData.addProperty("removalReason", removalReason == null ? "null" : removalReason.name());
+
+ if (!worlds.contains(craftWorld)) {
+ worlds.add(craftWorld);
+ }
+ }
+ }
+
+ JsonArray chunkWaitInformation = new JsonArray();
+ data.add("chunk-wait-infos", chunkWaitInformation);
+
+ for (io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.ChunkInfo chunkInfo : io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.getChunkInfos()) {
+ chunkWaitInformation.add(chunkInfo.toString());
+ }
+
+ JsonArray worldsData = new JsonArray();
+
+ for (org.bukkit.World bukkitWorld : worlds) {
+ JsonObject worldData = new JsonObject();
+
+ ServerLevel world = ((org.bukkit.craftbukkit.CraftWorld)bukkitWorld).getHandle();
+ List<ServerPlayer> players = world.players();
+
+ worldData.addProperty("is-loaded", loadedWorlds.contains(bukkitWorld));
+ worldData.addProperty("name", world.getWorld().getName());
+ worldData.addProperty("view-distance", world.getChunkSource().chunkMap.playerChunkManager.getTargetNoTickViewDistance()); // Paper - replace chunk loader system
+ worldData.addProperty("tick-view-distance", world.getChunkSource().chunkMap.playerChunkManager.getTargetTickViewDistance()); // Paper - replace chunk loader system
+ worldData.addProperty("keep-spawn-loaded", world.keepSpawnInMemory);
+ worldData.addProperty("keep-spawn-loaded-range", world.paperConfig().spawn.keepSpawnLoadedRange * 16);
+
+ JsonArray playersData = new JsonArray();
+
+ for (ServerPlayer player : players) {
+ JsonObject playerData = new JsonObject();
+
+ playerData.addProperty("name", player.getScoreboardName());
+ playerData.addProperty("x", player.getX());
+ playerData.addProperty("y", player.getY());
+ playerData.addProperty("z", player.getZ());
+
+ playersData.add(playerData);
+ }
+
+ worldData.add("players", playersData);
+ worldData.add("chunk-data", watchdog ? world.chunkTaskScheduler.chunkHolderManager.getDebugJsonForWatchdog() : world.chunkTaskScheduler.chunkHolderManager.getDebugJson());
+ worldsData.add(worldData);
+ }
+
+ data.add("worlds", worldsData);
+
+ StringWriter stringWriter = new StringWriter();
+ JsonWriter jsonWriter = new JsonWriter(stringWriter);
+ jsonWriter.setIndent(" ");
+ jsonWriter.setLenient(false);
+ Streams.write(data, jsonWriter);
+
+ String fileData = stringWriter.toString();
+
+ try (PrintStream out = new PrintStream(new FileOutputStream(file), false, StandardCharsets.UTF_8)) {
+ out.print(fileData);
+ }
+ }
+
public static int getTicketLevelFor(net.minecraft.world.level.chunk.ChunkStatus status) {
return net.minecraft.server.level.ChunkMap.MAX_VIEW_DISTANCE + net.minecraft.world.level.chunk.ChunkStatus.getDistance(status);
}
diff --git a/src/main/java/io/papermc/paper/util/TickThread.java b/src/main/java/io/papermc/paper/util/TickThread.java
index d59885ee9c8b29d5bac34dce0597e345e5358c77..fc57850b80303fcade89ca95794f63910404a407 100644
--- a/src/main/java/io/papermc/paper/util/TickThread.java
+++ b/src/main/java/io/papermc/paper/util/TickThread.java
@@ -6,7 +6,7 @@ import net.minecraft.world.entity.Entity;
import org.bukkit.Bukkit;
import java.util.concurrent.atomic.AtomicInteger;
-public final class TickThread extends Thread {
+public class TickThread extends Thread {
public static final boolean STRICT_THREAD_CHECKS = Boolean.getBoolean("paper.strict-thread-checks");
@@ -16,6 +16,10 @@ public final class TickThread extends Thread {
}
}
+ /**
+ * @deprecated
+ */
+ @Deprecated
public static void softEnsureTickThread(final String reason) {
if (!STRICT_THREAD_CHECKS) {
return;
@@ -23,6 +27,10 @@ public final class TickThread extends Thread {
ensureTickThread(reason);
}
+ /**
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ * @deprecated
+ */
+ @Deprecated
public static void ensureTickThread(final String reason) {
if (!isTickThread()) {
MinecraftServer.LOGGER.error("Thread " + Thread.currentThread().getName() + " failed main thread check: " + reason, new Throwable());
@@ -66,14 +74,14 @@ public final class TickThread extends Thread {
}
public static boolean isTickThread() {
- return Bukkit.isPrimaryThread();
+ return Thread.currentThread() instanceof TickThread;
}
public static boolean isTickThreadFor(final ServerLevel world, final int chunkX, final int chunkZ) {
- return Bukkit.isPrimaryThread();
+ return Thread.currentThread() instanceof TickThread;
}
public static boolean isTickThreadFor(final Entity entity) {
- return Bukkit.isPrimaryThread();
+ return Thread.currentThread() instanceof TickThread;
}
}
diff --git a/src/main/java/io/papermc/paper/world/ChunkEntitySlices.java b/src/main/java/io/papermc/paper/world/ChunkEntitySlices.java
new file mode 100644
index 0000000000000000000000000000000000000000..f597d65d56964297eeeed6c7e77703764178fee0
--- /dev/null
+++ b/src/main/java/io/papermc/paper/world/ChunkEntitySlices.java
@@ -0,0 +1,601 @@
+package io.papermc.paper.world;
+
+import com.destroystokyo.paper.util.maplist.EntityList;
+import io.papermc.paper.chunk.system.entity.EntityLookup;
+import io.papermc.paper.util.TickThread;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectMap;
+import it.unimi.dsi.fastutil.objects.Reference2ObjectOpenHashMap;
+import net.minecraft.nbt.CompoundTag;
+import net.minecraft.server.level.ChunkHolder;
+import net.minecraft.server.level.ServerLevel;
+import net.minecraft.util.Mth;
+import net.minecraft.world.entity.Entity;
+import net.minecraft.world.entity.EntityType;
+import net.minecraft.world.entity.boss.EnderDragonPart;
+import net.minecraft.world.entity.boss.enderdragon.EnderDragon;
+import net.minecraft.world.level.ChunkPos;
+import net.minecraft.world.level.chunk.storage.EntityStorage;
+import net.minecraft.world.level.entity.Visibility;
+import net.minecraft.world.phys.AABB;
+import org.bukkit.craftbukkit.event.CraftEventFactory;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Iterator;
+import java.util.List;
+import java.util.function.Predicate;
+
+public final class ChunkEntitySlices {
+
+ protected final int minSection;
+ protected final int maxSection;
+ public final int chunkX;
+ public final int chunkZ;
+ protected final ServerLevel world;
+
+ protected final EntityCollectionBySection allEntities;
+ protected final EntityCollectionBySection hardCollidingEntities;
+ protected final Reference2ObjectOpenHashMap<Class<? extends Entity>, EntityCollectionBySection> entitiesByClass;
+ protected final EntityList entities = new EntityList();
+
+ public ChunkHolder.FullChunkStatus status;
+
+ protected boolean isTransient;
+
+ public boolean isTransient() {
+ return this.isTransient;
+ }
+
+ public void setTransient(final boolean value) {
+ this.isTransient = value;
+ }
+
+ // TODO implement container search optimisations
+
+ public ChunkEntitySlices(final ServerLevel world, final int chunkX, final int chunkZ, final ChunkHolder.FullChunkStatus status,
+ final int minSection, final int maxSection) { // inclusive, inclusive
+ this.minSection = minSection;
+ this.maxSection = maxSection;
+ this.chunkX = chunkX;
+ this.chunkZ = chunkZ;
+ this.world = world;
+
+ this.allEntities = new EntityCollectionBySection(this);
+ this.hardCollidingEntities = new EntityCollectionBySection(this);
+ this.entitiesByClass = new Reference2ObjectOpenHashMap<>();
+
+ this.status = status;
+ }
+
+ // Paper start - optimise CraftChunk#getEntities
+ public org.bukkit.entity.Entity[] getChunkEntities() {
+ List<org.bukkit.entity.Entity> ret = new java.util.ArrayList<>();
+ final Entity[] entities = this.entities.getRawData();
+ for (int i = 0, size = Math.min(entities.length, this.entities.size()); i < size; ++i) {
+ final Entity entity = entities[i];
+ if (entity == null) {
+ continue;
+ }
+ final org.bukkit.entity.Entity bukkit = entity.getBukkitEntity();
+ if (bukkit != null && bukkit.isValid()) {
+ ret.add(bukkit);
+ }
+ }
+
+ return ret.toArray(new org.bukkit.entity.Entity[0]);
+ }
+
+ public CompoundTag save() {
+ final int len = this.entities.size();
+ if (len == 0) {
+ return null;
+ }
+
+ final Entity[] rawData = this.entities.getRawData();
+ final List<Entity> collectedEntities = new ArrayList<>(len);
+ for (int i = 0; i < len; ++i) {
+ final Entity entity = rawData[i];
+ if (entity.shouldBeSaved()) {
+ collectedEntities.add(entity);
+ }
+ }
+
+ if (collectedEntities.isEmpty()) {
+ return null;
+ }
+
+ return EntityStorage.saveEntityChunk(collectedEntities, new ChunkPos(this.chunkX, this.chunkZ), this.world);
+ }
+
+ // returns true if this chunk has transient entities remaining
+ public boolean unload() {
+ final int len = this.entities.size();
+ final Entity[] collectedEntities = Arrays.copyOf(this.entities.getRawData(), len);
+
+ for (int i = 0; i < len; ++i) {
+ final Entity entity = collectedEntities[i];
+ if (entity.isRemoved()) {
+ // removed by us below
+ continue;
+ }
+ if (entity.shouldBeSaved()) {
+ entity.setRemoved(Entity.RemovalReason.UNLOADED_TO_CHUNK);
+ if (entity.isVehicle()) {
+ // we cannot assume that these entities are contained within this chunk, because entities can
+ // desync - so we need to remove them all
+ for (final Entity passenger : entity.getIndirectPassengers()) {
+ passenger.setRemoved(Entity.RemovalReason.UNLOADED_TO_CHUNK);
+ }
+ }
+ }
+ }
+
+ return this.entities.size() != 0;
+ }
+
+ private List<Entity> getAllEntities() {
+ final int len = this.entities.size();
+ if (len == 0) {
+ return new ArrayList<>();
+ }
+
+ final Entity[] rawData = this.entities.getRawData();
+ final List<Entity> collectedEntities = new ArrayList<>(len);
+ for (int i = 0; i < len; ++i) {
+ collectedEntities.add(rawData[i]);
+ }
+
+ return collectedEntities;
+ }
+
+ public void callEntitiesLoadEvent() {
+ CraftEventFactory.callEntitiesLoadEvent(this.world, new ChunkPos(this.chunkX, this.chunkZ), this.getAllEntities());
+ }
+
+ public void callEntitiesUnloadEvent() {
+ CraftEventFactory.callEntitiesUnloadEvent(this.world, new ChunkPos(this.chunkX, this.chunkZ), this.getAllEntities());
+ }
+ // Paper end - optimise CraftChunk#getEntities
+
+ public boolean isEmpty() {
+ return this.entities.size() == 0;
+ }
+
+ public void mergeInto(final ChunkEntitySlices slices) {
+ final Entity[] entities = this.entities.getRawData();
+ for (int i = 0, size = Math.min(entities.length, this.entities.size()); i < size; ++i) {
+ final Entity entity = entities[i];
+ slices.addEntity(entity, entity.sectionY);
+ }
+ }
+
+ private boolean preventStatusUpdates;
+ public boolean startPreventingStatusUpdates() {
+ final boolean ret = this.preventStatusUpdates;
+ this.preventStatusUpdates = true;
+ return ret;
+ }
+
+ public void stopPreventingStatusUpdates(final boolean prev) {
+ this.preventStatusUpdates = prev;
+ }
+
+ public void updateStatus(final ChunkHolder.FullChunkStatus status, final EntityLookup lookup) {
+ this.status = status;
+
+ final Entity[] entities = this.entities.getRawData();
+
+ for (int i = 0, size = this.entities.size(); i < size; ++i) {
+ final Entity entity = entities[i];
+
+ final Visibility oldVisibility = EntityLookup.getEntityStatus(entity);
+ entity.chunkStatus = status;
+ final Visibility newVisibility = EntityLookup.getEntityStatus(entity);
+
+ lookup.entityStatusChange(entity, this, oldVisibility, newVisibility, false, false, false);
+ }
+ }
+
+ public boolean addEntity(final Entity entity, final int chunkSection) {
+ if (!this.entities.add(entity)) {
+ return false;
+ }
+ entity.chunkStatus = this.status;
+ final int sectionIndex = chunkSection - this.minSection;
+
+ this.allEntities.addEntity(entity, sectionIndex);
+
+ if (entity.hardCollides()) {
+ this.hardCollidingEntities.addEntity(entity, sectionIndex);
+ }
+
+ for (final Iterator<Reference2ObjectMap.Entry<Class<? extends Entity>, EntityCollectionBySection>> iterator =
+ this.entitiesByClass.reference2ObjectEntrySet().fastIterator(); iterator.hasNext();) {
+ final Reference2ObjectMap.Entry<Class<? extends Entity>, EntityCollectionBySection> entry = iterator.next();
+
+ if (entry.getKey().isInstance(entity)) {
+ entry.getValue().addEntity(entity, sectionIndex);
+ }
+ }
+
+ return true;
+ }
+
+ public boolean removeEntity(final Entity entity, final int chunkSection) {
+ if (!this.entities.remove(entity)) {
+ return false;
+ }
+ entity.chunkStatus = null;
+ final int sectionIndex = chunkSection - this.minSection;
+
+ this.allEntities.removeEntity(entity, sectionIndex);
+
+ if (entity.hardCollides()) {
+ this.hardCollidingEntities.removeEntity(entity, sectionIndex);
+ }
+
+ for (final Iterator<Reference2ObjectMap.Entry<Class<? extends Entity>, EntityCollectionBySection>> iterator =
+ this.entitiesByClass.reference2ObjectEntrySet().fastIterator(); iterator.hasNext();) {
+ final Reference2ObjectMap.Entry<Class<? extends Entity>, EntityCollectionBySection> entry = iterator.next();
+
+ if (entry.getKey().isInstance(entity)) {
+ entry.getValue().removeEntity(entity, sectionIndex);
+ }
+ }
+
+ return true;
+ }
+
+ public void getHardCollidingEntities(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ this.hardCollidingEntities.getEntities(except, box, into, predicate);
+ }
+
+ public void getEntities(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ this.allEntities.getEntitiesWithEnderDragonParts(except, box, into, predicate);
+ }
+
+ public void getEntitiesWithoutDragonParts(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ this.allEntities.getEntities(except, box, into, predicate);
+ }
+
+ public <T extends Entity> void getEntities(final EntityType<?> type, final AABB box, final List<? super T> into,
+ final Predicate<? super T> predicate) {
+ this.allEntities.getEntities(type, box, (List)into, (Predicate)predicate);
+ }
+
+ protected EntityCollectionBySection initClass(final Class<? extends Entity> clazz) {
+ final EntityCollectionBySection ret = new EntityCollectionBySection(this);
+
+ for (int sectionIndex = 0; sectionIndex < this.allEntities.entitiesBySection.length; ++sectionIndex) {
+ final BasicEntityList<Entity> sectionEntities = this.allEntities.entitiesBySection[sectionIndex];
+ if (sectionEntities == null) {
+ continue;
+ }
+
+ final Entity[] storage = sectionEntities.storage;
+
+ for (int i = 0, len = Math.min(storage.length, sectionEntities.size()); i < len; ++i) {
+ final Entity entity = storage[i];
+
+ if (clazz.isInstance(entity)) {
+ ret.addEntity(entity, sectionIndex);
+ }
+ }
+ }
+
+ return ret;
+ }
+
+ public <T extends Entity> void getEntities(final Class<? extends T> clazz, final Entity except, final AABB box, final List<? super T> into,
+ final Predicate<? super T> predicate) {
+ EntityCollectionBySection collection = this.entitiesByClass.get(clazz);
+ if (collection != null) {
+ collection.getEntitiesWithEnderDragonParts(except, clazz, box, (List)into, (Predicate)predicate);
+ } else {
+ this.entitiesByClass.putIfAbsent(clazz, collection = this.initClass(clazz));
+ collection.getEntitiesWithEnderDragonParts(except, clazz, box, (List)into, (Predicate)predicate);
+ }
+ }
+
+ protected static final class BasicEntityList<E extends Entity> {
+
+ protected static final Entity[] EMPTY = new Entity[0];
+ protected static final int DEFAULT_CAPACITY = 4;
+
+ protected E[] storage;
+ protected int size;
+
+ public BasicEntityList() {
+ this(0);
+ }
+
+ public BasicEntityList(final int cap) {
+ this.storage = (E[])(cap <= 0 ? EMPTY : new Entity[cap]);
+ }
+
+ public boolean isEmpty() {
+ return this.size == 0;
+ }
+
+ public int size() {
+ return this.size;
+ }
+
+ private void resize() {
+ if (this.storage == EMPTY) {
+ this.storage = (E[])new Entity[DEFAULT_CAPACITY];
+ } else {
+ this.storage = Arrays.copyOf(this.storage, this.storage.length * 2);
+ }
+ }
+
+ public void add(final E entity) {
+ final int idx = this.size++;
+ if (idx >= this.storage.length) {
+ this.resize();
+ this.storage[idx] = entity;
+ } else {
+ this.storage[idx] = entity;
+ }
+ }
+
+ public int indexOf(final E entity) {
+ final E[] storage = this.storage;
+
+ for (int i = 0, len = Math.min(this.storage.length, this.size); i < len; ++i) {
+ if (storage[i] == entity) {
+ return i;
+ }
+ }
+
+ return -1;
+ }
+
+ public boolean remove(final E entity) {
+ final int idx = this.indexOf(entity);
+ if (idx == -1) {
+ return false;
+ }
+
+ final int size = --this.size;
+ final E[] storage = this.storage;
+ if (idx != size) {
+ System.arraycopy(storage, idx + 1, storage, idx, size - idx);
+ }
+
+ storage[size] = null;
+
+ return true;
+ }
+
+ public boolean has(final E entity) {
+ return this.indexOf(entity) != -1;
+ }
+ }
+
+ protected static final class EntityCollectionBySection {
+
+ protected final ChunkEntitySlices manager;
+ protected final long[] nonEmptyBitset;
+ protected final BasicEntityList<Entity>[] entitiesBySection;
+ protected int count;
+
+ public EntityCollectionBySection(final ChunkEntitySlices manager) {
+ this.manager = manager;
+
+ final int sectionCount = manager.maxSection - manager.minSection + 1;
+
+ this.nonEmptyBitset = new long[(sectionCount + (Long.SIZE - 1)) >>> 6]; // (sectionCount + (Long.SIZE - 1)) / Long.SIZE
+ this.entitiesBySection = new BasicEntityList[sectionCount];
+ }
+
+ public void addEntity(final Entity entity, final int sectionIndex) {
+ BasicEntityList<Entity> list = this.entitiesBySection[sectionIndex];
+
+ if (list != null && list.has(entity)) {
+ return;
+ }
+
+ if (list == null) {
+ this.entitiesBySection[sectionIndex] = list = new BasicEntityList<>();
+ this.nonEmptyBitset[sectionIndex >>> 6] |= (1L << (sectionIndex & (Long.SIZE - 1)));
+ }
+
+ list.add(entity);
+ ++this.count;
+ }
+
+ public void removeEntity(final Entity entity, final int sectionIndex) {
+ final BasicEntityList<Entity> list = this.entitiesBySection[sectionIndex];
+
+ if (list == null || !list.remove(entity)) {
+ return;
+ }
+
+ --this.count;
+
+ if (list.isEmpty()) {
+ this.entitiesBySection[sectionIndex] = null;
+ this.nonEmptyBitset[sectionIndex >>> 6] ^= (1L << (sectionIndex & (Long.SIZE - 1)));
+ }
+ }
+
+ public void getEntities(final Entity except, final AABB box, final List<Entity> into, final Predicate<? super Entity> predicate) {
+ if (this.count == 0) {
+ return;
+ }
+
+ final int minSection = this.manager.minSection;
+ final int maxSection = this.manager.maxSection;
+
+ final int min = Mth.clamp(Mth.floor(box.minY - 2.0) >> 4, minSection, maxSection);
+ final int max = Mth.clamp(Mth.floor(box.maxY + 2.0) >> 4, minSection, maxSection);
+
+ final BasicEntityList<Entity>[] entitiesBySection = this.entitiesBySection;
+
+ for (int section = min; section <= max; ++section) {
+ final BasicEntityList<Entity> list = entitiesBySection[section - minSection];
+
+ if (list == null) {
+ continue;
+ }
+
+ final Entity[] storage = list.storage;
+
+ for (int i = 0, len = Math.min(storage.length, list.size()); i < len; ++i) {
+ final Entity entity = storage[i];
+
+ if (entity == null || entity == except || !entity.getBoundingBox().intersects(box)) {
+ continue;
+ }
+
+ if (predicate != null && !predicate.test(entity)) {
+ continue;
+ }
+
+ into.add(entity);
+ }
+ }
+ }
+
+ public void getEntitiesWithEnderDragonParts(final Entity except, final AABB box, final List<Entity> into,
+ final Predicate<? super Entity> predicate) {
+ if (this.count == 0) {
+ return;
+ }
+
+ final int minSection = this.manager.minSection;
+ final int maxSection = this.manager.maxSection;
+
+ final int min = Mth.clamp(Mth.floor(box.minY - 2.0) >> 4, minSection, maxSection);
+ final int max = Mth.clamp(Mth.floor(box.maxY + 2.0) >> 4, minSection, maxSection);
+
+ final BasicEntityList<Entity>[] entitiesBySection = this.entitiesBySection;
+
+ for (int section = min; section <= max; ++section) {
+ final BasicEntityList<Entity> list = entitiesBySection[section - minSection];
+
+ if (list == null) {
+ continue;
+ }
+
+ final Entity[] storage = list.storage;
+
+ for (int i = 0, len = Math.min(storage.length, list.size()); i < len; ++i) {
+ final Entity entity = storage[i];
+
+ if (entity == null || entity == except || !entity.getBoundingBox().intersects(box)) {
+ continue;
+ }
+
+ if (predicate == null || predicate.test(entity)) {
+ into.add(entity);
+ } // else: continue to test the ender dragon parts
+
+ if (entity instanceof EnderDragon) {
+ for (final EnderDragonPart part : ((EnderDragon)entity).subEntities) {
+ if (part == except || !part.getBoundingBox().intersects(box)) {
+ continue;
+ }
+
+ if (predicate != null && !predicate.test(part)) {
+ continue;
+ }
+
+ into.add(part);
+ }
+ }
+ }
+ }
+ }
+
+ public void getEntitiesWithEnderDragonParts(final Entity except, final Class<?> clazz, final AABB box, final List<Entity> into,
+ final Predicate<? super Entity> predicate) {
+ if (this.count == 0) {
+ return;
+ }
+
+ final int minSection = this.manager.minSection;
+ final int maxSection = this.manager.maxSection;
+
+ final int min = Mth.clamp(Mth.floor(box.minY - 2.0) >> 4, minSection, maxSection);
+ final int max = Mth.clamp(Mth.floor(box.maxY + 2.0) >> 4, minSection, maxSection);
+
+ final BasicEntityList<Entity>[] entitiesBySection = this.entitiesBySection;
+
+ for (int section = min; section <= max; ++section) {
+ final BasicEntityList<Entity> list = entitiesBySection[section - minSection];
+
+ if (list == null) {
+ continue;
+ }
+
+ final Entity[] storage = list.storage;
+
+ for (int i = 0, len = Math.min(storage.length, list.size()); i < len; ++i) {
+ final Entity entity = storage[i];
+
+ if (entity == null || entity == except || !entity.getBoundingBox().intersects(box)) {
+ continue;
+ }
+
+ if (predicate == null || predicate.test(entity)) {
+ into.add(entity);
+ } // else: continue to test the ender dragon parts
+
+ if (entity instanceof EnderDragon) {
+ for (final EnderDragonPart part : ((EnderDragon)entity).subEntities) {
+ if (part == except || !part.getBoundingBox().intersects(box) || !clazz.isInstance(part)) {
+ continue;
+ }
+
+ if (predicate != null && !predicate.test(part)) {
+ continue;
+ }
+
+ into.add(part);
+ }
+ }
+ }
+ }
+ }
+
+ public <T extends Entity> void getEntities(final EntityType<?> type, final AABB box, final List<? super T> into,
+ final Predicate<? super T> predicate) {
+ if (this.count == 0) {
+ return;
+ }
+
+ final int minSection = this.manager.minSection;
+ final int maxSection = this.manager.maxSection;
+
+ final int min = Mth.clamp(Mth.floor(box.minY - 2.0) >> 4, minSection, maxSection);
+ final int max = Mth.clamp(Mth.floor(box.maxY + 2.0) >> 4, minSection, maxSection);
+
+ final BasicEntityList<Entity>[] entitiesBySection = this.entitiesBySection;
+
+ for (int section = min; section <= max; ++section) {
+ final BasicEntityList<Entity> list = entitiesBySection[section - minSection];
+
+ if (list == null) {
+ continue;
+ }
+
+ final Entity[] storage = list.storage;
+
+ for (int i = 0, len = Math.min(storage.length, list.size()); i < len; ++i) {
+ final Entity entity = storage[i];
+
+ if (entity == null || (type != null && entity.getType() != type) || !entity.getBoundingBox().intersects(box)) {
+ continue;
+ }
+
+ if (predicate != null && !predicate.test((T)entity)) {
+ continue;
+ }
+
+ into.add((T)entity);
+ }
+ }
+ }
+ }
+}
diff --git a/src/main/java/net/minecraft/network/Connection.java b/src/main/java/net/minecraft/network/Connection.java
2023-03-14 19:05:23 +01:00
index 95bc21fdb97d41dc1adddfaf3303d41740df9ed6..da13fd2cde9b54ab5bf87fbb4e90b5da47b6b8f2 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/network/Connection.java
+++ b/src/main/java/net/minecraft/network/Connection.java
2023-03-14 19:05:23 +01:00
@@ -92,6 +92,28 @@ public class Connection extends SimpleChannelInboundHandler<Packet<?>> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private int tickCount;
private boolean handlingFault;
public String hostname = ""; // CraftBukkit - add field
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - add pending task queue
+ private final Queue<Runnable> pendingTasks = new java.util.concurrent.ConcurrentLinkedQueue<>();
+ public void execute(final Runnable run) {
+ if (this.channel == null || !this.channel.isRegistered()) {
+ run.run();
+ return;
+ }
+ final boolean queue = !this.queue.isEmpty();
+ if (!queue) {
+ this.channel.eventLoop().execute(run);
+ } else {
+ this.pendingTasks.add(run);
+ if (this.queue.isEmpty()) {
+ // something flushed async, dump tasks now
+ Runnable r;
+ while ((r = this.pendingTasks.poll()) != null) {
+ this.channel.eventLoop().execute(r);
+ }
+ }
+ }
+ }
+ // Paper end - add pending task queue
public Connection(PacketFlow side) {
this.receiving = side;
2023-03-14 19:05:23 +01:00
@@ -255,6 +277,7 @@ public class Connection extends SimpleChannelInboundHandler<Packet<?>> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private void flushQueue() {
+ try { // Paper - add pending task queue
if (this.channel != null && this.channel.isOpen()) {
Queue queue = this.queue;
2023-03-14 19:05:23 +01:00
@@ -267,6 +290,12 @@ public class Connection extends SimpleChannelInboundHandler<Packet<?>> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
}
+ } finally { // Paper start - add pending task queue
+ Runnable r;
+ while ((r = this.pendingTasks.poll()) != null) {
+ this.channel.eventLoop().execute(r);
+ }
+ } // Paper end - add pending task queue
}
public void tick() {
diff --git a/src/main/java/net/minecraft/network/protocol/game/ServerboundCommandSuggestionPacket.java b/src/main/java/net/minecraft/network/protocol/game/ServerboundCommandSuggestionPacket.java
index a5e438a834826161c52ca9db57d234d9ff80a591..b8bc1b9b8e8a33df90a963f9f9769292bf595642 100644
--- a/src/main/java/net/minecraft/network/protocol/game/ServerboundCommandSuggestionPacket.java
+++ b/src/main/java/net/minecraft/network/protocol/game/ServerboundCommandSuggestionPacket.java
@@ -14,7 +14,7 @@ public class ServerboundCommandSuggestionPacket implements Packet<ServerGamePack
public ServerboundCommandSuggestionPacket(FriendlyByteBuf buf) {
this.id = buf.readVarInt();
- this.command = buf.readUtf(32500);
+ this.command = buf.readUtf(2048);
}
@Override
diff --git a/src/main/java/net/minecraft/server/Main.java b/src/main/java/net/minecraft/server/Main.java
2023-03-14 19:05:23 +01:00
index d9f2518a08bc4ae978051be51e467597f9325ecb..26fa789a44a1d934944e78eae21f489213f50ae3 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/Main.java
+++ b/src/main/java/net/minecraft/server/Main.java
2023-03-14 19:05:23 +01:00
@@ -271,6 +271,7 @@ public class Main {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
convertable_conversionsession.saveDataTag(iregistrycustom_dimension, savedata);
*/
+ Class.forName(net.minecraft.world.entity.npc.VillagerTrades.class.getName());// Paper - load this sync so it won't fail later async
final DedicatedServer dedicatedserver = (DedicatedServer) MinecraftServer.spin((thread) -> {
2022-12-07 19:32:25 +01:00
DedicatedServer dedicatedserver1 = new DedicatedServer(optionset, worldLoader.get(), thread, convertable_conversionsession, resourcepackrepository, worldstem, dedicatedserversettings, DataFixers.getDataFixer(), services, LoggerChunkProgressListener::new);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
diff --git a/src/main/java/net/minecraft/server/MinecraftServer.java b/src/main/java/net/minecraft/server/MinecraftServer.java
index 8b4933653ca2a40d171ae34f7d79022db631a903..6251d93c2ea61c471b4e1069048327782acc78e7 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/MinecraftServer.java
+++ b/src/main/java/net/minecraft/server/MinecraftServer.java
@@ -302,7 +302,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public static <S extends MinecraftServer> S spin(Function<Thread, S> serverFactory) {
AtomicReference<S> atomicreference = new AtomicReference();
- Thread thread = new Thread(() -> {
+ Thread thread = new io.papermc.paper.util.TickThread(() -> { // Paper - rewrite chunk system
((MinecraftServer) atomicreference.get()).runServer();
}, "Server thread");
@@ -585,7 +585,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.forceDifficulty();
for (ServerLevel worldserver : this.getAllLevels()) {
this.prepareLevels(worldserver.getChunkSource().chunkMap.progressListener, worldserver);
- worldserver.entityManager.tick(); // SPIGOT-6526: Load pending entities so they are available to the API
+ //worldserver.entityManager.tick(); // SPIGOT-6526: Load pending entities so they are available to the API // Paper - rewrite chunk system, not required to "tick" anything
this.server.getPluginManager().callEvent(new org.bukkit.event.world.WorldLoadEvent(worldserver.getWorld()));
}
@@ -787,6 +787,12 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public abstract boolean shouldRconBroadcast();
public boolean saveAllChunks(boolean suppressLogs, boolean flush, boolean force) {
+ // Paper start - rewrite chunk system - add close param
+ // This allows us to avoid double saving chunks by closing instead of saving then closing
+ return this.saveAllChunks(suppressLogs, flush, force, false);
+ }
+ public boolean saveAllChunks(boolean suppressLogs, boolean flush, boolean force, boolean close) {
+ // Paper end - rewrite chunk system - add close param
boolean flag3 = false;
for (Iterator iterator = this.getAllLevels().iterator(); iterator.hasNext(); flag3 = true) {
@@ -795,8 +801,12 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
if (!suppressLogs) {
MinecraftServer.LOGGER.info("Saving chunks for level '{}'/{}", worldserver, worldserver.dimension().location());
}
-
- worldserver.save((ProgressListener) null, flush, worldserver.noSave && !force);
+ // Paper start - rewrite chunk system
+ worldserver.save((ProgressListener) null, flush, worldserver.noSave && !force, close);
+ if (flush) {
+ MinecraftServer.LOGGER.info("ThreadedAnvilChunkStorage ({}): All chunks are saved", worldserver.getChunkSource().chunkMap.getStorageName());
+ }
+ // Paper end - rewrite chunk system
}
// CraftBukkit start - moved to WorldServer.save
@@ -815,7 +825,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
while (iterator1.hasNext()) {
ServerLevel worldserver2 = (ServerLevel) iterator1.next();
- MinecraftServer.LOGGER.info("ThreadedAnvilChunkStorage ({}): All chunks are saved", worldserver2.getChunkSource().chunkMap.getStorageName());
+ //MinecraftServer.LOGGER.info("ThreadedAnvilChunkStorage ({}): All chunks are saved", worldserver2.getChunkSource().chunkMap.getStorageName()); // Paper - move up
}
MinecraftServer.LOGGER.info("ThreadedAnvilChunkStorage: All dimensions are saved");
@@ -895,36 +905,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
}
- while (this.levels.values().stream().anyMatch((worldserver1) -> {
- return worldserver1.getChunkSource().chunkMap.hasWork();
- })) {
- this.nextTickTime = Util.getMillis() + 1L;
- iterator = this.getAllLevels().iterator();
-
- while (iterator.hasNext()) {
- worldserver = (ServerLevel) iterator.next();
- worldserver.getChunkSource().removeTicketsOnClosing();
- worldserver.getChunkSource().tick(() -> {
- return true;
- }, false);
- }
-
- this.waitUntilNextTick();
- }
-
- this.saveAllChunks(false, true, false);
- iterator = this.getAllLevels().iterator();
-
- while (iterator.hasNext()) {
- worldserver = (ServerLevel) iterator.next();
- if (worldserver != null) {
- try {
- worldserver.close();
- } catch (IOException ioexception) {
- MinecraftServer.LOGGER.error("Exception closing the level", ioexception);
- }
- }
- }
+ this.saveAllChunks(false, true, false, true); // Paper - rewrite chunk system - move closing into here
this.isSaving = false;
this.resources.close();
@@ -943,7 +924,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.getProfileCache().save();
}
// Spigot end
-
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.close(true); // Paper // Paper - rewrite chunk system
}
public String getLocalIp() {
@@ -977,6 +958,8 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
// Spigot End
+ public static volatile RuntimeException chunkSystemCrash; // Paper - rewrite chunk system
+
protected void runServer() {
try {
if (!this.initServer()) {
@@ -991,6 +974,12 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
Arrays.fill( recentTps, 20 );
long curTime, tickSection = Util.getMillis(), tickCount = 1;
while (this.running) {
+ // Paper start - rewrite chunk system
+ // guarantee that nothing can stop the server from halting if it can at least still tick
+ if (this.chunkSystemCrash != null) {
+ throw this.chunkSystemCrash;
+ }
+ // Paper end - rewrite chunk system
long i = (curTime = Util.getMillis()) - this.nextTickTime;
if (i > 5000L && this.nextTickTime - this.lastOverloadWarning >= 30000L) { // CraftBukkit
@@ -1103,6 +1092,11 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private boolean haveTime() {
+ // Paper start
+ if (this.forceTicks) {
+ return true;
+ }
+ // Paper end
// CraftBukkit start
if (isOversleep) return canOversleep();// Paper - because of our changes, this logic is broken
return this.forceTicks || this.runningTask() || Util.getMillis() < (this.mayHaveDelayedTasks ? this.delayedTasksMaxNextTickTime : this.nextTickTime);
@@ -2258,7 +2252,7 @@ public abstract class MinecraftServer extends ReentrantBlockableEventLoop<TickTa
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// CraftBukkit start
@Override
public boolean isSameThread() {
- return super.isSameThread() || this.isStopped(); // CraftBukkit - MC-142590
+ return io.papermc.paper.util.TickThread.isTickThread(); // Paper - rewrite chunk system
}
public boolean isDebugging() {
diff --git a/src/main/java/net/minecraft/server/dedicated/DedicatedServer.java b/src/main/java/net/minecraft/server/dedicated/DedicatedServer.java
2022-12-07 19:32:25 +01:00
index 5e5c4de89784db702256ee765091e929066116e4..8318af4db5fa7242b5cb2ae50fd08174017d6f7d 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/dedicated/DedicatedServer.java
+++ b/src/main/java/net/minecraft/server/dedicated/DedicatedServer.java
2022-12-07 19:32:25 +01:00
@@ -400,7 +400,34 @@ public class DedicatedServer extends MinecraftServer implements ServerInterface
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return this.getProperties().allowNether;
}
+ static final java.util.concurrent.atomic.AtomicInteger ASYNC_DEBUG_CHUNKS_COUNT = new java.util.concurrent.atomic.AtomicInteger(); // Paper - rewrite chunk system
+
public void handleConsoleInput(String command, CommandSourceStack commandSource) {
+ // Paper start - rewrite chunk system
+ if (command.equalsIgnoreCase("paper debug chunks --async")) {
+ LOGGER.info("Scheduling async debug chunks");
+ Runnable run = () -> {
+ LOGGER.info("Async debug chunks executing");
+ io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.dumpAllChunkLoadInfo(false);
+ CommandSender sender = MinecraftServer.getServer().console;
+ java.io.File file = new java.io.File(new java.io.File(new java.io.File("."), "debug"),
+ "chunks-" + java.time.format.DateTimeFormatter.ofPattern("yyyy-MM-dd_HH.mm.ss").format(java.time.LocalDateTime.now()) + ".txt");
+ sender.sendMessage(net.kyori.adventure.text.Component.text("Writing chunk information dump to " + file, net.kyori.adventure.text.format.NamedTextColor.GREEN));
+ try {
+ io.papermc.paper.util.MCUtil.dumpChunks(file, true);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ sender.sendMessage(net.kyori.adventure.text.Component.text("Successfully written chunk information!", net.kyori.adventure.text.format.NamedTextColor.GREEN));
+ } catch (Throwable thr) {
+ MinecraftServer.LOGGER.warn("Failed to dump chunk information to file " + file.toString(), thr);
+ sender.sendMessage(net.kyori.adventure.text.Component.text("Failed to dump chunk information, see console", net.kyori.adventure.text.format.NamedTextColor.RED));
+ }
+ };
+ Thread t = new Thread(run);
+ t.setName("Async debug thread #" + ASYNC_DEBUG_CHUNKS_COUNT.getAndIncrement());
+ t.setDaemon(true);
+ t.start();
+ return;
+ }
+ // Paper end - rewrite chunk system
this.consoleInput.add(new ConsoleInput(command, commandSource));
}
diff --git a/src/main/java/net/minecraft/server/level/ChunkHolder.java b/src/main/java/net/minecraft/server/level/ChunkHolder.java
2023-03-14 19:05:23 +01:00
index 30e330cc3a2f51e4ec9555671e3eeb7bb14bdfa3..214e0657035f82e5266de06e31975d6b45405019 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/ChunkHolder.java
+++ b/src/main/java/net/minecraft/server/level/ChunkHolder.java
@@ -50,17 +50,12 @@ public class ChunkHolder {
private static final List<ChunkStatus> CHUNK_STATUSES = ChunkStatus.getStatusList();
private static final ChunkHolder.FullChunkStatus[] FULL_CHUNK_STATUSES = ChunkHolder.FullChunkStatus.values();
private static final int BLOCKS_BEFORE_RESEND_FUDGE = 64;
- private final AtomicReferenceArray<CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>>> futures;
+ // Paper - rewrite chunk system
private final LevelHeightAccessor levelHeightAccessor;
- private volatile CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> fullChunkFuture; private int fullChunkCreateCount; private volatile boolean isFullChunkReady; // Paper - cache chunk ticking stage
- private volatile CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> tickingChunkFuture; private volatile boolean isTickingReady; // Paper - cache chunk ticking stage
- private volatile CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> entityTickingChunkFuture; private volatile boolean isEntityTickingReady; // Paper - cache chunk ticking stage
- public CompletableFuture<ChunkAccess> chunkToSave; // Paper - public
+ // Paper - rewrite chunk system
@Nullable
private final DebugBuffer<ChunkHolder.ChunkSaveDebug> chunkToSaveHistory;
- public int oldTicketLevel;
- private int ticketLevel;
- private int queueLevel;
+ // Paper - rewrite chunk system
public final ChunkPos pos;
private boolean hasChangedSections;
private final ShortSet[] changedBlocksPerSection;
@@ -69,11 +64,22 @@ public class ChunkHolder {
private final LevelLightEngine lightEngine;
private final ChunkHolder.LevelChangeListener onLevelChange;
public final ChunkHolder.PlayerProvider playerProvider;
- private boolean wasAccessibleSinceLastSave;
+ // Paper - rewrite chunk system
private boolean resendLight;
- private CompletableFuture<Void> pendingFullStateConfirmation;
+ // Paper - rewrite chunk system
private final ChunkMap chunkMap; // Paper
+ // Paper start - no-tick view distance
+ public final LevelChunk getSendingChunk() {
+ // it's important that we use getChunkAtIfLoadedImmediately to mirror the chunk sending logic used
+ // in Chunk's neighbour callback
+ LevelChunk ret = this.chunkMap.level.getChunkSource().getChunkAtIfLoadedImmediately(this.pos.x, this.pos.z);
+ if (ret != null && ret.areNeighboursLoaded(1)) {
+ return ret;
+ }
+ return null;
+ }
+ // Paper end - no-tick view distance
// Paper start
public void onChunkAdd() {
@@ -85,148 +91,106 @@ public class ChunkHolder {
}
// Paper end
- public ChunkHolder(ChunkPos pos, int level, LevelHeightAccessor world, LevelLightEngine lightingProvider, ChunkHolder.LevelChangeListener levelUpdateListener, ChunkHolder.PlayerProvider playersWatchingChunkProvider) {
- this.futures = new AtomicReferenceArray(ChunkHolder.CHUNK_STATUSES.size());
- this.fullChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- this.tickingChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- this.entityTickingChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- this.chunkToSave = CompletableFuture.completedFuture(null); // CraftBukkit - decompile error
+ public final io.papermc.paper.chunk.system.scheduling.NewChunkHolder newChunkHolder; // Paper - rewrite chunk system
+
+ public ChunkHolder(ChunkPos pos, LevelHeightAccessor world, LevelLightEngine lightingProvider, ChunkHolder.PlayerProvider playersWatchingChunkProvider, io.papermc.paper.chunk.system.scheduling.NewChunkHolder newChunkHolder) { // Paper - rewrite chunk system
+ this.newChunkHolder = newChunkHolder; // Paper - rewrite chunk system
this.chunkToSaveHistory = null;
this.blockChangedLightSectionFilter = new BitSet();
this.skyChangedLightSectionFilter = new BitSet();
- this.pendingFullStateConfirmation = CompletableFuture.completedFuture(null); // CraftBukkit - decompile error
+ // Paper - rewrite chunk system
this.pos = pos;
this.levelHeightAccessor = world;
this.lightEngine = lightingProvider;
- this.onLevelChange = levelUpdateListener;
+ this.onLevelChange = null; // Paper - rewrite chunk system
this.playerProvider = playersWatchingChunkProvider;
- this.oldTicketLevel = ChunkMap.MAX_CHUNK_DISTANCE + 1;
- this.ticketLevel = this.oldTicketLevel;
- this.queueLevel = this.oldTicketLevel;
- this.setTicketLevel(level);
+ // Paper - rewrite chunk system
this.changedBlocksPerSection = new ShortSet[world.getSectionsCount()];
this.chunkMap = (ChunkMap)playersWatchingChunkProvider; // Paper
}
// Paper start
public @Nullable ChunkAccess getAvailableChunkNow() {
- // TODO can we just getStatusFuture(EMPTY)?
- for (ChunkStatus curr = ChunkStatus.FULL, next = curr.getParent(); curr != next; curr = next, next = next.getParent()) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> future = this.getFutureIfPresentUnchecked(curr);
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = future.getNow(null);
- if (either == null || either.left().isEmpty()) {
- continue;
- }
- return either.left().get();
- }
- return null;
+ return this.newChunkHolder.getCurrentChunk(); // Paper - rewrite chunk system
}
// Paper end
// CraftBukkit start
public LevelChunk getFullChunkNow() {
- // Note: We use the oldTicketLevel for isLoaded checks.
- if (!ChunkHolder.getFullChunkStatus(this.oldTicketLevel).isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) return null;
- return this.getFullChunkNowUnchecked();
+ // Paper start - rewrite chunk system
+ ChunkAccess chunk = this.getAvailableChunkNow();
+ if (!this.isFullChunkReady() || !(chunk instanceof LevelChunk)) return null; // instanceof to avoid a race condition on off-main threads
+ return (LevelChunk)chunk;
+ // Paper end - rewrite chunk system
}
public LevelChunk getFullChunkNowUnchecked() {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> statusFuture = this.getFutureIfPresentUnchecked(ChunkStatus.FULL);
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = (Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>) statusFuture.getNow(null);
- return (either == null) ? null : (LevelChunk) either.left().orElse(null);
+ // Paper start - rewrite chunk system
+ ChunkAccess chunk = this.getAvailableChunkNow();
+ return chunk instanceof LevelChunk ? (LevelChunk)chunk : null;
+ // Paper end - rewrite chunk system
}
// CraftBukkit end
public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getFutureIfPresentUnchecked(ChunkStatus leastStatus) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = (CompletableFuture) this.futures.get(leastStatus.getIndex());
-
- return completablefuture == null ? ChunkHolder.UNLOADED_CHUNK_FUTURE : completablefuture;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getFutureIfPresent(ChunkStatus leastStatus) {
- return ChunkHolder.getStatus(this.ticketLevel).isOrAfter(leastStatus) ? this.getFutureIfPresentUnchecked(leastStatus) : ChunkHolder.UNLOADED_CHUNK_FUTURE;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public final CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> getTickingChunkFuture() { // Paper - final for inline
- return this.tickingChunkFuture;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public final CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> getEntityTickingChunkFuture() { // Paper - final for inline
- return this.entityTickingChunkFuture;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public final CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> getFullChunkFuture() { // Paper - final for inline
- return this.fullChunkFuture;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@Nullable
public final LevelChunk getTickingChunk() { // Paper - final for inline
- CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getTickingChunkFuture();
- Either<LevelChunk, ChunkHolder.ChunkLoadingFailure> either = (Either) completablefuture.getNow(null); // CraftBukkit - decompile error
-
- return either == null ? null : (LevelChunk) either.left().orElse(null); // CraftBukkit - decompile error
+ // Paper start - rewrite chunk system
+ if (!this.isTickingReady()) {
+ return null;
+ }
+ return (LevelChunk)this.getAvailableChunkNow();
+ // Paper end - rewrite chunk system
}
@Nullable
public final LevelChunk getFullChunk() { // Paper - final for inline
- CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getFullChunkFuture();
- Either<LevelChunk, ChunkHolder.ChunkLoadingFailure> either = (Either) completablefuture.getNow(null); // CraftBukkit - decompile error
-
- return either == null ? null : (LevelChunk) either.left().orElse(null); // CraftBukkit - decompile error
+ // Paper start - rewrite chunk system
+ if (!this.isFullChunkReady()) {
+ return null;
+ }
+ return (LevelChunk)this.getAvailableChunkNow();
+ // Paper end - rewrite chunk system
}
@Nullable
public ChunkStatus getLastAvailableStatus() {
- for (int i = ChunkHolder.CHUNK_STATUSES.size() - 1; i >= 0; --i) {
- ChunkStatus chunkstatus = (ChunkStatus) ChunkHolder.CHUNK_STATUSES.get(i);
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getFutureIfPresentUnchecked(chunkstatus);
-
- if (((Either) completablefuture.getNow(ChunkHolder.UNLOADED_CHUNK)).left().isPresent()) {
- return chunkstatus;
- }
- }
-
- return null;
+ return this.newChunkHolder.getCurrentGenStatus(); // Paper - rewrite chunk system
}
// Paper start
public ChunkStatus getChunkHolderStatus() {
- for (ChunkStatus curr = ChunkStatus.FULL, next = curr.getParent(); curr != next; curr = next, next = next.getParent()) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> future = this.getFutureIfPresentUnchecked(curr);
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = future.getNow(null);
- if (either == null || !either.left().isPresent()) {
- continue;
- }
- return curr;
- }
-
- return null;
+ return this.newChunkHolder.getCurrentGenStatus(); // Paper - rewrite chunk system
}
// Paper end
@Nullable
public ChunkAccess getLastAvailable() {
- for (int i = ChunkHolder.CHUNK_STATUSES.size() - 1; i >= 0; --i) {
- ChunkStatus chunkstatus = (ChunkStatus) ChunkHolder.CHUNK_STATUSES.get(i);
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getFutureIfPresentUnchecked(chunkstatus);
-
- if (!completablefuture.isCompletedExceptionally()) {
- Optional<ChunkAccess> optional = ((Either) completablefuture.getNow(ChunkHolder.UNLOADED_CHUNK)).left();
-
- if (optional.isPresent()) {
- return (ChunkAccess) optional.get();
- }
- }
- }
-
- return null;
+ return this.newChunkHolder.getCurrentChunk(); // Paper - rewrite chunk system
}
- public final CompletableFuture<ChunkAccess> getChunkToSave() { // Paper - final for inline
- return this.chunkToSave;
- }
+ // Paper - rewrite chunk system
public void blockChanged(BlockPos pos) {
- LevelChunk chunk = this.getTickingChunk();
+ LevelChunk chunk = this.getSendingChunk(); // Paper - no-tick view distance
if (chunk != null) {
int i = this.levelHeightAccessor.getSectionIndex(pos.getY());
@@ -242,14 +206,15 @@ public class ChunkHolder {
}
public void sectionLightChanged(LightLayer lightType, int y) {
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = (Either) this.getFutureIfPresent(ChunkStatus.FEATURES).getNow(null); // CraftBukkit - decompile error
+ // Paper start - no-tick view distance
- if (either != null) {
- ChunkAccess ichunkaccess = (ChunkAccess) either.left().orElse(null); // CraftBukkit - decompile error
+ if (true) {
+ ChunkAccess ichunkaccess = this.getAvailableChunkNow();
if (ichunkaccess != null) {
ichunkaccess.setUnsaved(true);
- LevelChunk chunk = this.getTickingChunk();
+ LevelChunk chunk = this.getSendingChunk();
+ // Paper end - no-tick view distance
if (chunk != null) {
int j = this.lightEngine.getMinLightSection();
@@ -340,66 +305,32 @@ public class ChunkHolder {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public void broadcast(Packet<?> packet, boolean onlyOnWatchDistanceEdge) {
- this.playerProvider.getPlayers(this.pos, onlyOnWatchDistanceEdge).forEach((entityplayer) -> {
- entityplayer.connection.send(packet);
- });
- }
2023-03-23 22:57:03 +01:00
-
- public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getOrScheduleFuture(ChunkStatus targetStatus, ChunkMap chunkStorage) {
- int i = targetStatus.getIndex();
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = (CompletableFuture) this.futures.get(i);
-
- if (completablefuture != null) {
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = (Either) completablefuture.getNow(ChunkHolder.NOT_DONE_YET);
-
- if (either == null) {
- String s = "value in future for status: " + targetStatus + " was incorrectly set to null at chunk: " + this.pos;
2023-03-23 22:57:03 +01:00
+ // Paper start - per player view distance
+ // there can be potential desync with player's last mapped section and the view distance map, so use the
+ // view distance map here.
+ com.destroystokyo.paper.util.misc.PlayerAreaMap viewDistanceMap = this.chunkMap.playerChunkManager.broadcastMap; // Paper - replace old player chunk manager
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> players = viewDistanceMap.getObjectsInRange(this.pos);
+ if (players == null) {
+ return;
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- throw chunkStorage.debugFuturesAndCreateReportedException(new IllegalStateException("null value previously set for chunk status"), s);
+ Object[] backingSet = players.getBackingSet();
+ for (int i = 0, len = backingSet.length; i < len; ++i) {
+ if (!(backingSet[i] instanceof ServerPlayer player)) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ continue;
}
-
- if (either == ChunkHolder.NOT_DONE_YET || either.right().isEmpty()) {
- return completablefuture;
+ if (!this.chunkMap.playerChunkManager.isChunkSent(player, this.pos.x, this.pos.z, onlyOnWatchDistanceEdge)) {
+ continue;
}
+ player.connection.send(packet);
}
-
- if (ChunkHolder.getStatus(this.ticketLevel).isOrAfter(targetStatus)) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture1 = chunkStorage.schedule(this, targetStatus);
-
- this.updateChunkToSave(completablefuture1, "schedule " + targetStatus);
- this.futures.set(i, completablefuture1);
- return completablefuture1;
- } else {
- return completablefuture == null ? ChunkHolder.UNLOADED_CHUNK_FUTURE : completablefuture;
- }
+ // Paper end - per player view distance
}
- protected void addSaveDependency(String thenDesc, CompletableFuture<?> then) {
- if (this.chunkToSaveHistory != null) {
- this.chunkToSaveHistory.push(new ChunkHolder.ChunkSaveDebug(Thread.currentThread(), then, thenDesc));
- }
-
- this.chunkToSave = this.chunkToSave.thenCombine(then, (ichunkaccess, object) -> {
- return ichunkaccess;
- });
- }
-
- private void updateChunkToSave(CompletableFuture<? extends Either<? extends ChunkAccess, ChunkHolder.ChunkLoadingFailure>> then, String thenDesc) {
- if (this.chunkToSaveHistory != null) {
- this.chunkToSaveHistory.push(new ChunkHolder.ChunkSaveDebug(Thread.currentThread(), then, thenDesc));
- }
-
- this.chunkToSave = this.chunkToSave.thenCombine(then, (ichunkaccess, either) -> {
- return (ChunkAccess) either.map((ichunkaccess1) -> {
- return ichunkaccess1;
- }, (playerchunk_failure) -> {
- return ichunkaccess;
- });
- });
- }
+ // Paper - rewrite chunk system
public ChunkHolder.FullChunkStatus getFullStatus() {
- return ChunkHolder.getFullChunkStatus(this.ticketLevel);
+ return this.newChunkHolder.getChunkStatus(); // Paper - rewrite chunk system
}
public final ChunkPos getPos() { // Paper - final for inline
@@ -407,207 +338,10 @@ public class ChunkHolder {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public final int getTicketLevel() { // Paper - final for inline
- return this.ticketLevel;
2023-03-23 22:57:03 +01:00
- }
-
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- public int getQueueLevel() {
- return this.queueLevel;
- }
-
- private void setQueueLevel(int level) {
- this.queueLevel = level;
- }
-
- public void setTicketLevel(int level) {
- this.ticketLevel = level;
- }
-
- private void scheduleFullChunkPromotion(ChunkMap playerchunkmap, CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> completablefuture, Executor executor, ChunkHolder.FullChunkStatus playerchunk_state) {
- this.pendingFullStateConfirmation.cancel(false);
- CompletableFuture<Void> completablefuture1 = new CompletableFuture();
-
- completablefuture1.thenRunAsync(() -> {
- playerchunkmap.onFullChunkStatusChange(this.pos, playerchunk_state);
- }, executor);
- this.pendingFullStateConfirmation = completablefuture1;
- completablefuture.thenAccept((either) -> {
- either.ifLeft((chunk) -> {
- completablefuture1.complete(null); // CraftBukkit - decompile error
- });
- });
2023-03-23 22:57:03 +01:00
+ return this.newChunkHolder.getTicketLevel(); // Paper - rewrite chunk system
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- private void demoteFullChunk(ChunkMap playerchunkmap, ChunkHolder.FullChunkStatus playerchunk_state) {
- this.pendingFullStateConfirmation.cancel(false);
- playerchunkmap.onFullChunkStatusChange(this.pos, playerchunk_state);
- }
-
- protected void updateFutures(ChunkMap chunkStorage, Executor executor) {
- ChunkStatus chunkstatus = ChunkHolder.getStatus(this.oldTicketLevel);
- ChunkStatus chunkstatus1 = ChunkHolder.getStatus(this.ticketLevel);
- boolean flag = this.oldTicketLevel <= ChunkMap.MAX_CHUNK_DISTANCE;
- boolean flag1 = this.ticketLevel <= ChunkMap.MAX_CHUNK_DISTANCE;
- ChunkHolder.FullChunkStatus playerchunk_state = ChunkHolder.getFullChunkStatus(this.oldTicketLevel);
- ChunkHolder.FullChunkStatus playerchunk_state1 = ChunkHolder.getFullChunkStatus(this.ticketLevel);
- // CraftBukkit start
- // ChunkUnloadEvent: Called before the chunk is unloaded: isChunkLoaded is still true and chunk can still be modified by plugins.
- if (playerchunk_state.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && !playerchunk_state1.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
- this.getFutureIfPresentUnchecked(ChunkStatus.FULL).thenAccept((either) -> {
- LevelChunk chunk = (LevelChunk)either.left().orElse(null);
- if (chunk != null) {
- chunkStorage.callbackExecutor.execute(() -> {
- // Minecraft will apply the chunks tick lists to the world once the chunk got loaded, and then store the tick
- // lists again inside the chunk once the chunk becomes inaccessible and set the chunk's needsSaving flag.
- // These actions may however happen deferred, so we manually set the needsSaving flag already here.
- chunk.setUnsaved(true);
- chunk.unloadCallback();
- });
- }
- }).exceptionally((throwable) -> {
- // ensure exceptions are printed, by default this is not the case
- MinecraftServer.LOGGER.error("Failed to schedule unload callback for chunk " + ChunkHolder.this.pos, throwable);
- return null;
- });
-
- // Run callback right away if the future was already done
- chunkStorage.callbackExecutor.run();
- }
- // CraftBukkit end
-
- if (flag) {
- Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = Either.right(new ChunkHolder.ChunkLoadingFailure() {
- public String toString() {
- return "Unloaded ticket level " + ChunkHolder.this.pos;
- }
- });
-
- for (int i = flag1 ? chunkstatus1.getIndex() + 1 : 0; i <= chunkstatus.getIndex(); ++i) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = (CompletableFuture) this.futures.get(i);
-
- if (completablefuture == null) {
- this.futures.set(i, CompletableFuture.completedFuture(either));
- }
- }
- }
-
- boolean flag2 = playerchunk_state.isOrAfter(ChunkHolder.FullChunkStatus.BORDER);
- boolean flag3 = playerchunk_state1.isOrAfter(ChunkHolder.FullChunkStatus.BORDER);
-
- this.wasAccessibleSinceLastSave |= flag3;
- if (!flag2 && flag3) {
- int expectCreateCount = ++this.fullChunkCreateCount; // Paper
- this.fullChunkFuture = chunkStorage.prepareAccessibleChunk(this);
- this.scheduleFullChunkPromotion(chunkStorage, this.fullChunkFuture, executor, ChunkHolder.FullChunkStatus.BORDER);
- // Paper start - cache ticking ready status
- this.fullChunkFuture.thenAccept(either -> {
- final Optional<LevelChunk> left = either.left();
- if (left.isPresent() && ChunkHolder.this.fullChunkCreateCount == expectCreateCount) {
- LevelChunk fullChunk = either.left().get();
- ChunkHolder.this.isFullChunkReady = true;
- io.papermc.paper.chunk.system.ChunkSystem.onChunkBorder(fullChunk, this);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
- });
- this.updateChunkToSave(this.fullChunkFuture, "full");
- }
-
- if (flag2 && !flag3) {
- // Paper start
- if (this.isFullChunkReady) {
- io.papermc.paper.chunk.system.ChunkSystem.onChunkNotBorder(this.fullChunkFuture.join().left().get(), this); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
- // Paper end
- this.fullChunkFuture.complete(ChunkHolder.UNLOADED_LEVEL_CHUNK);
- this.fullChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- ++this.fullChunkCreateCount; // Paper - cache ticking ready status
- this.isFullChunkReady = false; // Paper - cache ticking ready status
- }
-
- boolean flag4 = playerchunk_state.isOrAfter(ChunkHolder.FullChunkStatus.TICKING);
- boolean flag5 = playerchunk_state1.isOrAfter(ChunkHolder.FullChunkStatus.TICKING);
-
- if (!flag4 && flag5) {
- this.tickingChunkFuture = chunkStorage.prepareTickingChunk(this);
- this.scheduleFullChunkPromotion(chunkStorage, this.tickingChunkFuture, executor, ChunkHolder.FullChunkStatus.TICKING);
- // Paper start - cache ticking ready status
- this.tickingChunkFuture.thenAccept(either -> {
- either.ifLeft(chunk -> {
- // note: Here is a very good place to add callbacks to logic waiting on this.
- ChunkHolder.this.isTickingReady = true;
- io.papermc.paper.chunk.system.ChunkSystem.onChunkTicking(chunk, this);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- });
- });
- // Paper end
- this.updateChunkToSave(this.tickingChunkFuture, "ticking");
- }
-
- if (flag4 && !flag5) {
- // Paper start
- if (this.isTickingReady) {
- io.papermc.paper.chunk.system.ChunkSystem.onChunkNotTicking(this.tickingChunkFuture.join().left().get(), this); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
- // Paper end
- this.tickingChunkFuture.complete(ChunkHolder.UNLOADED_LEVEL_CHUNK); this.isTickingReady = false; // Paper - cache chunk ticking stage
- this.tickingChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- }
-
- boolean flag6 = playerchunk_state.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING);
- boolean flag7 = playerchunk_state1.isOrAfter(ChunkHolder.FullChunkStatus.ENTITY_TICKING);
-
- if (!flag6 && flag7) {
- if (this.entityTickingChunkFuture != ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE) {
- throw (IllegalStateException) Util.pauseInIde(new IllegalStateException());
- }
-
- this.entityTickingChunkFuture = chunkStorage.prepareEntityTickingChunk(this.pos);
- this.scheduleFullChunkPromotion(chunkStorage, this.entityTickingChunkFuture, executor, ChunkHolder.FullChunkStatus.ENTITY_TICKING);
- // Paper start - cache ticking ready status
- this.entityTickingChunkFuture.thenAccept(either -> {
- either.ifLeft(chunk -> {
- ChunkHolder.this.isEntityTickingReady = true;
- io.papermc.paper.chunk.system.ChunkSystem.onChunkEntityTicking(chunk, this);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- });
- });
- // Paper end
- this.updateChunkToSave(this.entityTickingChunkFuture, "entity ticking");
- }
-
- if (flag6 && !flag7) {
- // Paper start
- if (this.isEntityTickingReady) {
- io.papermc.paper.chunk.system.ChunkSystem.onChunkNotEntityTicking(this.entityTickingChunkFuture.join().left().get(), this);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
- // Paper end
- this.entityTickingChunkFuture.complete(ChunkHolder.UNLOADED_LEVEL_CHUNK); this.isEntityTickingReady = false; // Paper - cache chunk ticking stage
- this.entityTickingChunkFuture = ChunkHolder.UNLOADED_LEVEL_CHUNK_FUTURE;
- }
-
- if (!playerchunk_state1.isOrAfter(playerchunk_state)) {
- this.demoteFullChunk(chunkStorage, playerchunk_state1);
- }
-
- this.onLevelChange.onLevelChange(this.pos, this::getQueueLevel, this.ticketLevel, this::setQueueLevel);
- this.oldTicketLevel = this.ticketLevel;
- // CraftBukkit start
- // ChunkLoadEvent: Called after the chunk is loaded: isChunkLoaded returns true and chunk is ready to be modified by plugins.
- if (!playerchunk_state.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && playerchunk_state1.isOrAfter(ChunkHolder.FullChunkStatus.BORDER)) {
- this.getFutureIfPresentUnchecked(ChunkStatus.FULL).thenAccept((either) -> {
- LevelChunk chunk = (LevelChunk)either.left().orElse(null);
- if (chunk != null) {
- chunkStorage.callbackExecutor.execute(() -> {
- chunk.loadCallback();
- });
- }
- }).exceptionally((throwable) -> {
- // ensure exceptions are printed, by default this is not the case
- MinecraftServer.LOGGER.error("Failed to schedule load callback for chunk " + ChunkHolder.this.pos, throwable);
- return null;
- });
-
- // Run callback right away if the future was already done
- chunkStorage.callbackExecutor.run();
- }
- // CraftBukkit end
- }
+ // Paper - rewrite chunk system
public static ChunkStatus getStatus(int level) {
return level < 33 ? ChunkStatus.FULL : ChunkStatus.getStatusAroundFullChunk(level - 33);
@@ -617,38 +351,14 @@ public class ChunkHolder {
2023-03-14 19:05:23 +01:00
return ChunkHolder.FULL_CHUNK_STATUSES[Mth.clamp(33 - distance + 1, 0, ChunkHolder.FULL_CHUNK_STATUSES.length - 1)];
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
- public boolean wasAccessibleSinceLastSave() {
- return this.wasAccessibleSinceLastSave;
- }
-
- public void refreshAccessibility() {
- this.wasAccessibleSinceLastSave = ChunkHolder.getFullChunkStatus(this.ticketLevel).isOrAfter(ChunkHolder.FullChunkStatus.BORDER);
- }
+ // Paper - rewrite chunk system
public void replaceProtoChunk(ImposterProtoChunk chunk) {
- for (int i = 0; i < this.futures.length(); ++i) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = (CompletableFuture) this.futures.get(i);
-
- if (completablefuture != null) {
- Optional<ChunkAccess> optional = ((Either) completablefuture.getNow(ChunkHolder.UNLOADED_CHUNK)).left();
-
- if (!optional.isEmpty() && optional.get() instanceof ProtoChunk) {
- this.futures.set(i, CompletableFuture.completedFuture(Either.left(chunk)));
- }
- }
- }
-
- this.updateChunkToSave(CompletableFuture.completedFuture(Either.left(chunk.getWrapped())), "replaceProto");
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public List<Pair<ChunkStatus, CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>>>> getAllFutures() {
- List<Pair<ChunkStatus, CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>>>> list = new ArrayList();
-
- for (int i = 0; i < ChunkHolder.CHUNK_STATUSES.size(); ++i) {
- list.add(Pair.of((ChunkStatus) ChunkHolder.CHUNK_STATUSES.get(i), (CompletableFuture) this.futures.get(i)));
- }
-
- return list;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@FunctionalInterface
@@ -697,15 +407,15 @@ public class ChunkHolder {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// Paper start
public final boolean isEntityTickingReady() {
- return this.isEntityTickingReady;
+ return this.newChunkHolder.isEntityTickingReady(); // Paper - rewrite chunk system
}
public final boolean isTickingReady() {
- return this.isTickingReady;
+ return this.newChunkHolder.isTickingReady(); // Paper - rewrite chunk system
}
public final boolean isFullChunkReady() {
- return this.isFullChunkReady;
+ return this.newChunkHolder.isFullChunkReady(); // Paper - rewrite chunk system
}
// Paper end
}
diff --git a/src/main/java/net/minecraft/server/level/ChunkMap.java b/src/main/java/net/minecraft/server/level/ChunkMap.java
2023-03-14 19:05:23 +01:00
index 578849f64de38dc75cdefeb243422c4c89a5c69c..9246ce9f654ad4db6155b026778a83d4ee79624d 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/ChunkMap.java
+++ b/src/main/java/net/minecraft/server/level/ChunkMap.java
2023-03-14 19:05:23 +01:00
@@ -128,10 +128,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public static final int MAX_VIEW_DISTANCE = 33;
public static final int MAX_CHUNK_DISTANCE = 33 + ChunkStatus.maxDistance();
public static final int FORCED_TICKET_LEVEL = 31;
- public final Long2ObjectLinkedOpenHashMap<ChunkHolder> updatingChunkMap = new Long2ObjectLinkedOpenHashMap();
- public volatile Long2ObjectLinkedOpenHashMap<ChunkHolder> visibleChunkMap;
- private final Long2ObjectLinkedOpenHashMap<ChunkHolder> pendingUnloads;
- private final LongSet entitiesInLevel;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper - rewrite chunk system
public final ServerLevel level;
private final ThreadedLevelLightEngine lightEngine;
public final BlockableEventLoop<Runnable> mainThreadExecutor; // Paper - public
2023-03-14 19:05:23 +01:00
@@ -140,16 +137,14 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
2022-12-07 19:32:25 +01:00
private final ChunkGeneratorStructureState chunkGeneratorState;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public final Supplier<DimensionDataStorage> overworldDataStorage;
private final PoiManager poiManager;
- public final LongSet toDrop;
+ // Paper - rewrite chunk system
private boolean modified;
- private final ChunkTaskPriorityQueueSorter queueSorter;
- private final ProcessorHandle<ChunkTaskPriorityQueueSorter.Message<Runnable>> worldgenMailbox;
- private final ProcessorHandle<ChunkTaskPriorityQueueSorter.Message<Runnable>> mainThreadMailbox;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper - rewrite chunk system
public final ChunkProgressListener progressListener;
private final ChunkStatusUpdateListener chunkStatusListener;
public final ChunkMap.ChunkDistanceManager distanceManager;
private final AtomicInteger tickingGenerated;
- private final StructureTemplateManager structureTemplateManager;
+ public final StructureTemplateManager structureTemplateManager; // Paper - rewrite chunk system
private final String storageName;
private final PlayerMap playerMap;
public final Int2ObjectMap<ChunkMap.TrackedEntity> entityMap;
2023-03-14 19:05:23 +01:00
@@ -158,37 +153,21 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private final Queue<Runnable> unloadQueue;
int viewDistance;
- // CraftBukkit start - recursion-safe executor for Chunk loadCallback() and unloadCallback()
- public final CallbackExecutor callbackExecutor = new CallbackExecutor();
- public static final class CallbackExecutor implements java.util.concurrent.Executor, Runnable {
-
- private final java.util.Queue<Runnable> queue = new java.util.ArrayDeque<>();
-
- @Override
- public void execute(Runnable runnable) {
- this.queue.add(runnable);
- }
-
- @Override
- public void run() {
- Runnable task;
- while ((task = this.queue.poll()) != null) {
- task.run();
- }
- }
- };
- // CraftBukkit end
+ // Paper - rewrite chunk system
// Paper start - distance maps
private final com.destroystokyo.paper.util.misc.PooledLinkedHashSets<ServerPlayer> pooledLinkedPlayerHashSets = new com.destroystokyo.paper.util.misc.PooledLinkedHashSets<>();
+ public final io.papermc.paper.chunk.PlayerChunkLoader playerChunkManager = new io.papermc.paper.chunk.PlayerChunkLoader(this, this.pooledLinkedPlayerHashSets); // Paper - replace chunk loader
void addPlayerToDistanceMaps(ServerPlayer player) {
+ this.playerChunkManager.addPlayer(player); // Paper - replace chunk loader
int chunkX = MCUtil.getChunkCoordinate(player.getX());
int chunkZ = MCUtil.getChunkCoordinate(player.getZ());
// Note: players need to be explicitly added to distance maps before they can be updated
}
void removePlayerFromDistanceMaps(ServerPlayer player) {
+ this.playerChunkManager.removePlayer(player); // Paper - replace chunk loader
}
2023-03-14 19:05:23 +01:00
@@ -196,6 +175,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
int chunkX = MCUtil.getChunkCoordinate(player.getX());
int chunkZ = MCUtil.getChunkCoordinate(player.getZ());
// Note: players need to be explicitly added to distance maps before they can be updated
+ this.playerChunkManager.updatePlayer(player); // Paper - replace chunk loader
}
// Paper end
// Paper start
2023-03-14 19:05:23 +01:00
@@ -225,16 +205,13 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public final ChunkHolder getUnloadingChunkHolder(int chunkX, int chunkZ) {
- return this.pendingUnloads.get(io.papermc.paper.util.CoordinateUtils.getChunkKey(chunkX, chunkZ));
+ return null; // Paper - rewrite chunk system
}
// Paper end
public ChunkMap(ServerLevel world, LevelStorageSource.LevelStorageAccess session, DataFixer dataFixer, StructureTemplateManager structureTemplateManager, Executor executor, BlockableEventLoop<Runnable> mainThreadExecutor, LightChunkGetter chunkProvider, ChunkGenerator chunkGenerator, ChunkProgressListener worldGenerationProgressListener, ChunkStatusUpdateListener chunkStatusChangeListener, Supplier<DimensionDataStorage> persistentStateManagerFactory, int viewDistance, boolean dsync) {
super(session.getDimensionPath(world.dimension()).resolve("region"), dataFixer, dsync);
- this.visibleChunkMap = this.updatingChunkMap.clone();
- this.pendingUnloads = new Long2ObjectLinkedOpenHashMap();
- this.entitiesInLevel = new LongOpenHashSet();
- this.toDrop = new LongOpenHashSet();
+ // Paper - rewrite chunk system
this.tickingGenerated = new AtomicInteger();
this.playerMap = new PlayerMap();
this.entityMap = new Int2ObjectOpenHashMap();
2023-03-14 19:05:23 +01:00
@@ -265,19 +242,17 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.chunkGeneratorState = chunkGenerator.createState(iregistrycustom.lookupOrThrow(Registries.STRUCTURE_SET), this.randomState, j, world.spigotConfig); // Spigot
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.mainThreadExecutor = mainThreadExecutor;
- ProcessorMailbox<Runnable> threadedmailbox = ProcessorMailbox.create(executor, "worldgen");
+ // Paper - rewrite chunk system
Objects.requireNonNull(mainThreadExecutor);
- ProcessorHandle<Runnable> mailbox = ProcessorHandle.of("main", mainThreadExecutor::tell);
+ // Paper - rewrite chunk system
this.progressListener = worldGenerationProgressListener;
this.chunkStatusListener = chunkStatusChangeListener;
- ProcessorMailbox<Runnable> threadedmailbox1 = ProcessorMailbox.create(executor, "light");
+ // Paper - rewrite chunk system
- this.queueSorter = new ChunkTaskPriorityQueueSorter(ImmutableList.of(threadedmailbox, mailbox, threadedmailbox1), executor, Integer.MAX_VALUE);
- this.worldgenMailbox = this.queueSorter.getProcessor(threadedmailbox, false);
- this.mainThreadMailbox = this.queueSorter.getProcessor(mailbox, false);
- this.lightEngine = new ThreadedLevelLightEngine(chunkProvider, this, this.level.dimensionType().hasSkyLight(), threadedmailbox1, this.queueSorter.getProcessor(threadedmailbox1, false));
+ // Paper - rewrite chunk system
+ this.lightEngine = new ThreadedLevelLightEngine(chunkProvider, this, this.level.dimensionType().hasSkyLight(), null, null); // Paper - rewrite chunk system
this.distanceManager = new ChunkMap.ChunkDistanceManager(executor, mainThreadExecutor);
this.overworldDataStorage = persistentStateManagerFactory;
2022-12-07 19:32:25 +01:00
this.poiManager = new PoiManager(path.resolve("poi"), dataFixer, dsync, iregistrycustom, world);
2023-03-14 19:05:23 +01:00
@@ -342,20 +317,22 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Nullable
protected ChunkHolder getUpdatingChunkIfPresent(long pos) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- return (ChunkHolder) this.updatingChunkMap.get(pos);
+ // Paper start - rewrite chunk system
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder holder = this.level.chunkTaskScheduler.chunkHolderManager.getChunkHolder(pos);
+ return holder == null ? null : holder.vanillaChunkHolder;
+ // Paper end - rewrite chunk system
}
@Nullable
public ChunkHolder getVisibleChunkIfPresent(long pos) {
- return (ChunkHolder) this.visibleChunkMap.get(pos);
+ // Paper start - rewrite chunk system
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder holder = this.level.chunkTaskScheduler.chunkHolderManager.getChunkHolder(pos);
+ return holder == null ? null : holder.vanillaChunkHolder;
+ // Paper end - rewrite chunk system
}
protected IntSupplier getChunkQueueLevel(long pos) {
- return () -> {
- ChunkHolder playerchunk = this.getVisibleChunkIfPresent(pos);
-
- return playerchunk == null ? ChunkTaskPriorityQueue.PRIORITY_LEVEL_COUNT - 1 : Math.min(playerchunk.getQueueLevel(), ChunkTaskPriorityQueue.PRIORITY_LEVEL_COUNT - 1);
- };
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public String getChunkDebugData(ChunkPos chunkPos) {
2023-03-14 19:05:23 +01:00
@@ -392,75 +369,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// Paper end
private CompletableFuture<Either<List<ChunkAccess>, ChunkHolder.ChunkLoadingFailure>> getChunkRangeFuture(ChunkPos centerChunk, int margin, IntFunction<ChunkStatus> distanceToStatus) {
- List<CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>>> list = new ArrayList();
- List<ChunkHolder> list1 = new ArrayList();
- int j = centerChunk.x;
- int k = centerChunk.z;
-
- for (int l = -margin; l <= margin; ++l) {
- for (int i1 = -margin; i1 <= margin; ++i1) {
- int j1 = Math.max(Math.abs(i1), Math.abs(l));
- final ChunkPos chunkcoordintpair1 = new ChunkPos(j + i1, k + l);
- long k1 = chunkcoordintpair1.toLong();
- ChunkHolder playerchunk = this.getUpdatingChunkIfPresent(k1);
-
- if (playerchunk == null) {
- return CompletableFuture.completedFuture(Either.right(new ChunkHolder.ChunkLoadingFailure() {
- public String toString() {
- return "Unloaded " + chunkcoordintpair1;
- }
- }));
- }
-
- ChunkStatus chunkstatus = (ChunkStatus) distanceToStatus.apply(j1);
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = playerchunk.getOrScheduleFuture(chunkstatus, this);
-
- list1.add(playerchunk);
- list.add(completablefuture);
- }
- }
-
- CompletableFuture<List<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>>> completablefuture1 = Util.sequence(list);
- CompletableFuture<Either<List<ChunkAccess>, ChunkHolder.ChunkLoadingFailure>> completablefuture2 = completablefuture1.thenApply((list2) -> {
- List<ChunkAccess> list3 = Lists.newArrayList();
- // CraftBukkit start - decompile error
- int cnt = 0;
-
- for (Iterator iterator = list2.iterator(); iterator.hasNext(); ++cnt) {
- final int l1 = cnt;
- // CraftBukkit end
- final Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure> either = (Either) iterator.next();
-
- if (either == null) {
- throw this.debugFuturesAndCreateReportedException(new IllegalStateException("At least one of the chunk futures were null"), "n/a");
- }
-
- Optional<ChunkAccess> optional = either.left();
-
- if (!optional.isPresent()) {
- return Either.right(new ChunkHolder.ChunkLoadingFailure() {
- public String toString() {
- ChunkPos chunkcoordintpair2 = new ChunkPos(j + l1 % (margin * 2 + 1), k + l1 / (margin * 2 + 1));
-
- return "Unloaded " + chunkcoordintpair2 + " " + either.right().get();
- }
- });
- }
-
- list3.add((ChunkAccess) optional.get());
- }
-
- return Either.left(list3);
- });
- Iterator iterator = list1.iterator();
-
- while (iterator.hasNext()) {
- ChunkHolder playerchunk1 = (ChunkHolder) iterator.next();
-
- playerchunk1.addSaveDependency("getChunkRangeFuture " + centerChunk + " " + margin, completablefuture2);
- }
-
- return completablefuture2;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public ReportedException debugFuturesAndCreateReportedException(IllegalStateException exception, String details) {
2023-03-14 19:05:23 +01:00
@@ -490,261 +399,72 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> prepareEntityTickingChunk(ChunkPos pos) {
- return this.getChunkRangeFuture(pos, 2, (i) -> {
- return ChunkStatus.FULL;
- }).thenApplyAsync((either) -> {
- return either.mapLeft((list) -> {
- return (LevelChunk) list.get(list.size() / 2);
- });
- }, this.mainThreadExecutor);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@Nullable
ChunkHolder updateChunkScheduling(long pos, int level, @Nullable ChunkHolder holder, int k) {
- if (k > ChunkMap.MAX_CHUNK_DISTANCE && level > ChunkMap.MAX_CHUNK_DISTANCE) {
- return holder;
- } else {
- if (holder != null) {
- holder.setTicketLevel(level);
- }
-
- if (holder != null) {
- if (level > ChunkMap.MAX_CHUNK_DISTANCE) {
- this.toDrop.add(pos);
- } else {
- this.toDrop.remove(pos);
- }
- }
-
- if (level <= ChunkMap.MAX_CHUNK_DISTANCE && holder == null) {
- holder = (ChunkHolder) this.pendingUnloads.remove(pos);
- if (holder != null) {
- holder.setTicketLevel(level);
- } else {
- holder = new ChunkHolder(new ChunkPos(pos), level, this.level, this.lightEngine, this.queueSorter, this);
- // Paper start
- io.papermc.paper.chunk.system.ChunkSystem.onChunkHolderCreate(this.level, holder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- // Paper end
- }
-
- // Paper start
- holder.onChunkAdd();
- // Paper end
- this.updatingChunkMap.put(pos, holder);
- this.modified = true;
- }
-
- return holder;
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@Override
public void close() throws IOException {
- try {
- this.queueSorter.close();
- this.poiManager.close();
- } finally {
- super.close();
- }
+ throw new UnsupportedOperationException("Use ServerChunkCache#close"); // Paper - rewrite chunk system
2023-03-23 22:57:03 +01:00
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - rewrite chunk system
+ protected void saveIncrementally() {
+ this.level.chunkTaskScheduler.chunkHolderManager.autoSave(); // Paper - rewrite chunk system
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper end - - rewrite chunk system
2023-03-23 22:57:03 +01:00
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
protected void saveAllChunks(boolean flush) {
- if (flush) {
- List<ChunkHolder> list = (List) io.papermc.paper.chunk.system.ChunkSystem.getVisibleChunkHolders(this.level).stream().filter(ChunkHolder::wasAccessibleSinceLastSave).peek(ChunkHolder::refreshAccessibility).collect(Collectors.toList()); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- MutableBoolean mutableboolean = new MutableBoolean();
-
- do {
- mutableboolean.setFalse();
- list.stream().map((playerchunk) -> {
- CompletableFuture completablefuture;
-
- do {
- completablefuture = playerchunk.getChunkToSave();
- BlockableEventLoop iasynctaskhandler = this.mainThreadExecutor;
-
- Objects.requireNonNull(completablefuture);
- iasynctaskhandler.managedBlock(completablefuture::isDone);
- } while (completablefuture != playerchunk.getChunkToSave());
-
- return (ChunkAccess) completablefuture.join();
- }).filter((ichunkaccess) -> {
- return ichunkaccess instanceof ImposterProtoChunk || ichunkaccess instanceof LevelChunk;
- }).filter(this::save).forEach((ichunkaccess) -> {
- mutableboolean.setTrue();
- });
- } while (mutableboolean.isTrue());
-
- this.processUnloads(() -> {
- return true;
- });
- this.flushWorker();
- } else {
- io.papermc.paper.chunk.system.ChunkSystem.getVisibleChunkHolders(this.level).forEach(this::saveChunkIfNeeded);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
-
+ this.level.chunkTaskScheduler.chunkHolderManager.saveAllChunks(flush, false, false); // Paper - rewrite chunk system
}
protected void tick(BooleanSupplier shouldKeepTicking) {
ProfilerFiller gameprofilerfiller = this.level.getProfiler();
+ try (Timing ignored = this.level.timings.poiUnload.startTiming()) { // Paper
gameprofilerfiller.push("poi");
this.poiManager.tick(shouldKeepTicking);
+ } // Paper
gameprofilerfiller.popPush("chunk_unload");
if (!this.level.noSave()) {
+ try (Timing ignored = this.level.timings.chunkUnload.startTiming()) { // Paper
this.processUnloads(shouldKeepTicking);
+ } // Paper
}
gameprofilerfiller.pop();
}
public boolean hasWork() {
- return this.lightEngine.hasLightWork() || !this.pendingUnloads.isEmpty() || io.papermc.paper.chunk.system.ChunkSystem.hasAnyChunkHolders(this.level) || this.poiManager.hasWork() || !this.toDrop.isEmpty() || !this.unloadQueue.isEmpty() || this.queueSorter.hasWork() || this.distanceManager.hasTickets(); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
private void processUnloads(BooleanSupplier shouldKeepTicking) {
- LongIterator longiterator = this.toDrop.iterator();
-
- for (int i = 0; longiterator.hasNext() && (shouldKeepTicking.getAsBoolean() || i < 200 || this.toDrop.size() > 2000); longiterator.remove()) {
- long j = longiterator.nextLong();
- ChunkHolder playerchunk = (ChunkHolder) this.updatingChunkMap.remove(j);
-
- if (playerchunk != null) {
- playerchunk.onChunkRemove(); // Paper
- this.pendingUnloads.put(j, playerchunk);
- this.modified = true;
- ++i;
- this.scheduleUnload(j, playerchunk);
- }
- }
-
- int k = Math.max(0, this.unloadQueue.size() - 2000);
-
- Runnable runnable;
-
- while ((shouldKeepTicking.getAsBoolean() || k > 0) && (runnable = (Runnable) this.unloadQueue.poll()) != null) {
- --k;
- runnable.run();
- }
-
- int l = 0;
- Iterator objectiterator = io.papermc.paper.chunk.system.ChunkSystem.getVisibleChunkHolders(this.level).iterator(); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
- while (l < 20 && shouldKeepTicking.getAsBoolean() && objectiterator.hasNext()) {
- if (this.saveChunkIfNeeded((ChunkHolder) objectiterator.next())) {
- ++l;
- }
- }
+ this.level.chunkTaskScheduler.chunkHolderManager.processUnloads(); // Paper - rewrite chunk system
}
private void scheduleUnload(long pos, ChunkHolder holder) {
- CompletableFuture<ChunkAccess> completablefuture = holder.getChunkToSave();
- Consumer<ChunkAccess> consumer = (ichunkaccess) -> { // CraftBukkit - decompile error
- CompletableFuture<ChunkAccess> completablefuture1 = holder.getChunkToSave();
-
- if (completablefuture1 != completablefuture) {
- this.scheduleUnload(pos, holder);
- } else {
- // Paper start
- boolean removed;
- if ((removed = this.pendingUnloads.remove(pos, holder)) && ichunkaccess != null) {
- io.papermc.paper.chunk.system.ChunkSystem.onChunkHolderDelete(this.level, holder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- // Paper end
- if (ichunkaccess instanceof LevelChunk) {
- ((LevelChunk) ichunkaccess).setLoaded(false);
- }
-
- this.save(ichunkaccess);
- if (this.entitiesInLevel.remove(pos) && ichunkaccess instanceof LevelChunk) {
- LevelChunk chunk = (LevelChunk) ichunkaccess;
-
- this.level.unload(chunk);
- }
-
- this.lightEngine.updateChunkStatus(ichunkaccess.getPos());
- this.lightEngine.tryScheduleUpdate();
- this.progressListener.onStatusChange(ichunkaccess.getPos(), (ChunkStatus) null);
- this.chunkSaveCooldowns.remove(ichunkaccess.getPos().toLong());
- } else if (removed) { // Paper start
- io.papermc.paper.chunk.system.ChunkSystem.onChunkHolderDelete(this.level, holder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- } // Paper end
-
- }
- };
- Queue queue = this.unloadQueue;
-
- Objects.requireNonNull(this.unloadQueue);
- completablefuture.thenAcceptAsync(consumer, queue::add).whenComplete((ovoid, throwable) -> {
- if (throwable != null) {
- ChunkMap.LOGGER.error("Failed to save chunk {}", holder.getPos(), throwable);
- }
-
- });
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
protected boolean promoteChunkMap() {
- if (!this.modified) {
- return false;
- } else {
- this.visibleChunkMap = this.updatingChunkMap.clone();
- this.modified = false;
- return true;
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> schedule(ChunkHolder holder, ChunkStatus requiredStatus) {
- ChunkPos chunkcoordintpair = holder.getPos();
-
- if (requiredStatus == ChunkStatus.EMPTY) {
- return this.scheduleChunkLoad(chunkcoordintpair);
- } else {
- if (requiredStatus == ChunkStatus.LIGHT) {
- this.distanceManager.addTicket(TicketType.LIGHT, chunkcoordintpair, 33 + ChunkStatus.getDistance(ChunkStatus.LIGHT), chunkcoordintpair);
- }
-
- Optional<ChunkAccess> optional = ((Either) holder.getOrScheduleFuture(requiredStatus.getParent(), this).getNow(ChunkHolder.UNLOADED_CHUNK)).left();
-
- if (optional.isPresent() && ((ChunkAccess) optional.get()).getStatus().isOrAfter(requiredStatus)) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = requiredStatus.load(this.level, this.structureTemplateManager, this.lightEngine, (ichunkaccess) -> {
- return this.protoChunkToFullChunk(holder);
- }, (ChunkAccess) optional.get());
-
- this.progressListener.onStatusChange(chunkcoordintpair, requiredStatus);
- return completablefuture;
- } else {
- return this.scheduleChunkGeneration(holder, requiredStatus);
- }
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
private CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> scheduleChunkLoad(ChunkPos pos) {
- return this.readChunk(pos).thenApply((optional) -> {
- return optional.filter((nbttagcompound) -> {
- boolean flag = ChunkMap.isChunkDataValid(nbttagcompound);
-
- if (!flag) {
- ChunkMap.LOGGER.error("Chunk file at {} is missing level data, skipping", pos);
- }
-
- return flag;
- });
- }).thenApplyAsync((optional) -> {
- this.level.getProfiler().incrementCounter("chunkLoad");
- if (optional.isPresent()) {
- ProtoChunk protochunk = ChunkSerializer.read(this.level, this.poiManager, pos, (CompoundTag) optional.get());
-
- this.markPosition(pos, protochunk.getStatus().getChunkType());
- return Either.<ChunkAccess, ChunkHolder.ChunkLoadingFailure>left(protochunk); // CraftBukkit - decompile error
- } else {
- return Either.<ChunkAccess, ChunkHolder.ChunkLoadingFailure>left(this.createEmptyChunk(pos)); // CraftBukkit - decompile error
- }
- }, this.mainThreadExecutor).exceptionallyAsync((throwable) -> {
- return this.handleChunkLoadFailure(throwable, pos);
- }, this.mainThreadExecutor);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
- private static boolean isChunkDataValid(CompoundTag nbt) {
+ public static boolean isChunkDataValid(CompoundTag nbt) { // Paper - async chunk loading
return nbt.contains("Status", 8);
}
2023-03-14 19:05:23 +01:00
@@ -780,45 +500,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> scheduleChunkGeneration(ChunkHolder holder, ChunkStatus requiredStatus) {
- ChunkPos chunkcoordintpair = holder.getPos();
- CompletableFuture<Either<List<ChunkAccess>, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getChunkRangeFuture(chunkcoordintpair, requiredStatus.getRange(), (i) -> {
- return this.getDependencyStatus(requiredStatus, i);
- });
-
- this.level.getProfiler().incrementCounter(() -> {
- return "chunkGenerate " + requiredStatus.getName();
- });
- Executor executor = (runnable) -> {
- this.worldgenMailbox.tell(ChunkTaskPriorityQueueSorter.message(holder, runnable));
- };
-
- return completablefuture.thenComposeAsync((either) -> {
- return (CompletionStage) either.map((list) -> {
- try {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture1 = requiredStatus.generate(executor, this.level, this.generator, this.structureTemplateManager, this.lightEngine, (ichunkaccess) -> {
- return this.protoChunkToFullChunk(holder);
- }, list, false);
-
- this.progressListener.onStatusChange(chunkcoordintpair, requiredStatus);
- return completablefuture1;
- } catch (Exception exception) {
- exception.getStackTrace();
- CrashReport crashreport = CrashReport.forThrowable(exception, "Exception generating new chunk");
- CrashReportCategory crashreportsystemdetails = crashreport.addCategory("Chunk to be generated");
-
- crashreportsystemdetails.setDetail("Location", (Object) String.format(Locale.ROOT, "%d,%d", chunkcoordintpair.x, chunkcoordintpair.z));
- crashreportsystemdetails.setDetail("Position hash", (Object) ChunkPos.asLong(chunkcoordintpair.x, chunkcoordintpair.z));
- crashreportsystemdetails.setDetail("Generator", (Object) this.generator);
- this.mainThreadExecutor.execute(() -> {
- throw new ReportedException(crashreport);
- });
- throw new ReportedException(crashreport);
- }
- }, (playerchunk_failure) -> {
- this.releaseLightTicket(chunkcoordintpair);
- return CompletableFuture.completedFuture(Either.right(playerchunk_failure));
- });
- }, executor);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
protected void releaseLightTicket(ChunkPos pos) {
2023-03-14 19:05:23 +01:00
@@ -829,7 +511,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}));
}
- private ChunkStatus getDependencyStatus(ChunkStatus centerChunkTargetStatus, int distance) {
+ public static ChunkStatus getDependencyStatus(ChunkStatus centerChunkTargetStatus, int distance) { // Paper -> public, static
ChunkStatus chunkstatus1;
if (distance == 0) {
2023-03-14 19:05:23 +01:00
@@ -841,7 +523,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return chunkstatus1;
}
- private static void postLoadProtoChunk(ServerLevel world, List<CompoundTag> nbt) {
+ public static void postLoadProtoChunk(ServerLevel world, List<CompoundTag> nbt) { // Paper - public
if (!nbt.isEmpty()) {
// CraftBukkit start - these are spawned serialized (DefinedStructure) and we don't call an add event below at the moment due to ordering complexities
world.addWorldGenChunkEntities(EntityType.loadEntitiesRecursive(nbt, world).filter((entity) -> {
2023-03-14 19:05:23 +01:00
@@ -863,91 +545,15 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> protoChunkToFullChunk(ChunkHolder chunkHolder) {
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = chunkHolder.getFutureIfPresentUnchecked(ChunkStatus.FULL.getParent());
-
- return completablefuture.thenApplyAsync((either) -> {
- ChunkStatus chunkstatus = ChunkHolder.getStatus(chunkHolder.getTicketLevel());
-
- return !chunkstatus.isOrAfter(ChunkStatus.FULL) ? ChunkHolder.UNLOADED_CHUNK : either.mapLeft((ichunkaccess) -> {
- try (Timing ignored = level.timings.chunkPostLoad.startTimingIfSync()) { // Paper
- ChunkPos chunkcoordintpair = chunkHolder.getPos();
- ProtoChunk protochunk = (ProtoChunk) ichunkaccess;
- LevelChunk chunk;
-
- if (protochunk instanceof ImposterProtoChunk) {
- chunk = ((ImposterProtoChunk) protochunk).getWrapped();
- } else {
- chunk = new LevelChunk(this.level, protochunk, (chunk1) -> {
- ChunkMap.postLoadProtoChunk(this.level, protochunk.getEntities());
- });
- chunkHolder.replaceProtoChunk(new ImposterProtoChunk(chunk, false));
- }
-
- chunk.setFullStatus(() -> {
- return ChunkHolder.getFullChunkStatus(chunkHolder.getTicketLevel());
- });
- chunk.runPostLoad();
- if (this.entitiesInLevel.add(chunkcoordintpair.toLong())) {
- chunk.setLoaded(true);
- chunk.registerAllBlockEntitiesAfterLevelLoad();
- chunk.registerTickContainerInLevel(this.level);
- }
-
- return chunk;
- } // Paper
- });
- }, (runnable) -> {
- ProcessorHandle mailbox = this.mainThreadMailbox;
- long i = chunkHolder.getPos().toLong();
-
- Objects.requireNonNull(chunkHolder);
- mailbox.tell(ChunkTaskPriorityQueueSorter.message(runnable, i, chunkHolder::getTicketLevel));
- });
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> prepareTickingChunk(ChunkHolder holder) {
- ChunkPos chunkcoordintpair = holder.getPos();
- CompletableFuture<Either<List<ChunkAccess>, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getChunkRangeFuture(chunkcoordintpair, 1, (i) -> {
- return ChunkStatus.FULL;
- });
- CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> completablefuture1 = completablefuture.thenApplyAsync((either) -> {
- return either.mapLeft((list) -> {
- return (LevelChunk) list.get(list.size() / 2);
- });
- }, (runnable) -> {
- this.mainThreadMailbox.tell(ChunkTaskPriorityQueueSorter.message(holder, runnable));
- }).thenApplyAsync((either) -> {
- return either.ifLeft((chunk) -> {
- chunk.postProcessGeneration();
- this.level.startTickingChunk(chunk);
- });
- }, this.mainThreadExecutor);
-
- completablefuture1.thenAcceptAsync((either) -> {
- either.ifLeft((chunk) -> {
- this.tickingGenerated.getAndIncrement();
- MutableObject<ClientboundLevelChunkWithLightPacket> mutableobject = new MutableObject();
-
- this.getPlayers(chunkcoordintpair, false).forEach((entityplayer) -> {
- this.playerLoadedChunk(entityplayer, mutableobject, chunk);
- });
- });
- }, (runnable) -> {
- this.mainThreadMailbox.tell(ChunkTaskPriorityQueueSorter.message(holder, runnable));
- });
- return completablefuture1;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> prepareAccessibleChunk(ChunkHolder holder) {
- return this.getChunkRangeFuture(holder.getPos(), 1, ChunkStatus::getStatusAroundFullChunk).thenApplyAsync((either) -> {
- return either.mapLeft((list) -> {
- LevelChunk chunk = (LevelChunk) list.get(list.size() / 2);
-
- return chunk;
- });
- }, (runnable) -> {
- this.mainThreadMailbox.tell(ChunkTaskPriorityQueueSorter.message(holder, runnable));
- });
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public int getTickingGenerated() {
2023-03-14 19:05:23 +01:00
@@ -955,94 +561,22 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private boolean saveChunkIfNeeded(ChunkHolder chunkHolder) {
- if (!chunkHolder.wasAccessibleSinceLastSave()) {
- return false;
- } else {
- ChunkAccess ichunkaccess = (ChunkAccess) chunkHolder.getChunkToSave().getNow(null); // CraftBukkit - decompile error
-
- if (!(ichunkaccess instanceof ImposterProtoChunk) && !(ichunkaccess instanceof LevelChunk)) {
- return false;
- } else {
- long i = ichunkaccess.getPos().toLong();
- long j = this.chunkSaveCooldowns.getOrDefault(i, -1L);
- long k = System.currentTimeMillis();
-
- if (k < j) {
- return false;
- } else {
- boolean flag = this.save(ichunkaccess);
-
- chunkHolder.refreshAccessibility();
- if (flag) {
- this.chunkSaveCooldowns.put(i, k + 10000L);
- }
-
- return flag;
- }
- }
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public boolean save(ChunkAccess chunk) {
- this.poiManager.flush(chunk.getPos());
- if (!chunk.isUnsaved()) {
- return false;
- } else {
- chunk.setUnsaved(false);
- ChunkPos chunkcoordintpair = chunk.getPos();
-
- try {
- ChunkStatus chunkstatus = chunk.getStatus();
-
- if (chunkstatus.getChunkType() != ChunkStatus.ChunkType.LEVELCHUNK) {
- if (this.isExistingChunkFull(chunkcoordintpair)) {
- return false;
- }
-
- if (chunkstatus == ChunkStatus.EMPTY && chunk.getAllStarts().values().stream().noneMatch(StructureStart::isValid)) {
- return false;
- }
- }
-
- this.level.getProfiler().incrementCounter("chunkSave");
- CompoundTag nbttagcompound = ChunkSerializer.write(this.level, chunk);
-
- this.write(chunkcoordintpair, nbttagcompound);
- this.markPosition(chunkcoordintpair, chunkstatus.getChunkType());
- return true;
- } catch (Exception exception) {
- ChunkMap.LOGGER.error("Failed to save chunk {},{}", new Object[]{chunkcoordintpair.x, chunkcoordintpair.z, exception});
- return false;
- }
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
private boolean isExistingChunkFull(ChunkPos pos) {
- byte b0 = this.chunkTypeCache.get(pos.toLong());
-
- if (b0 != 0) {
- return b0 == 1;
- } else {
- CompoundTag nbttagcompound;
-
- try {
- nbttagcompound = (CompoundTag) ((Optional) this.readChunk(pos).join()).orElse((Object) null);
- if (nbttagcompound == null) {
- this.markPositionReplaceable(pos);
- return false;
- }
- } catch (Exception exception) {
- ChunkMap.LOGGER.error("Failed to read chunk {}", pos, exception);
- this.markPositionReplaceable(pos);
- return false;
- }
-
- ChunkStatus.ChunkType chunkstatus_type = ChunkSerializer.getChunkTypeFromTag(nbttagcompound);
-
- return this.markPosition(pos, chunkstatus_type) == 1;
- }
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
+ // Paper start - replace player loader system
+ public void setTickViewDistance(int distance) {
+ this.playerChunkManager.setTickDistance(distance);
+ }
+ // Paper end - replace player loader system
public void setViewDistance(int watchDistance) {
2023-03-14 19:05:23 +01:00
int j = Mth.clamp(watchDistance + 1, 3, 33);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
2023-03-14 19:05:23 +01:00
@@ -1050,33 +584,18 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
int k = this.viewDistance;
this.viewDistance = j;
- this.distanceManager.updatePlayerTickets(this.viewDistance + 1);
- Iterator objectiterator = io.papermc.paper.chunk.system.ChunkSystem.getUpdatingChunkHolders(this.level).iterator(); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
- while (objectiterator.hasNext()) {
- ChunkHolder playerchunk = (ChunkHolder) objectiterator.next();
- ChunkPos chunkcoordintpair = playerchunk.getPos();
- MutableObject<ClientboundLevelChunkWithLightPacket> mutableobject = new MutableObject();
-
- this.getPlayers(chunkcoordintpair, false).forEach((entityplayer) -> {
- SectionPos sectionposition = entityplayer.getLastSectionPos();
- boolean flag = ChunkMap.isChunkInRange(chunkcoordintpair.x, chunkcoordintpair.z, sectionposition.x(), sectionposition.z(), k);
- boolean flag1 = ChunkMap.isChunkInRange(chunkcoordintpair.x, chunkcoordintpair.z, sectionposition.x(), sectionposition.z(), this.viewDistance);
-
- this.updateChunkTracking(entityplayer, chunkcoordintpair, mutableobject, flag, flag1);
- });
- }
+ this.playerChunkManager.setLoadDistance(this.viewDistance); // Paper - replace player loader system
}
}
- protected void updateChunkTracking(ServerPlayer player, ChunkPos pos, MutableObject<ClientboundLevelChunkWithLightPacket> packet, boolean oldWithinViewDistance, boolean newWithinViewDistance) {
+ public void updateChunkTracking(ServerPlayer player, ChunkPos pos, MutableObject<ClientboundLevelChunkWithLightPacket> packet, boolean oldWithinViewDistance, boolean newWithinViewDistance) { // Paper - public
if (player.level == this.level) {
if (newWithinViewDistance && !oldWithinViewDistance) {
ChunkHolder playerchunk = this.getVisibleChunkIfPresent(pos.toLong());
if (playerchunk != null) {
- LevelChunk chunk = playerchunk.getTickingChunk();
+ LevelChunk chunk = playerchunk.getSendingChunk(); // Paper - replace chunk loader system
if (chunk != null) {
this.playerLoadedChunk(player, packet, chunk);
2023-03-14 19:05:23 +01:00
@@ -1106,30 +625,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
void dumpChunks(Writer writer) throws IOException {
- CsvOutput csvwriter = CsvOutput.builder().addColumn("x").addColumn("z").addColumn("level").addColumn("in_memory").addColumn("status").addColumn("full_status").addColumn("accessible_ready").addColumn("ticking_ready").addColumn("entity_ticking_ready").addColumn("ticket").addColumn("spawning").addColumn("block_entity_count").addColumn("ticking_ticket").addColumn("ticking_level").addColumn("block_ticks").addColumn("fluid_ticks").build(writer);
- TickingTracker tickingtracker = this.distanceManager.tickingTracker();
- Iterator<ChunkHolder> objectbidirectionaliterator = io.papermc.paper.chunk.system.ChunkSystem.getVisibleChunkHolders(this.level).iterator(); // Paper
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
- while (objectbidirectionaliterator.hasNext()) {
- ChunkHolder playerchunk = objectbidirectionaliterator.next(); // Paper
- long i = playerchunk.pos.toLong(); // Paper
- ChunkPos chunkcoordintpair = new ChunkPos(i);
- // Paper
- Optional<ChunkAccess> optional = Optional.ofNullable(playerchunk.getLastAvailable());
- Optional<LevelChunk> optional1 = optional.flatMap((ichunkaccess) -> {
- return ichunkaccess instanceof LevelChunk ? Optional.of((LevelChunk) ichunkaccess) : Optional.empty();
- });
-
- // CraftBukkit - decompile error
- csvwriter.writeRow(chunkcoordintpair.x, chunkcoordintpair.z, playerchunk.getTicketLevel(), optional.isPresent(), optional.map(ChunkAccess::getStatus).orElse(null), optional1.map(LevelChunk::getFullStatus).orElse(null), ChunkMap.printFuture(playerchunk.getFullChunkFuture()), ChunkMap.printFuture(playerchunk.getTickingChunkFuture()), ChunkMap.printFuture(playerchunk.getEntityTickingChunkFuture()), this.distanceManager.getTicketDebugString(i), this.anyPlayerCloseEnoughForSpawning(chunkcoordintpair), optional1.map((chunk) -> {
- return chunk.getBlockEntities().size();
- }).orElse(0), tickingtracker.getTicketDebugString(i), tickingtracker.getLevel(i), optional1.map((chunk) -> {
- return chunk.getBlockTicks().count();
- }).orElse(0), optional1.map((chunk) -> {
- return chunk.getFluidTicks().count();
- }).orElse(0));
- }
-
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
private static String printFuture(CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> future) {
2023-03-14 19:05:23 +01:00
@@ -1148,6 +644,35 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
}
+ // Paper start - Asynchronous chunk io
+ @Nullable
+ @Override
+ public CompoundTag readSync(ChunkPos chunkcoordintpair) throws IOException {
+ // Paper start - rewrite chunk system
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ return io.papermc.paper.chunk.system.io.RegionFileIOThread.loadData(
+ this.level, chunkcoordintpair.x, chunkcoordintpair.z, io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.getIOBlockingPriorityForCurrentThread()
+ );
+ }
+ // Paper end - rewrite chunk system
+ return super.readSync(chunkcoordintpair);
+ }
+
+ @Override
+ public void write(ChunkPos chunkcoordintpair, CompoundTag nbttagcompound) throws IOException {
+ // Paper start - rewrite chunk system
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.scheduleSave(
+ this.level, chunkcoordintpair.x, chunkcoordintpair.z, nbttagcompound,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA);
+ return;
+ }
+ // Paper end - rewrite chunk system
+ super.write(chunkcoordintpair, nbttagcompound);
+ }
+ // Paper end
+
private CompletableFuture<Optional<CompoundTag>> readChunk(ChunkPos chunkPos) {
return this.read(chunkPos).thenApplyAsync((optional) -> {
return optional.map((nbttagcompound) -> this.upgradeChunkTag(nbttagcompound, chunkPos)); // CraftBukkit
2023-03-14 19:05:23 +01:00
@@ -1251,15 +776,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.removePlayerFromDistanceMaps(player); // Paper - distance maps
}
- for (int k = i - this.viewDistance - 1; k <= i + this.viewDistance + 1; ++k) {
- for (int l = j - this.viewDistance - 1; l <= j + this.viewDistance + 1; ++l) {
- if (ChunkMap.isChunkInRange(k, l, i, j, this.viewDistance)) {
- ChunkPos chunkcoordintpair = new ChunkPos(k, l);
-
- this.updateChunkTracking(player, chunkcoordintpair, new MutableObject(), !added, added);
- }
- }
- }
+ // Paper - handled by player chunk loader
}
2023-03-14 19:05:23 +01:00
@@ -1267,7 +784,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
SectionPos sectionposition = SectionPos.of((EntityAccess) player);
player.setLastSectionPos(sectionposition);
- player.connection.send(new ClientboundSetChunkCacheCenterPacket(sectionposition.x(), sectionposition.z()));
+ //player.connection.send(new ClientboundSetChunkCacheCenterPacket(sectionposition.x(), sectionposition.z())); // Paper - handled by player chunk loader
return sectionposition;
}
2023-03-14 19:05:23 +01:00
@@ -1322,65 +839,38 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
int k1;
int l1;
- if (Math.abs(i1 - i) <= this.viewDistance * 2 && Math.abs(j1 - j) <= this.viewDistance * 2) {
- k1 = Math.min(i, i1) - this.viewDistance - 1;
- l1 = Math.min(j, j1) - this.viewDistance - 1;
- int i2 = Math.max(i, i1) + this.viewDistance + 1;
- int j2 = Math.max(j, j1) + this.viewDistance + 1;
-
- for (int k2 = k1; k2 <= i2; ++k2) {
- for (int l2 = l1; l2 <= j2; ++l2) {
- boolean flag3 = ChunkMap.isChunkInRange(k2, l2, i1, j1, this.viewDistance);
- boolean flag4 = ChunkMap.isChunkInRange(k2, l2, i, j, this.viewDistance);
-
- this.updateChunkTracking(player, new ChunkPos(k2, l2), new MutableObject(), flag3, flag4);
- }
- }
- } else {
- boolean flag5;
- boolean flag6;
-
- for (k1 = i1 - this.viewDistance - 1; k1 <= i1 + this.viewDistance + 1; ++k1) {
- for (l1 = j1 - this.viewDistance - 1; l1 <= j1 + this.viewDistance + 1; ++l1) {
- if (ChunkMap.isChunkInRange(k1, l1, i1, j1, this.viewDistance)) {
- flag5 = true;
- flag6 = false;
- this.updateChunkTracking(player, new ChunkPos(k1, l1), new MutableObject(), true, false);
- }
- }
- }
-
- for (k1 = i - this.viewDistance - 1; k1 <= i + this.viewDistance + 1; ++k1) {
- for (l1 = j - this.viewDistance - 1; l1 <= j + this.viewDistance + 1; ++l1) {
- if (ChunkMap.isChunkInRange(k1, l1, i, j, this.viewDistance)) {
- flag5 = false;
- flag6 = true;
- this.updateChunkTracking(player, new ChunkPos(k1, l1), new MutableObject(), false, true);
- }
- }
- }
- }
+ // Paper - replaced by PlayerChunkLoader
this.updateMaps(player); // Paper - distance maps
+ this.playerChunkManager.updatePlayer(player); // Paper - respond to movement immediately
}
@Override
public List<ServerPlayer> getPlayers(ChunkPos chunkPos, boolean onlyOnWatchDistanceEdge) {
- Set<ServerPlayer> set = this.playerMap.getPlayers(chunkPos.toLong());
- Builder<ServerPlayer> builder = ImmutableList.builder();
- Iterator iterator = set.iterator();
+ // Paper start - per player view distance
+ // there can be potential desync with player's last mapped section and the view distance map, so use the
+ // view distance map here.
+ com.destroystokyo.paper.util.misc.PooledLinkedHashSets.PooledObjectLinkedOpenHashSet<ServerPlayer> players = this.playerChunkManager.broadcastMap.getObjectsInRange(chunkPos);
+ if (players == null) {
+ return java.util.Collections.emptyList();
2023-03-23 22:57:03 +01:00
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
2023-03-23 22:57:03 +01:00
- while (iterator.hasNext()) {
- ServerPlayer entityplayer = (ServerPlayer) iterator.next();
- SectionPos sectionposition = entityplayer.getLastSectionPos();
+ List<ServerPlayer> ret = new java.util.ArrayList<>(players.size());
2023-03-23 22:57:03 +01:00
- if (onlyOnWatchDistanceEdge && ChunkMap.isChunkOnRangeBorder(chunkPos.x, chunkPos.z, sectionposition.x(), sectionposition.z(), this.viewDistance) || !onlyOnWatchDistanceEdge && ChunkMap.isChunkInRange(chunkPos.x, chunkPos.z, sectionposition.x(), sectionposition.z(), this.viewDistance)) {
- builder.add(entityplayer);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ Object[] backingSet = players.getBackingSet();
+ for (int i = 0, len = backingSet.length; i < len; ++i) {
+ if (!(backingSet[i] instanceof ServerPlayer player)) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ continue;
+ }
+ if (!this.playerChunkManager.isChunkSent(player, chunkPos.x, chunkPos.z, onlyOnWatchDistanceEdge)) {
+ continue;
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ret.add(player);
2023-03-23 22:57:03 +01:00
}
- return builder.build();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ return ret;
+ // Paper end - per player view distance
}
public void addEntity(Entity entity) {
2023-03-14 19:05:23 +01:00
@@ -1608,7 +1098,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
protected boolean isChunkToRemove(long pos) {
- return ChunkMap.this.toDrop.contains(pos);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@Nullable
2023-03-14 19:05:23 +01:00
@@ -1689,7 +1179,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
org.spigotmc.AsyncCatcher.catchOp("player tracker update"); // Spigot
if (player != this.entity) {
Vec3 vec3d = player.position().subtract(this.entity.position());
- double d0 = (double) Math.min(this.getEffectiveRange(), (ChunkMap.this.viewDistance - 1) * 16);
+ double d0 = (double) Math.min(this.getEffectiveRange(), io.papermc.paper.chunk.PlayerChunkLoader.getSendViewDistance(player) * 16); // Paper - per player view distance
double d1 = vec3d.x * vec3d.x + vec3d.z * vec3d.z;
double d2 = d0 * d0;
boolean flag = d1 <= d2 && this.entity.broadcastToPlayer(player);
diff --git a/src/main/java/net/minecraft/server/level/DistanceManager.java b/src/main/java/net/minecraft/server/level/DistanceManager.java
index d38ad1b1eee92a6dbd2b79b4fcdb8959cdb4007d..ffa1e457decf8502c3283352bf5be94d419ff165 100644
--- a/src/main/java/net/minecraft/server/level/DistanceManager.java
+++ b/src/main/java/net/minecraft/server/level/DistanceManager.java
@@ -40,6 +40,12 @@ import org.slf4j.Logger;
public abstract class DistanceManager {
+ // Paper start - rewrite chunk system
+ public io.papermc.paper.chunk.system.scheduling.ChunkHolderManager getChunkHolderManager() {
+ return this.chunkMap.level.chunkTaskScheduler.chunkHolderManager;
+ }
+ // Paper end - rewrite chunk system
+
static final Logger LOGGER = LogUtils.getLogger();
private static final int ENTITY_TICKING_RANGE = 2;
static final int PLAYER_TICKET_LEVEL = 33 + ChunkStatus.getDistance(ChunkStatus.FULL) - 2;
@@ -47,61 +53,20 @@ public abstract class DistanceManager {
private static final int ENTITY_TICKING_LEVEL_THRESHOLD = 32;
private static final int BLOCK_TICKING_LEVEL_THRESHOLD = 33;
final Long2ObjectMap<ObjectSet<ServerPlayer>> playersPerChunk = new Long2ObjectOpenHashMap();
- public final Long2ObjectOpenHashMap<SortedArraySet<Ticket<?>>> tickets = new Long2ObjectOpenHashMap();
- private final DistanceManager.ChunkTicketTracker ticketTracker = new DistanceManager.ChunkTicketTracker();
+ // Paper - rewrite chunk system
private final DistanceManager.FixedPlayerDistanceChunkTracker naturalSpawnChunkCounter = new DistanceManager.FixedPlayerDistanceChunkTracker(8);
- private final TickingTracker tickingTicketsTracker = new TickingTracker();
- private final DistanceManager.PlayerTicketTracker playerTicketManager = new DistanceManager.PlayerTicketTracker(33);
- final Set<ChunkHolder> chunksToUpdateFutures = Sets.newHashSet();
- final ChunkTaskPriorityQueueSorter ticketThrottler;
- final ProcessorHandle<ChunkTaskPriorityQueueSorter.Message<Runnable>> ticketThrottlerInput;
- final ProcessorHandle<ChunkTaskPriorityQueueSorter.Release> ticketThrottlerReleaser;
- final LongSet ticketsToRelease = new LongOpenHashSet();
- final Executor mainThreadExecutor;
- private long ticketTickCounter;
- private int simulationDistance = 10;
+ //private final TickingTracker tickingTicketsTracker = new TickingTracker(); // Paper - no longer used
+ //private final DistanceManager.PlayerTicketTracker playerTicketManager = new DistanceManager.PlayerTicketTracker(33); // Paper - no longer used
+ // Paper - rewrite chunk system
private final ChunkMap chunkMap; // Paper
protected DistanceManager(Executor workerExecutor, Executor mainThreadExecutor, ChunkMap chunkMap) {
- Objects.requireNonNull(mainThreadExecutor);
- ProcessorHandle<Runnable> mailbox = ProcessorHandle.of("player ticket throttler", mainThreadExecutor::execute);
- ChunkTaskPriorityQueueSorter chunktaskqueuesorter = new ChunkTaskPriorityQueueSorter(ImmutableList.of(mailbox), workerExecutor, 4);
-
- this.ticketThrottler = chunktaskqueuesorter;
- this.ticketThrottlerInput = chunktaskqueuesorter.getProcessor(mailbox, true);
- this.ticketThrottlerReleaser = chunktaskqueuesorter.getReleaseProcessor(mailbox);
- this.mainThreadExecutor = mainThreadExecutor;
+ // Paper - rewrite chunk system
this.chunkMap = chunkMap; // Paper
}
protected void purgeStaleTickets() {
- ++this.ticketTickCounter;
- ObjectIterator objectiterator = this.tickets.long2ObjectEntrySet().fastIterator();
-
- while (objectiterator.hasNext()) {
- Entry<SortedArraySet<Ticket<?>>> entry = (Entry) objectiterator.next();
- Iterator<Ticket<?>> iterator = ((SortedArraySet) entry.getValue()).iterator();
- boolean flag = false;
-
- while (iterator.hasNext()) {
- Ticket<?> ticket = (Ticket) iterator.next();
-
- if (ticket.timedOut(this.ticketTickCounter)) {
- iterator.remove();
- flag = true;
- this.tickingTicketsTracker.removeTicket(entry.getLongKey(), ticket);
- }
- }
-
- if (flag) {
- this.ticketTracker.update(entry.getLongKey(), DistanceManager.getTicketLevelAt((SortedArraySet) entry.getValue()), false);
- }
-
- if (((SortedArraySet) entry.getValue()).isEmpty()) {
- objectiterator.remove();
- }
- }
-
+ this.getChunkHolderManager().tick(); // Paper - rewrite chunk system
}
private static int getTicketLevelAt(SortedArraySet<Ticket<?>> tickets) {
@@ -117,108 +82,25 @@ public abstract class DistanceManager {
protected abstract ChunkHolder updateChunkScheduling(long pos, int level, @Nullable ChunkHolder holder, int k);
public boolean runAllUpdates(ChunkMap chunkStorage) {
- this.naturalSpawnChunkCounter.runAllUpdates();
- this.tickingTicketsTracker.runAllUpdates();
- this.playerTicketManager.runAllUpdates();
- int i = Integer.MAX_VALUE - this.ticketTracker.runDistanceUpdates(Integer.MAX_VALUE);
- boolean flag = i != 0;
-
- if (flag) {
- ;
- }
-
- if (!this.chunksToUpdateFutures.isEmpty()) {
- // CraftBukkit start
- // Iterate pending chunk updates with protection against concurrent modification exceptions
- java.util.Iterator<ChunkHolder> iter = this.chunksToUpdateFutures.iterator();
- int expectedSize = this.chunksToUpdateFutures.size();
- do {
- ChunkHolder playerchunk = iter.next();
- iter.remove();
- expectedSize--;
-
- playerchunk.updateFutures(chunkStorage, this.mainThreadExecutor);
-
- // Reset iterator if set was modified using add()
- if (this.chunksToUpdateFutures.size() != expectedSize) {
- expectedSize = this.chunksToUpdateFutures.size();
- iter = this.chunksToUpdateFutures.iterator();
- }
- } while (iter.hasNext());
- // CraftBukkit end
-
- return true;
- } else {
- if (!this.ticketsToRelease.isEmpty()) {
- LongIterator longiterator = this.ticketsToRelease.iterator();
-
- while (longiterator.hasNext()) {
- long j = longiterator.nextLong();
-
- if (this.getTickets(j).stream().anyMatch((ticket) -> {
- return ticket.getType() == TicketType.PLAYER;
- })) {
- ChunkHolder playerchunk = chunkStorage.getUpdatingChunkIfPresent(j);
-
- if (playerchunk == null) {
- throw new IllegalStateException();
- }
-
- CompletableFuture<Either<LevelChunk, ChunkHolder.ChunkLoadingFailure>> completablefuture = playerchunk.getEntityTickingChunkFuture();
-
- completablefuture.thenAccept((either) -> {
- this.mainThreadExecutor.execute(() -> {
- this.ticketThrottlerReleaser.tell(ChunkTaskPriorityQueueSorter.release(() -> {
- }, j, false));
- });
- });
- }
- }
-
- this.ticketsToRelease.clear();
- }
-
- return flag;
- }
+ return this.getChunkHolderManager().processTicketUpdates(); // Paper - rewrite chunk system
}
boolean addTicket(long i, Ticket<?> ticket) { // CraftBukkit - void -> boolean
- SortedArraySet<Ticket<?>> arraysetsorted = this.getTickets(i);
- int j = DistanceManager.getTicketLevelAt(arraysetsorted);
- Ticket<?> ticket1 = (Ticket) arraysetsorted.addOrGet(ticket);
-
- ticket1.setCreatedTick(this.ticketTickCounter);
- if (ticket.getTicketLevel() < j) {
- this.ticketTracker.update(i, ticket.getTicketLevel(), true);
- }
-
- return ticket == ticket1; // CraftBukkit
+ org.spigotmc.AsyncCatcher.catchOp("ChunkMapDistance::addTicket"); // Paper
+ return this.getChunkHolderManager().addTicketAtLevel((TicketType)ticket.getType(), i, ticket.getTicketLevel(), ticket.key); // Paper - rewrite chunk system
}
boolean removeTicket(long i, Ticket<?> ticket) { // CraftBukkit - void -> boolean
- SortedArraySet<Ticket<?>> arraysetsorted = this.getTickets(i);
-
- boolean removed = false; // CraftBukkit
- if (arraysetsorted.remove(ticket)) {
- removed = true; // CraftBukkit
- }
-
- if (arraysetsorted.isEmpty()) {
- this.tickets.remove(i);
- }
-
- this.ticketTracker.update(i, DistanceManager.getTicketLevelAt(arraysetsorted), false);
- return removed; // CraftBukkit
+ org.spigotmc.AsyncCatcher.catchOp("ChunkMapDistance::removeTicket"); // Paper
+ return this.getChunkHolderManager().removeTicketAtLevel((TicketType)ticket.getType(), i, ticket.getTicketLevel(), ticket.key); // Paper - rewrite chunk system
}
public <T> void addTicket(TicketType<T> type, ChunkPos pos, int level, T argument) {
- this.addTicket(pos.toLong(), new Ticket<>(type, level, argument));
+ this.getChunkHolderManager().addTicketAtLevel(type, pos, level, argument); // Paper - rewrite chunk system
}
public <T> void removeTicket(TicketType<T> type, ChunkPos pos, int level, T argument) {
- Ticket<T> ticket = new Ticket<>(type, level, argument);
-
- this.removeTicket(pos.toLong(), ticket);
+ this.getChunkHolderManager().removeTicketAtLevel(type, pos, level, argument); // Paper - rewrite chunk system
}
public <T> void addRegionTicket(TicketType<T> type, ChunkPos pos, int radius, T argument) {
@@ -227,13 +109,7 @@ public abstract class DistanceManager {
}
public <T> boolean addRegionTicketAtDistance(TicketType<T> tickettype, ChunkPos chunkcoordintpair, int i, T t0) {
- // CraftBukkit end
- Ticket<T> ticket = new Ticket<>(tickettype, 33 - i, t0);
- long j = chunkcoordintpair.toLong();
-
- boolean added = this.addTicket(j, ticket); // CraftBukkit
- this.tickingTicketsTracker.addTicket(j, ticket);
- return added; // CraftBukkit
+ return this.getChunkHolderManager().addTicketAtLevel(tickettype, chunkcoordintpair, 33 - i, t0); // Paper - rewrite chunk system
}
public <T> void removeRegionTicket(TicketType<T> type, ChunkPos pos, int radius, T argument) {
@@ -242,31 +118,21 @@ public abstract class DistanceManager {
}
public <T> boolean removeRegionTicketAtDistance(TicketType<T> tickettype, ChunkPos chunkcoordintpair, int i, T t0) {
- // CraftBukkit end
- Ticket<T> ticket = new Ticket<>(tickettype, 33 - i, t0);
- long j = chunkcoordintpair.toLong();
-
- boolean removed = this.removeTicket(j, ticket); // CraftBukkit
- this.tickingTicketsTracker.removeTicket(j, ticket);
- return removed; // CraftBukkit
+ return this.getChunkHolderManager().removeTicketAtLevel(tickettype, chunkcoordintpair, 33 - i, t0); // Paper - rewrite chunk system
}
- private SortedArraySet<Ticket<?>> getTickets(long position) {
- return (SortedArraySet) this.tickets.computeIfAbsent(position, (j) -> {
- return SortedArraySet.create(4);
- });
- }
+ // Paper - rewrite chunk system
protected void updateChunkForced(ChunkPos pos, boolean forced) {
- Ticket<ChunkPos> ticket = new Ticket<>(TicketType.FORCED, 31, pos);
+ Ticket<ChunkPos> ticket = new Ticket<>(TicketType.FORCED, 31, pos, 0L); // Paper - rewrite chunk system
long i = pos.toLong();
if (forced) {
this.addTicket(i, ticket);
- this.tickingTicketsTracker.addTicket(i, ticket);
+ //this.tickingTicketsTracker.addTicket(i, ticket); // Paper - no longer used
} else {
this.removeTicket(i, ticket);
- this.tickingTicketsTracker.removeTicket(i, ticket);
+ //this.tickingTicketsTracker.removeTicket(i, ticket); // Paper - no longer used
}
}
@@ -275,12 +141,10 @@ public abstract class DistanceManager {
ChunkPos chunkcoordintpair = pos.chunk();
long i = chunkcoordintpair.toLong();
- ((ObjectSet) this.playersPerChunk.computeIfAbsent(i, (j) -> {
- return new ObjectOpenHashSet();
- })).add(player);
+ // Paper - no longer used
this.naturalSpawnChunkCounter.update(i, 0, true);
- this.playerTicketManager.update(i, 0, true);
- this.tickingTicketsTracker.addTicket(TicketType.PLAYER, chunkcoordintpair, this.getPlayerTicketLevel(), chunkcoordintpair);
+ //this.playerTicketManager.update(i, 0, true); // Paper - no longer used
+ //this.tickingTicketsTracker.addTicket(TicketType.PLAYER, chunkcoordintpair, this.getPlayerTicketLevel(), chunkcoordintpair); // Paper - no longer used
}
public void removePlayer(SectionPos pos, ServerPlayer player) {
@@ -293,46 +157,44 @@ public abstract class DistanceManager {
if (objectset.isEmpty()) {
this.playersPerChunk.remove(i);
this.naturalSpawnChunkCounter.update(i, Integer.MAX_VALUE, false);
- this.playerTicketManager.update(i, Integer.MAX_VALUE, false);
- this.tickingTicketsTracker.removeTicket(TicketType.PLAYER, chunkcoordintpair, this.getPlayerTicketLevel(), chunkcoordintpair);
+ //this.playerTicketManager.update(i, Integer.MAX_VALUE, false); // Paper - no longer used
+ //this.tickingTicketsTracker.removeTicket(TicketType.PLAYER, chunkcoordintpair, this.getPlayerTicketLevel(), chunkcoordintpair); // Paper - no longer used
}
}
- private int getPlayerTicketLevel() {
- return Math.max(0, 31 - this.simulationDistance);
- }
+ // Paper - rewrite chunk system
public boolean inEntityTickingRange(long chunkPos) {
- return this.tickingTicketsTracker.getLevel(chunkPos) < 32;
+ // Paper start - replace player chunk loader system
+ ChunkHolder holder = this.chunkMap.getVisibleChunkIfPresent(chunkPos);
+ return holder != null && holder.isEntityTickingReady();
+ // Paper end - replace player chunk loader system
}
public boolean inBlockTickingRange(long chunkPos) {
- return this.tickingTicketsTracker.getLevel(chunkPos) < 33;
+ // Paper start - replace player chunk loader system
+ ChunkHolder holder = this.chunkMap.getVisibleChunkIfPresent(chunkPos);
+ return holder != null && holder.isTickingReady();
+ // Paper end - replace player chunk loader system
}
protected String getTicketDebugString(long pos) {
- SortedArraySet<Ticket<?>> arraysetsorted = (SortedArraySet) this.tickets.get(pos);
-
- return arraysetsorted != null && !arraysetsorted.isEmpty() ? ((Ticket) arraysetsorted.first()).toString() : "no_ticket";
+ return this.getChunkHolderManager().getTicketDebugString(pos); // Paper - rewrite chunk system
}
protected void updatePlayerTickets(int viewDistance) {
- this.playerTicketManager.updateViewDistance(viewDistance);
+ this.chunkMap.playerChunkManager.setTargetNoTickViewDistance(viewDistance); // Paper - route to player chunk manager
}
// Paper start
public int getSimulationDistance() {
- return this.simulationDistance;
+ return this.chunkMap.playerChunkManager.getTargetTickViewDistance(); // Paper - route to player chunk manager
}
// Paper end
public void updateSimulationDistance(int simulationDistance) {
- if (simulationDistance != this.simulationDistance) {
- this.simulationDistance = simulationDistance;
- this.tickingTicketsTracker.replacePlayerTicketsLevel(this.getPlayerTicketLevel());
- }
-
+ this.chunkMap.playerChunkManager.setTargetTickViewDistance(simulationDistance); // Paper - route to player chunk manager
}
public int getNaturalSpawnChunkCount() {
@@ -346,103 +208,28 @@ public abstract class DistanceManager {
}
public String getDebugStatus() {
- return this.ticketThrottler.getDebugStatus();
+ return "No DistanceManager stats available"; // Paper - rewrite chunk system
}
- private void dumpTickets(String path) {
- try {
- FileOutputStream fileoutputstream = new FileOutputStream(new File(path));
2023-03-23 22:57:03 +01:00
-
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- try {
- ObjectIterator objectiterator = this.tickets.long2ObjectEntrySet().iterator();
-
- while (objectiterator.hasNext()) {
- Entry<SortedArraySet<Ticket<?>>> entry = (Entry) objectiterator.next();
- ChunkPos chunkcoordintpair = new ChunkPos(entry.getLongKey());
- Iterator iterator = ((SortedArraySet) entry.getValue()).iterator();
-
- while (iterator.hasNext()) {
- Ticket<?> ticket = (Ticket) iterator.next();
-
- fileoutputstream.write((chunkcoordintpair.x + "\t" + chunkcoordintpair.z + "\t" + ticket.getType() + "\t" + ticket.getTicketLevel() + "\t\n").getBytes(StandardCharsets.UTF_8));
- }
- }
- } catch (Throwable throwable) {
- try {
- fileoutputstream.close();
- } catch (Throwable throwable1) {
- throwable.addSuppressed(throwable1);
- }
2023-03-23 22:57:03 +01:00
+ // Paper - rewrite chunk system
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- throw throwable;
- }
-
- fileoutputstream.close();
- } catch (IOException ioexception) {
- DistanceManager.LOGGER.error("Failed to dump tickets to {}", path, ioexception);
- }
-
- }
-
- @VisibleForTesting
- TickingTracker tickingTracker() {
- return this.tickingTicketsTracker;
- }
+ // Paper - replace player chunk loader
public void removeTicketsOnClosing() {
- ImmutableSet<TicketType<?>> immutableset = ImmutableSet.of(TicketType.UNKNOWN, TicketType.POST_TELEPORT, TicketType.LIGHT, TicketType.FUTURE_AWAIT, TicketType.CHUNK_RELIGHT, ca.spottedleaf.starlight.common.light.StarLightInterface.CHUNK_WORK_TICKET); // Paper - add additional tickets to preserve
- ObjectIterator objectiterator = this.tickets.long2ObjectEntrySet().fastIterator();
-
- while (objectiterator.hasNext()) {
- Entry<SortedArraySet<Ticket<?>>> entry = (Entry) objectiterator.next();
- Iterator<Ticket<?>> iterator = ((SortedArraySet) entry.getValue()).iterator();
- boolean flag = false;
-
- while (iterator.hasNext()) {
- Ticket<?> ticket = (Ticket) iterator.next();
-
- if (!immutableset.contains(ticket.getType())) {
- iterator.remove();
- flag = true;
- this.tickingTicketsTracker.removeTicket(entry.getLongKey(), ticket);
- }
- }
-
- if (flag) {
- this.ticketTracker.update(entry.getLongKey(), DistanceManager.getTicketLevelAt((SortedArraySet) entry.getValue()), false);
- }
-
- if (((SortedArraySet) entry.getValue()).isEmpty()) {
- objectiterator.remove();
- }
- }
-
+ // Paper - rewrite chunk system - this stupid hack ain't needed anymore
}
public boolean hasTickets() {
- return !this.tickets.isEmpty();
+ return this.getChunkHolderManager().hasTickets(); // Paper - rewrite chunk system
}
// CraftBukkit start
public <T> void removeAllTicketsFor(TicketType<T> ticketType, int ticketLevel, T ticketIdentifier) {
- Ticket<T> target = new Ticket<>(ticketType, ticketLevel, ticketIdentifier);
-
- for (java.util.Iterator<Entry<SortedArraySet<Ticket<?>>>> iterator = this.tickets.long2ObjectEntrySet().fastIterator(); iterator.hasNext();) {
- Entry<SortedArraySet<Ticket<?>>> entry = iterator.next();
- SortedArraySet<Ticket<?>> tickets = entry.getValue();
- if (tickets.remove(target)) {
- // copied from removeTicket
- this.ticketTracker.update(entry.getLongKey(), DistanceManager.getTicketLevelAt(tickets), false);
-
- // can't use entry after it's removed
- if (tickets.isEmpty()) {
- iterator.remove();
- }
- }
- }
+ this.getChunkHolderManager().removeAllTicketsFor(ticketType, ticketLevel, ticketIdentifier); // Paper - rewrite chunk system
}
// CraftBukkit end
+ /* Paper - rewrite chunk system
private class ChunkTicketTracker extends ChunkTracker {
public ChunkTicketTracker() {
@@ -487,6 +274,7 @@ public abstract class DistanceManager {
return this.runUpdates(distance);
}
}
+ */ // Paper - rewrite chunk system
private class FixedPlayerDistanceChunkTracker extends ChunkTracker {
@@ -566,6 +354,7 @@ public abstract class DistanceManager {
}
}
+ /* Paper - rewrite chunk system
private class PlayerTicketTracker extends DistanceManager.FixedPlayerDistanceChunkTracker {
private int viewDistance = 0;
@@ -661,4 +450,5 @@ public abstract class DistanceManager {
return distance <= this.viewDistance - 2;
}
}
+ */ // Paper - rewrite chunk system
}
diff --git a/src/main/java/net/minecraft/server/level/ServerChunkCache.java b/src/main/java/net/minecraft/server/level/ServerChunkCache.java
2022-12-07 19:32:25 +01:00
index 28c8a3ba1caddf0ea334a6ef43cae25f982743e4..80d108ae7faf3fdcb024931e93032215935fe70b 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/ServerChunkCache.java
+++ b/src/main/java/net/minecraft/server/level/ServerChunkCache.java
2022-12-07 19:32:25 +01:00
@@ -367,7 +367,7 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public LevelChunk getChunkAtIfLoadedImmediately(int x, int z) {
long k = ChunkPos.asLong(x, z);
- if (Thread.currentThread() == this.mainThread) {
+ if (io.papermc.paper.util.TickThread.isTickThread()) { // Paper - rewrite chunk system
return this.getChunkAtIfLoadedMainThread(x, z);
}
2022-12-07 19:32:25 +01:00
@@ -389,11 +389,34 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return ret;
}
// Paper end
+ // Paper start - async chunk io
+ public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getChunkAtAsynchronously(int x, int z, boolean gen, boolean isUrgent) {
+ CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> ret = new CompletableFuture<>();
+
+ ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority priority;
+ if (isUrgent) {
+ priority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.HIGHER;
+ } else {
+ priority = ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.NORMAL;
+ }
+
+ io.papermc.paper.chunk.system.ChunkSystem.scheduleChunkLoad(this.level, x, z, gen, ChunkStatus.FULL, true, priority, (chunk) -> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ if (chunk == null) {
+ ret.complete(ChunkHolder.UNLOADED_CHUNK);
+ } else {
+ ret.complete(Either.left(chunk));
+ }
+ });
+
+ return ret;
+ }
+ // Paper end - async chunk io
@Nullable
@Override
public ChunkAccess getChunk(int x, int z, ChunkStatus leastStatus, boolean create) {
- if (Thread.currentThread() != this.mainThread) {
+ final int x1 = x; final int z1 = z; // Paper - conflict on variable change
+ if (!io.papermc.paper.util.TickThread.isTickThread()) { // Paper - rewrite chunk system
return (ChunkAccess) CompletableFuture.supplyAsync(() -> {
return this.getChunk(x, z, leastStatus, create);
}, this.mainThreadProcessor).join();
2022-12-07 19:32:25 +01:00
@@ -405,23 +428,20 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
ChunkAccess ichunkaccess;
- for (int l = 0; l < 4; ++l) {
- if (k == this.lastChunkPos[l] && leastStatus == this.lastChunkStatus[l]) {
- ichunkaccess = this.lastChunk[l];
- if (ichunkaccess != null) { // CraftBukkit - the chunk can become accessible in the meantime TODO for non-null chunks it might also make sense to check that the chunk's state hasn't changed in the meantime
- return ichunkaccess;
- }
- }
- }
+ // Paper - rewrite chunk system - there are no correct callbacks to remove items from cache in the new chunk system
gameprofilerfiller.incrementCounter("getChunkCacheMiss");
- CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getChunkFutureMainThread(x, z, leastStatus, create);
+ CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> completablefuture = this.getChunkFutureMainThread(x, z, leastStatus, create, true); // Paper
ServerChunkCache.MainThreadExecutor chunkproviderserver_b = this.mainThreadProcessor;
Objects.requireNonNull(completablefuture);
if (!completablefuture.isDone()) { // Paper
+ // Paper start - async chunk io/loading
+ io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.pushChunkWait(this.level, x1, z1); // Paper - rewrite chunk system
+ // Paper end
this.level.timings.syncChunkLoad.startTiming(); // Paper
chunkproviderserver_b.managedBlock(completablefuture::isDone);
+ io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.popChunkWait(); // Paper - async chunk debug // Paper - rewrite chunk system
this.level.timings.syncChunkLoad.stopTiming(); // Paper
} // Paper
ichunkaccess = (ChunkAccess) ((Either) completablefuture.join()).map((ichunkaccess1) -> {
2022-12-07 19:32:25 +01:00
@@ -441,7 +461,7 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Nullable
@Override
public LevelChunk getChunkNow(int chunkX, int chunkZ) {
- if (Thread.currentThread() != this.mainThread) {
+ if (!io.papermc.paper.util.TickThread.isTickThread()) { // Paper - rewrite chunk system
return null;
} else {
this.level.getProfiler().incrementCounter("getChunkNow");
2022-12-07 19:32:25 +01:00
@@ -487,7 +507,7 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getChunkFuture(int chunkX, int chunkZ, ChunkStatus leastStatus, boolean create) {
- boolean flag1 = Thread.currentThread() == this.mainThread;
+ boolean flag1 = io.papermc.paper.util.TickThread.isTickThread(); // Paper - rewrite chunk system
CompletableFuture completablefuture;
if (flag1) {
2022-12-07 19:32:25 +01:00
@@ -508,47 +528,52 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getChunkFutureMainThread(int chunkX, int chunkZ, ChunkStatus leastStatus, boolean create) {
- ChunkPos chunkcoordintpair = new ChunkPos(chunkX, chunkZ);
- long k = chunkcoordintpair.toLong();
- int l = 33 + ChunkStatus.getDistance(leastStatus);
- ChunkHolder playerchunk = this.getVisibleChunkIfPresent(k);
+ // Paper start - add isUrgent - old sig left in place for dirty nms plugins
+ return getChunkFutureMainThread(chunkX, chunkZ, leastStatus, create, false);
+ }
+ private CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> getChunkFutureMainThread(int chunkX, int chunkZ, ChunkStatus leastStatus, boolean create, boolean isUrgent) {
+ // Paper start - rewrite chunk system
+ io.papermc.paper.util.TickThread.ensureTickThread(this.level, chunkX, chunkZ, "Scheduling chunk load off-main");
+ int minLevel = 33 + ChunkStatus.getDistance(leastStatus);
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder = this.level.chunkTaskScheduler.chunkHolderManager.getChunkHolder(chunkX, chunkZ);
- // CraftBukkit start - don't add new ticket for currently unloading chunk
- boolean currentlyUnloading = false;
- if (playerchunk != null) {
- ChunkHolder.FullChunkStatus oldChunkState = ChunkHolder.getFullChunkStatus(playerchunk.oldTicketLevel);
- ChunkHolder.FullChunkStatus currentChunkState = ChunkHolder.getFullChunkStatus(playerchunk.getTicketLevel());
- currentlyUnloading = (oldChunkState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER) && !currentChunkState.isOrAfter(ChunkHolder.FullChunkStatus.BORDER));
2023-03-23 22:57:03 +01:00
- }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- if (create && !currentlyUnloading) {
- // CraftBukkit end
- this.distanceManager.addTicket(TicketType.UNKNOWN, chunkcoordintpair, l, chunkcoordintpair);
- if (this.chunkAbsent(playerchunk, l)) {
- ProfilerFiller gameprofilerfiller = this.level.getProfiler();
2023-03-23 22:57:03 +01:00
-
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- gameprofilerfiller.push("chunkLoad");
- this.runDistanceManagerUpdates();
- playerchunk = this.getVisibleChunkIfPresent(k);
- gameprofilerfiller.pop();
- if (this.chunkAbsent(playerchunk, l)) {
- throw (IllegalStateException) Util.pauseInIde(new IllegalStateException("No chunk holder after ticket has been added"));
2023-03-23 22:57:03 +01:00
- }
- }
+ boolean needsFullScheduling = leastStatus == ChunkStatus.FULL && (chunkHolder == null || !chunkHolder.getChunkStatus().isOrAfter(ChunkHolder.FullChunkStatus.BORDER));
+
+ if ((chunkHolder == null || chunkHolder.getTicketLevel() > minLevel || needsFullScheduling) && !create) {
+ return ChunkHolder.UNLOADED_CHUNK_FUTURE;
}
- return this.chunkAbsent(playerchunk, l) ? ChunkHolder.UNLOADED_CHUNK_FUTURE : playerchunk.getOrScheduleFuture(leastStatus, this.chunkMap);
- }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder.ChunkCompletion chunkCompletion = chunkHolder == null ? null : chunkHolder.getLastChunkCompletion();
+ if (needsFullScheduling || chunkCompletion == null || !chunkCompletion.genStatus().isOrAfter(leastStatus)) {
+ // schedule
+ CompletableFuture<Either<ChunkAccess, ChunkHolder.ChunkLoadingFailure>> ret = new CompletableFuture<>();
+ Consumer<ChunkAccess> complete = (ChunkAccess chunk) -> {
+ if (chunk == null) {
+ ret.complete(Either.right(ChunkHolder.ChunkLoadingFailure.UNLOADED));
+ } else {
+ ret.complete(Either.left(chunk));
2023-03-23 22:57:03 +01:00
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ };
2023-03-23 22:57:03 +01:00
- private boolean chunkAbsent(@Nullable ChunkHolder holder, int maxLevel) {
- return holder == null || holder.oldTicketLevel > maxLevel; // CraftBukkit using oldTicketLevel for isLoaded checks
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ this.level.chunkTaskScheduler.scheduleChunkLoad(
+ chunkX, chunkZ, leastStatus, true,
+ isUrgent ? ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.BLOCKING : ca.spottedleaf.concurrentutil.executor.standard.PrioritisedExecutor.Priority.NORMAL,
+ complete
+ );
+
+ return ret;
+ } else {
+ // can return now
+ return CompletableFuture.completedFuture(Either.left(chunkCompletion.chunk()));
2023-03-23 22:57:03 +01:00
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper end - rewrite chunk system
}
+ // Paper - rewrite chunk system
2023-03-23 22:57:03 +01:00
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public boolean hasChunk(int x, int z) {
- ChunkHolder playerchunk = this.getVisibleChunkIfPresent((new ChunkPos(x, z)).toLong());
- int k = 33 + ChunkStatus.getDistance(ChunkStatus.FULL);
-
- return !this.chunkAbsent(playerchunk, k);
+ return this.getChunkAtIfLoadedImmediately(x, z) != null; // Paper - rewrite chunk system
}
@Override
2022-12-07 19:32:25 +01:00
@@ -559,22 +584,13 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
if (playerchunk == null) {
return null;
} else {
- int l = ServerChunkCache.CHUNK_STATUSES.size() - 1;
-
- while (true) {
- ChunkStatus chunkstatus = (ChunkStatus) ServerChunkCache.CHUNK_STATUSES.get(l);
- Optional<ChunkAccess> optional = ((Either) playerchunk.getFutureIfPresentUnchecked(chunkstatus).getNow(ChunkHolder.UNLOADED_CHUNK)).left();
-
- if (optional.isPresent()) {
- return (BlockGetter) optional.get();
- }
-
- if (chunkstatus == ChunkStatus.LIGHT.getParent()) {
- return null;
- }
-
- --l;
+ // Paper start - rewrite chunk system
+ ChunkStatus status = playerchunk.getChunkHolderStatus();
+ if (status != null && !status.isOrAfter(ChunkStatus.LIGHT.getParent())) {
+ return null;
}
+ return playerchunk.getAvailableChunkNow();
+ // Paper end - rewrite chunk system
}
}
2022-12-07 19:32:25 +01:00
@@ -588,15 +604,7 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
boolean runDistanceManagerUpdates() {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- boolean flag = this.distanceManager.runAllUpdates(this.chunkMap);
- boolean flag1 = this.chunkMap.promoteChunkMap();
-
- if (!flag && !flag1) {
- return false;
- } else {
- this.clearCache();
- return true;
- }
+ return this.level.chunkTaskScheduler.chunkHolderManager.processTicketUpdates(); // Paper - rewrite chunk system
}
// Paper start
2022-12-07 19:32:25 +01:00
@@ -606,17 +614,10 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// Paper end
public boolean isPositionTicking(long pos) {
- ChunkHolder playerchunk = this.getVisibleChunkIfPresent(pos);
-
- if (playerchunk == null) {
- return false;
- } else if (!this.level.shouldTickBlocksAt(pos)) {
- return false;
- } else {
- Either<LevelChunk, ChunkHolder.ChunkLoadingFailure> either = (Either) playerchunk.getTickingChunkFuture().getNow(null); // CraftBukkit - decompile error
-
- return either != null && either.left().isPresent();
- }
+ // Paper start - replace player chunk loader system
+ ChunkHolder holder = this.chunkMap.getVisibleChunkIfPresent(pos);
+ return holder != null && holder.isTickingReady();
+ // Paper end - replace player chunk loader system
}
public void save(boolean flush) {
2022-12-07 19:32:25 +01:00
@@ -632,17 +633,13 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.close(true);
}
- public void close(boolean save) throws IOException {
- if (save) {
- this.save(true);
- }
- // CraftBukkit end
- this.lightEngine.close();
- this.chunkMap.close();
+ public void close(boolean save) { // Paper - rewrite chunk system
+ this.level.chunkTaskScheduler.chunkHolderManager.close(save, true); // Paper - rewrite chunk system
}
// CraftBukkit start - modelled on below
public void purgeUnload() {
+ if (true) return; // Paper - tickets will be removed later, this behavior isn't really well accounted for by the chunk system
this.level.getProfiler().push("purge");
this.distanceManager.purgeStaleTickets();
this.runDistanceManagerUpdates();
2022-12-07 19:32:25 +01:00
@@ -663,6 +660,7 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.level.getProfiler().popPush("chunks");
if (tickChunks) {
this.level.timings.chunks.startTiming(); // Paper - timings
+ this.chunkMap.playerChunkManager.tick(); // Paper - this is mostly is to account for view distance changes
this.tickChunks();
this.level.timings.chunks.stopTiming(); // Paper - timings
}
2022-12-07 19:32:25 +01:00
@@ -759,7 +757,12 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
ChunkHolder playerchunk = this.getVisibleChunkIfPresent(pos);
if (playerchunk != null) {
- ((Either) playerchunk.getFullChunkFuture().getNow(ChunkHolder.UNLOADED_LEVEL_CHUNK)).left().ifPresent(chunkConsumer);
+ // Paper start - rewrite chunk system
+ LevelChunk chunk = playerchunk.getFullChunk();
+ if (chunk != null) {
+ chunkConsumer.accept(chunk);
+ }
+ // Paper end - rewrite chunk system
}
}
2022-12-07 19:32:25 +01:00
@@ -925,17 +928,11 @@ public class ServerChunkCache extends ChunkSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
// CraftBukkit start - process pending Chunk loadCallback() and unloadCallback() after each run task
public boolean pollTask() {
- try {
+ ServerChunkCache.this.chunkMap.playerChunkManager.tickMidTick();
if (ServerChunkCache.this.runDistanceManagerUpdates()) {
return true;
- } else {
- ServerChunkCache.this.lightEngine.tryScheduleUpdate();
- return super.pollTask();
}
- } finally {
- chunkMap.callbackExecutor.run();
- }
- // CraftBukkit end
+ return super.pollTask() | ServerChunkCache.this.level.chunkTaskScheduler.executeMainThreadTask(); // Paper - rewrite chunk system
}
}
diff --git a/src/main/java/net/minecraft/server/level/ServerLevel.java b/src/main/java/net/minecraft/server/level/ServerLevel.java
2023-03-14 19:05:23 +01:00
index f2798373e7d07cc0e46c39297c29e7be364a8dff..fb7930009db2c219347e6601ca2cde0c8601e2b4 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/ServerLevel.java
+++ b/src/main/java/net/minecraft/server/level/ServerLevel.java
2023-03-14 19:05:23 +01:00
@@ -191,7 +191,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private final MinecraftServer server;
public final PrimaryLevelData serverLevelData; // CraftBukkit - type
final EntityTickList entityTickList;
- public final PersistentEntitySectionManager<Entity> entityManager;
+ //public final PersistentEntitySectionManager<Entity> entityManager; // Paper - rewrite chunk system
2022-12-07 19:32:25 +01:00
private final GameEventDispatcher gameEventDispatcher;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public boolean noSave;
private final SleepStatus sleepStatus;
2023-03-14 19:05:23 +01:00
@@ -316,7 +316,108 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
}
}
- // Paper end
+
+ // Paper start - rewrite chunk system
+ public final io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler chunkTaskScheduler;
+ public final io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController chunkDataControllerNew
+ = new io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.CHUNK_DATA) {
+
+ @Override
+ public net.minecraft.world.level.chunk.storage.RegionFileStorage getCache() {
+ return ServerLevel.this.getChunkSource().chunkMap.regionFileCache;
+ }
+
+ @Override
+ public void writeData(int chunkX, int chunkZ, net.minecraft.nbt.CompoundTag compound) throws IOException {
+ ServerLevel.this.getChunkSource().chunkMap.write(new ChunkPos(chunkX, chunkZ), compound);
+ }
+
+ @Override
+ public net.minecraft.nbt.CompoundTag readData(int chunkX, int chunkZ) throws IOException {
+ return ServerLevel.this.getChunkSource().chunkMap.readSync(new ChunkPos(chunkX, chunkZ));
+ }
+ };
+ public final io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController poiDataControllerNew
+ = new io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA) {
+
+ @Override
+ public net.minecraft.world.level.chunk.storage.RegionFileStorage getCache() {
+ return ServerLevel.this.getChunkSource().chunkMap.getPoiManager();
+ }
+
+ @Override
+ public void writeData(int chunkX, int chunkZ, net.minecraft.nbt.CompoundTag compound) throws IOException {
+ ServerLevel.this.getChunkSource().chunkMap.getPoiManager().write(new ChunkPos(chunkX, chunkZ), compound);
+ }
+
+ @Override
+ public net.minecraft.nbt.CompoundTag readData(int chunkX, int chunkZ) throws IOException {
+ return ServerLevel.this.getChunkSource().chunkMap.getPoiManager().read(new ChunkPos(chunkX, chunkZ));
+ }
+ };
+ public final io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController entityDataControllerNew
+ = new io.papermc.paper.chunk.system.io.RegionFileIOThread.ChunkDataController(io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.ENTITY_DATA) {
+
+ @Override
+ public net.minecraft.world.level.chunk.storage.RegionFileStorage getCache() {
+ return ServerLevel.this.entityStorage;
+ }
+
+ @Override
+ public void writeData(int chunkX, int chunkZ, net.minecraft.nbt.CompoundTag compound) throws IOException {
+ ServerLevel.this.writeEntityChunk(chunkX, chunkZ, compound);
+ }
+
+ @Override
+ public net.minecraft.nbt.CompoundTag readData(int chunkX, int chunkZ) throws IOException {
+ return ServerLevel.this.readEntityChunk(chunkX, chunkZ);
+ }
+ };
+ private final EntityRegionFileStorage entityStorage;
+
+ private static final class EntityRegionFileStorage extends net.minecraft.world.level.chunk.storage.RegionFileStorage {
+
+ public EntityRegionFileStorage(Path directory, boolean dsync) {
+ super(directory, dsync);
+ }
+
+ protected void write(ChunkPos pos, net.minecraft.nbt.CompoundTag nbt) throws IOException {
+ ChunkPos nbtPos = nbt == null ? null : EntityStorage.readChunkPos(nbt);
+ if (nbtPos != null && !pos.equals(nbtPos)) {
+ throw new IllegalArgumentException(
+ "Entity chunk coordinate and serialized data do not have matching coordinates, trying to serialize coordinate " + pos.toString()
+ + " but compound says coordinate is " + nbtPos + " for world: " + this
+ );
+ }
+ super.write(pos, nbt);
+ }
+ }
+
+ private void writeEntityChunk(int chunkX, int chunkZ, net.minecraft.nbt.CompoundTag compound) throws IOException {
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.scheduleSave(
+ this, chunkX, chunkZ, compound,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.ENTITY_DATA);
+ return;
+ }
+ this.entityStorage.write(new ChunkPos(chunkX, chunkZ), compound);
+ }
+
+ private net.minecraft.nbt.CompoundTag readEntityChunk(int chunkX, int chunkZ) throws IOException {
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ return io.papermc.paper.chunk.system.io.RegionFileIOThread.loadData(
+ this, chunkX, chunkZ, io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.ENTITY_DATA,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.getIOBlockingPriorityForCurrentThread()
+ );
+ }
+ return this.entityStorage.read(new ChunkPos(chunkX, chunkZ));
+ }
+
+ private final io.papermc.paper.chunk.system.entity.EntityLookup entityLookup;
+ public final io.papermc.paper.chunk.system.entity.EntityLookup getEntityLookup() {
+ return this.entityLookup;
+ }
+ // Paper end - rewrite chunk system
// Add env and gen to constructor, IWorldDataServer -> WorldDataServer
public ServerLevel(MinecraftServer minecraftserver, Executor executor, LevelStorageSource.LevelStorageAccess convertable_conversionsession, PrimaryLevelData iworlddataserver, ResourceKey<Level> resourcekey, LevelStem worlddimension, ChunkProgressListener worldloadlistener, boolean flag, long i, List<CustomSpawner> list, boolean flag1, org.bukkit.World.Environment env, org.bukkit.generator.ChunkGenerator gen, org.bukkit.generator.BiomeProvider biomeProvider) {
@@ -360,16 +461,16 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// CraftBukkit end
boolean flag2 = minecraftserver.forceSynchronousWrites();
DataFixer datafixer = minecraftserver.getFixerUpper();
- EntityPersistentStorage<Entity> entitypersistentstorage = new EntityStorage(this, convertable_conversionsession.getDimensionPath(resourcekey).resolve("entities"), datafixer, flag2, minecraftserver);
+ this.entityStorage = new EntityRegionFileStorage(convertable_conversionsession.getDimensionPath(resourcekey).resolve("entities"), flag2); // Paper - rewrite chunk system //EntityPersistentStorage<Entity> entitypersistentstorage = new EntityStorage(this, convertable_conversionsession.getDimensionPath(resourcekey).resolve("entities"), datafixer, flag2, minecraftserver);
- this.entityManager = new PersistentEntitySectionManager<>(Entity.class, new ServerLevel.EntityCallbacks(), entitypersistentstorage);
+ // this.entityManager = new PersistentEntitySectionManager<>(Entity.class, new ServerLevel.EntityCallbacks(), entitypersistentstorage, this.entitySliceManager); // Paper // Paper - rewrite chunk system
StructureTemplateManager structuretemplatemanager = minecraftserver.getStructureManager();
int j = this.spigotConfig.viewDistance; // Spigot
int k = this.spigotConfig.simulationDistance; // Spigot
- PersistentEntitySectionManager persistententitysectionmanager = this.entityManager;
+ //PersistentEntitySectionManager persistententitysectionmanager = this.entityManager; // Paper - rewrite chunk system
- Objects.requireNonNull(this.entityManager);
- this.chunkSource = new ServerChunkCache(this, convertable_conversionsession, datafixer, structuretemplatemanager, executor, chunkgenerator, j, k, flag2, worldloadlistener, persistententitysectionmanager::updateChunkStatus, () -> {
+ //Objects.requireNonNull(this.entityManager); // Paper - rewrite chunk system
+ this.chunkSource = new ServerChunkCache(this, convertable_conversionsession, datafixer, structuretemplatemanager, executor, chunkgenerator, j, k, flag2, worldloadlistener, null, () -> { // Paper - rewrite chunk system
return minecraftserver.overworld().getDataStorage();
});
2022-12-07 19:32:25 +01:00
this.chunkSource.getGeneratorState().ensureStructuresGenerated();
@@ -399,6 +500,9 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.sleepStatus = new SleepStatus();
2022-12-07 19:32:25 +01:00
this.gameEventDispatcher = new GameEventDispatcher(this);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.getCraftServer().addWorld(this.getWorld()); // CraftBukkit
+
+ this.chunkTaskScheduler = new io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler(this, io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.workerThreads); // Paper - rewrite chunk system
+ this.entityLookup = new io.papermc.paper.chunk.system.entity.EntityLookup(this, new EntityCallbacks()); // Paper - rewrite chunk system
}
public void setWeatherParameters(int clearDuration, int rainDuration, boolean raining, boolean thundering) {
@@ -502,7 +606,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
gameprofilerfiller.push("checkDespawn");
entity.checkDespawn();
gameprofilerfiller.pop();
- if (this.chunkSource.chunkMap.getDistanceManager().inEntityTickingRange(entity.chunkPosition().toLong())) {
+ if (true || this.chunkSource.chunkMap.getDistanceManager().inEntityTickingRange(entity.chunkPosition().toLong())) { // Paper - now always true if in the ticking list
Entity entity1 = entity.getVehicle();
if (entity1 != null) {
@@ -527,13 +631,16 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
gameprofilerfiller.push("entityManagement");
- this.entityManager.tick();
+ //this.entityManager.tick(); // Paper - rewrite chunk system
gameprofilerfiller.pop();
2022-12-07 19:32:25 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public boolean shouldTickBlocksAt(long chunkPos) {
- return this.chunkSource.chunkMap.getDistanceManager().inBlockTickingRange(chunkPos);
+ // Paper start - replace player chunk loader system
+ ChunkHolder holder = this.chunkSource.chunkMap.getVisibleChunkIfPresent(chunkPos);
+ return holder != null && holder.isTickingReady();
+ // Paper end - replace player chunk loader system
}
protected void tickTime() {
2023-03-14 19:05:23 +01:00
@@ -994,6 +1101,11 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public void save(@Nullable ProgressListener progressListener, boolean flush, boolean savingDisabled) {
+ // Paper start - rewrite chunk system - add close param
+ this.save(progressListener, flush, savingDisabled, false);
+ }
+ public void save(@Nullable ProgressListener progressListener, boolean flush, boolean savingDisabled, boolean close) {
+ // Paper end - rewrite chunk system - add close param
ServerChunkCache chunkproviderserver = this.getChunkSource();
if (!savingDisabled) {
2023-03-14 19:05:23 +01:00
@@ -1009,16 +1121,13 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
timings.worldSaveChunks.startTiming(); // Paper
- chunkproviderserver.save(flush);
+ if (!close) chunkproviderserver.save(flush); // Paper - rewrite chunk system
+ if (close) chunkproviderserver.close(true); // Paper - rewrite chunk system
timings.worldSaveChunks.stopTiming(); // Paper
}// Paper
- if (flush) {
- this.entityManager.saveAll();
- } else {
- this.entityManager.autoSave();
- }
+ // Paper - rewrite chunk system - entity saving moved into ChunkHolder
- }
+ } else if (close) { chunkproviderserver.close(false); } // Paper - rewrite chunk system
// CraftBukkit start - moved from MinecraftServer.saveChunks
ServerLevel worldserver1 = this;
2023-03-14 19:05:23 +01:00
@@ -1154,7 +1263,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.removePlayerImmediately((ServerPlayer) entity, Entity.RemovalReason.DISCARDED);
}
- this.entityManager.addNewEntity(player);
+ this.entityLookup.addNewEntity(player); // Paper - rewite chunk system
}
// CraftBukkit start
2023-03-14 19:05:23 +01:00
@@ -1170,7 +1279,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
// CraftBukkit end
- return this.entityManager.addNewEntity(entity);
+ return this.entityLookup.addNewEntity(entity); // Paper - rewrite chunk system
}
}
2023-03-14 19:05:23 +01:00
@@ -1182,10 +1291,10 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public boolean tryAddFreshEntityWithPassengers(Entity entity, org.bukkit.event.entity.CreatureSpawnEvent.SpawnReason reason) {
// CraftBukkit end
Stream<UUID> stream = entity.getSelfAndPassengers().map(Entity::getUUID); // CraftBukkit - decompile error
- PersistentEntitySectionManager persistententitysectionmanager = this.entityManager;
+ //PersistentEntitySectionManager persistententitysectionmanager = this.entityManager; // Paper - rewrite chunk system
- Objects.requireNonNull(this.entityManager);
- if (stream.anyMatch(persistententitysectionmanager::isLoaded)) {
+ //Objects.requireNonNull(this.entityManager); // Paper - rewrite chunk system
+ if (stream.anyMatch(this.entityLookup::hasEntity)) { // Paper - rewrite chunk system
return false;
} else {
this.addFreshEntityWithPassengers(entity, reason); // CraftBukkit
2023-03-14 19:05:23 +01:00
@@ -1705,7 +1814,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
}
- bufferedwriter.write(String.format(Locale.ROOT, "entities: %s\n", this.entityManager.gatherStats()));
+ bufferedwriter.write(String.format(Locale.ROOT, "entities: %s\n", this.entityLookup.getDebugInfo())); // Paper - rewrite chunk system
bufferedwriter.write(String.format(Locale.ROOT, "block_entity_tickers: %d\n", this.blockEntityTickers.size()));
bufferedwriter.write(String.format(Locale.ROOT, "block_ticks: %d\n", this.getBlockTicks().count()));
bufferedwriter.write(String.format(Locale.ROOT, "fluid_ticks: %d\n", this.getFluidTicks().count()));
2023-03-14 19:05:23 +01:00
@@ -1754,7 +1863,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
BufferedWriter bufferedwriter2 = Files.newBufferedWriter(path1);
try {
- playerchunkmap.dumpChunks(bufferedwriter2);
+ //playerchunkmap.dumpChunks(bufferedwriter2); // Paper - rewrite chunk system
} catch (Throwable throwable4) {
if (bufferedwriter2 != null) {
try {
2023-03-14 19:05:23 +01:00
@@ -1775,7 +1884,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
BufferedWriter bufferedwriter3 = Files.newBufferedWriter(path2);
try {
- this.entityManager.dumpSections(bufferedwriter3);
+ //this.entityManager.dumpSections(bufferedwriter3); // Paper - rewrite chunk system
} catch (Throwable throwable6) {
if (bufferedwriter3 != null) {
try {
2023-03-14 19:05:23 +01:00
@@ -1917,7 +2026,7 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@VisibleForTesting
public String getWatchdogStats() {
- return String.format(Locale.ROOT, "players: %s, entities: %s [%s], block_entities: %d [%s], block_ticks: %d, fluid_ticks: %d, chunk_source: %s", this.players.size(), this.entityManager.gatherStats(), ServerLevel.getTypeCount(this.entityManager.getEntityGetter().getAll(), (entity) -> {
+ return String.format(Locale.ROOT, "players: %s, entities: %s [%s], block_entities: %d [%s], block_ticks: %d, fluid_ticks: %d, chunk_source: %s", this.players.size(), this.entityLookup.getDebugInfo(), ServerLevel.getTypeCount(this.entityLookup.getAll(), (entity) -> { // Paper - rewrite chunk system
2022-12-07 19:32:25 +01:00
return BuiltInRegistries.ENTITY_TYPE.getKey(entity.getType()).toString();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}), this.blockEntityTickers.size(), ServerLevel.getTypeCount(this.blockEntityTickers, TickingBlockEntity::getType), this.getBlockTicks().count(), this.getFluidTicks().count(), this.gatherChunkSourceStats());
}
2023-03-14 19:05:23 +01:00
@@ -1977,15 +2086,15 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public LevelEntityGetter<Entity> getEntities() {
org.spigotmc.AsyncCatcher.catchOp("Chunk getEntities call"); // Spigot
- return this.entityManager.getEntityGetter();
+ return this.entityLookup; // Paper - rewrite chunk system
}
public void addLegacyChunkEntities(Stream<Entity> entities) {
- this.entityManager.addLegacyChunkEntities(entities);
+ this.entityLookup.addLegacyChunkEntities(entities.toList()); // Paper - rewrite chunk system
}
public void addWorldGenChunkEntities(Stream<Entity> entities) {
- this.entityManager.addWorldGenChunkEntities(entities);
+ this.entityLookup.addWorldGenChunkEntities(entities.toList()); // Paper - rewrite chunk system
}
public void startTickingChunk(LevelChunk chunk) {
2023-03-14 19:05:23 +01:00
@@ -2001,34 +2110,49 @@ public class ServerLevel extends Level implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public void close() throws IOException {
super.close();
- this.entityManager.close();
+ //this.entityManager.close(); // Paper - rewrite chunk system
}
@Override
public String gatherChunkSourceStats() {
String s = this.chunkSource.gatherStats();
- return "Chunks[S] W: " + s + " E: " + this.entityManager.gatherStats();
+ return "Chunks[S] W: " + s + " E: " + this.entityLookup.getDebugInfo(); // Paper - rewrite chunk system
}
public boolean areEntitiesLoaded(long chunkPos) {
- return this.entityManager.areEntitiesLoaded(chunkPos);
+ // Paper start - rewrite chunk system
+ return this.getChunkIfLoadedImmediately(ChunkPos.getX(chunkPos), ChunkPos.getZ(chunkPos)) != null;
+ // Paper end - rewrite chunk system
}
private boolean isPositionTickingWithEntitiesLoaded(long chunkPos) {
- return this.areEntitiesLoaded(chunkPos) && this.chunkSource.isPositionTicking(chunkPos);
+ // Paper start - optimize is ticking ready type functions
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder = this.chunkTaskScheduler.chunkHolderManager.getChunkHolder(chunkPos);
+ // isTicking implies the chunk is loaded, and the chunk is loaded now implies the entities are loaded
+ return chunkHolder != null && chunkHolder.isTickingReady();
+ // Paper end
}
public boolean isPositionEntityTicking(BlockPos pos) {
- return this.entityManager.canPositionTick(pos) && this.chunkSource.chunkMap.getDistanceManager().inEntityTickingRange(ChunkPos.asLong(pos));
+ // Paper start - rewrite chunk system
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder = this.chunkTaskScheduler.chunkHolderManager.getChunkHolder(io.papermc.paper.util.CoordinateUtils.getChunkKey(pos));
+ return chunkHolder != null && chunkHolder.isEntityTickingReady();
+ // Paper end - rewrite chunk system
}
public boolean isNaturalSpawningAllowed(BlockPos pos) {
- return this.entityManager.canPositionTick(pos);
+ // Paper start - rewrite chunk system
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder = this.chunkTaskScheduler.chunkHolderManager.getChunkHolder(io.papermc.paper.util.CoordinateUtils.getChunkKey(pos));
+ return chunkHolder != null && chunkHolder.isEntityTickingReady();
+ // Paper end - rewrite chunk system
}
public boolean isNaturalSpawningAllowed(ChunkPos pos) {
- return this.entityManager.canPositionTick(pos);
+ // Paper start - rewrite chunk system
+ io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder = this.chunkTaskScheduler.chunkHolderManager.getChunkHolder(io.papermc.paper.util.CoordinateUtils.getChunkKey(pos));
+ return chunkHolder != null && chunkHolder.isEntityTickingReady();
+ // Paper end - rewrite chunk system
}
2022-12-07 19:32:25 +01:00
@Override
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
diff --git a/src/main/java/net/minecraft/server/level/ThreadedLevelLightEngine.java b/src/main/java/net/minecraft/server/level/ThreadedLevelLightEngine.java
index 275b7f7dd36a2073a3eb9f89f4c832839e5aa9af..660693c6dc0ef86f4013df980b6d0c11c03e46cd 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/ThreadedLevelLightEngine.java
+++ b/src/main/java/net/minecraft/server/level/ThreadedLevelLightEngine.java
@@ -36,15 +36,14 @@ import net.minecraft.world.level.chunk.ChunkStatus;
public class ThreadedLevelLightEngine extends LevelLightEngine implements AutoCloseable {
private static final Logger LOGGER = LogUtils.getLogger();
- private final ProcessorMailbox<Runnable> taskMailbox;
- private final ObjectList<Pair<ThreadedLevelLightEngine.TaskType, Runnable>> lightTasks = new ObjectArrayList<>();
+ // Paper - rewrite chunk system
private final ChunkMap chunkMap;
- private final ProcessorHandle<ChunkTaskPriorityQueueSorter.Message<Runnable>> sorterMailbox;
+ // Paper - rewrite chunk system
private volatile int taskPerBatch = 5;
- private final AtomicBoolean scheduled = new AtomicBoolean();
+ // Paper - rewrite chunk system
// Paper start - replace light engine impl
- protected final ca.spottedleaf.starlight.common.light.StarLightInterface theLightEngine;
+ public final ca.spottedleaf.starlight.common.light.StarLightInterface theLightEngine;
public final boolean hasBlockLight;
public final boolean hasSkyLight;
// Paper end - replace light engine impl
@@ -52,8 +51,7 @@ public class ThreadedLevelLightEngine extends LevelLightEngine implements AutoCl
public ThreadedLevelLightEngine(LightChunkGetter chunkProvider, ChunkMap chunkStorage, boolean hasBlockLight, ProcessorMailbox<Runnable> processor, ProcessorHandle<ChunkTaskPriorityQueueSorter.Message<Runnable>> executor) {
super(chunkProvider, false, false); // Paper - destroy vanilla light engine state
this.chunkMap = chunkStorage;
- this.sorterMailbox = executor;
- this.taskMailbox = processor;
+ // Paper - rewrite chunk system
// Paper start - replace light engine impl
this.hasBlockLight = true;
this.hasSkyLight = hasBlockLight; // Nice variable name.
@@ -97,7 +95,7 @@ public class ThreadedLevelLightEngine extends LevelLightEngine implements AutoCl
++totalChunks;
}
- this.taskMailbox.tell(() -> {
+ this.chunkMap.level.chunkTaskScheduler.lightExecutor.queueRunnable(() -> { // Paper - rewrite chunk system
this.theLightEngine.relightChunks(chunks, (ChunkPos chunkPos) -> {
chunkLightCallback.accept(chunkPos);
((java.util.concurrent.Executor)((ServerLevel)this.theLightEngine.getWorld()).getChunkSource().mainThreadProcessor).execute(() -> {
@@ -269,17 +267,11 @@ public class ThreadedLevelLightEngine extends LevelLightEngine implements AutoCl
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
private void addTask(int x, int z, ThreadedLevelLightEngine.TaskType stage, Runnable task) {
- this.addTask(x, z, this.chunkMap.getChunkQueueLevel(ChunkPos.asLong(x, z)), stage, task);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
private void addTask(int x, int z, IntSupplier completedLevelSupplier, ThreadedLevelLightEngine.TaskType stage, Runnable task) {
- this.sorterMailbox.tell(ChunkTaskPriorityQueueSorter.message(() -> {
- this.lightTasks.add(Pair.of(stage, task));
- if (this.lightTasks.size() >= this.taskPerBatch) {
- this.runUpdate();
- }
-
- }, ChunkPos.asLong(x, z), completedLevelSupplier));
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
@Override
@@ -337,74 +329,15 @@ public class ThreadedLevelLightEngine extends LevelLightEngine implements AutoCl
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
});
}
- // Paper end - replace light engine impl
- ChunkPos chunkPos = chunk.getPos();
- chunk.setLightCorrect(false);
- this.addTask(chunkPos.x, chunkPos.z, ThreadedLevelLightEngine.TaskType.PRE_UPDATE, Util.name(() -> {
- LevelChunkSection[] levelChunkSections = chunk.getSections();
-
- for(int i = 0; i < chunk.getSectionsCount(); ++i) {
- LevelChunkSection levelChunkSection = levelChunkSections[i];
- if (!levelChunkSection.hasOnlyAir()) {
- int j = this.levelHeightAccessor.getSectionYFromSectionIndex(i);
- super.updateSectionStatus(SectionPos.of(chunkPos, j), false);
- }
- }
-
- super.enableLightSources(chunkPos, true);
- if (!excludeBlocks) {
- chunk.getLights().forEach((pos) -> {
- super.onBlockEmissionIncrease(pos, chunk.getLightEmission(pos));
- });
- }
-
- }, () -> {
- return "lightChunk " + chunkPos + " " + excludeBlocks;
- }));
- return CompletableFuture.supplyAsync(() -> {
- chunk.setLightCorrect(true);
- super.retainData(chunkPos, false);
- this.chunkMap.releaseLightTicket(chunkPos);
- return chunk;
- }, (runnable) -> {
- this.addTask(chunkPos.x, chunkPos.z, ThreadedLevelLightEngine.TaskType.POST_UPDATE, runnable);
- });
+ throw new InternalError(); // Paper - rewrite chunk system
}
public void tryScheduleUpdate() {
- if (this.hasLightWork() && this.scheduled.compareAndSet(false, true)) { // Paper // Paper - rewrite light engine
- this.taskMailbox.tell(() -> {
- this.runUpdate();
- this.scheduled.set(false);
- });
- }
-
+ // Paper - rewrite chunk system
}
private void runUpdate() {
- int i = Math.min(this.lightTasks.size(), this.taskPerBatch);
- ObjectListIterator<Pair<ThreadedLevelLightEngine.TaskType, Runnable>> objectListIterator = this.lightTasks.iterator();
-
- int j;
- for(j = 0; objectListIterator.hasNext() && j < i; ++j) {
- Pair<ThreadedLevelLightEngine.TaskType, Runnable> pair = objectListIterator.next();
- if (pair.getFirst() == ThreadedLevelLightEngine.TaskType.PRE_UPDATE) {
- pair.getSecond().run();
- }
- }
-
- objectListIterator.back(j);
- this.theLightEngine.propagateChanges(); // Paper - rewrite light engine
-
- for(int var5 = 0; objectListIterator.hasNext() && var5 < i; ++var5) {
- Pair<ThreadedLevelLightEngine.TaskType, Runnable> pair2 = objectListIterator.next();
- if (pair2.getFirst() == ThreadedLevelLightEngine.TaskType.POST_UPDATE) {
- pair2.getSecond().run();
- }
-
- objectListIterator.remove();
- }
-
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
public void setTaskPerBatch(int taskBatchSize) {
diff --git a/src/main/java/net/minecraft/server/level/Ticket.java b/src/main/java/net/minecraft/server/level/Ticket.java
index b346fa94b23d81da7da073f71dd12e672e0f079c..768a2667f950a635a562fa8a0c75b31a3ae9190e 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/Ticket.java
+++ b/src/main/java/net/minecraft/server/level/Ticket.java
@@ -6,9 +6,12 @@ public final class Ticket<T> implements Comparable<Ticket<?>> {
private final TicketType<T> type;
private final int ticketLevel;
public final T key;
- private long createdTick;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - rewrite chunk system
+ public final long removalTick;
- protected Ticket(TicketType<T> type, int level, T argument) {
+ public Ticket(TicketType<T> type, int level, T argument, long removalTick) {
+ this.removalTick = removalTick;
+ // Paper end - rewrite chunk system
this.type = type;
this.ticketLevel = level;
this.key = argument;
@@ -44,7 +47,7 @@ public final class Ticket<T> implements Comparable<Ticket<?>> {
@Override
public String toString() {
- return "Ticket[" + this.type + " " + this.ticketLevel + " (" + this.key + ")] at " + this.createdTick;
+ return "Ticket[" + this.type + " " + this.ticketLevel + " (" + this.key + ")] to die on " + this.removalTick; // Paper - rewrite chunk system
}
public TicketType<T> getType() {
@@ -56,11 +59,10 @@ public final class Ticket<T> implements Comparable<Ticket<?>> {
}
protected void setCreatedTick(long tickCreated) {
- this.createdTick = tickCreated;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
protected boolean timedOut(long currentTick) {
- long l = this.type.timeout();
- return l != 0L && currentTick - this.createdTick > l;
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
}
diff --git a/src/main/java/net/minecraft/server/level/TicketType.java b/src/main/java/net/minecraft/server/level/TicketType.java
index 6051e5f272838ef23276a90e21c2fc821ca155d1..97d1ff2af23bac14e67bca5896843325aaa5bfc1 100644
--- a/src/main/java/net/minecraft/server/level/TicketType.java
+++ b/src/main/java/net/minecraft/server/level/TicketType.java
@@ -8,6 +8,7 @@ import net.minecraft.world.level.ChunkPos;
public class TicketType<T> {
public static final TicketType<Long> FUTURE_AWAIT = create("future_await", Long::compareTo); // Paper
+ public static final TicketType<Long> ASYNC_LOAD = create("async_load", Long::compareTo); // Paper
private final String name;
private final Comparator<T> comparator;
@@ -27,6 +28,13 @@ public class TicketType<T> {
public static final TicketType<Unit> PLUGIN = TicketType.create("plugin", (a, b) -> 0); // CraftBukkit
public static final TicketType<org.bukkit.plugin.Plugin> PLUGIN_TICKET = TicketType.create("plugin_ticket", (plugin1, plugin2) -> plugin1.getClass().getName().compareTo(plugin2.getClass().getName())); // CraftBukkit
public static final TicketType<Long> CHUNK_RELIGHT = create("light_update", Long::compareTo); // Paper - ensure chunks stay loaded for lighting
+ // Paper start - rewrite chunk system
+ public static final TicketType<Long> CHUNK_LOAD = create("chunk_load", Long::compareTo);
+ public static final TicketType<Long> STATUS_UPGRADE = create("status_upgrade", Long::compareTo);
+ public static final TicketType<Long> ENTITY_LOAD = create("entity_load", Long::compareTo);
+ public static final TicketType<Long> POI_LOAD = create("poi_load", Long::compareTo);
+ public static final TicketType<Unit> UNLOAD_COOLDOWN = create("unload_cooldown", (u1, u2) -> 0, 5 * 20);
+ // Paper end - rewrite chunk system
public static <T> TicketType<T> create(String name, Comparator<T> argumentComparator) {
return new TicketType<>(name, argumentComparator, 0L);
diff --git a/src/main/java/net/minecraft/server/level/WorldGenRegion.java b/src/main/java/net/minecraft/server/level/WorldGenRegion.java
2022-12-07 19:32:25 +01:00
index e96a0ca47e4701ba187555bd92c968345bc85677..73b96f804079288e9c5fcc11da54e61e89a6782a 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/level/WorldGenRegion.java
+++ b/src/main/java/net/minecraft/server/level/WorldGenRegion.java
2022-12-07 19:32:25 +01:00
@@ -504,4 +504,21 @@ public class WorldGenRegion implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public long nextSubTickCount() {
return this.subTickCount.getAndIncrement();
}
+
+ // Paper start
+ // No-op, this class doesn't provide entity access
+ @Override
+ public List<Entity> getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate) {
+ return Collections.emptyList();
+ }
+
+ @Override
+ public void getEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {}
+
+ @Override
+ public void getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {}
+
+ @Override
+ public <T> void getEntitiesByClass(Class<? extends T> clazz, Entity except, AABB box, List<? super T> into, Predicate<? super T> predicate) {}
+ // Paper end
}
diff --git a/src/main/java/net/minecraft/server/network/ServerGamePacketListenerImpl.java b/src/main/java/net/minecraft/server/network/ServerGamePacketListenerImpl.java
index e9c8c008ce37ceec3be993e672786eaeb8d01a7e..542c6ebbf01b91709b73052eca3889347a5aa90a 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/network/ServerGamePacketListenerImpl.java
+++ b/src/main/java/net/minecraft/server/network/ServerGamePacketListenerImpl.java
@@ -779,6 +779,13 @@ public class ServerGamePacketListenerImpl implements ServerPlayerConnection, Tic
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.disconnect(Component.translatable("disconnect.spam"));
return;
}
+ // Paper start
+ String str = packet.getCommand(); int index = -1;
+ if (str.length() > 64 && ((index = str.indexOf(' ')) == -1 || index >= 64)) {
+ server.scheduleOnMain(() -> this.disconnect(Component.translatable("disconnect.spam", new Object[0]))); // Paper
+ return;
+ }
+ // Paper end
// CraftBukkit end
StringReader stringreader = new StringReader(packet.getCommand());
diff --git a/src/main/java/net/minecraft/server/players/PlayerList.java b/src/main/java/net/minecraft/server/players/PlayerList.java
index 53e85923c417cf4d853c27e6fedbe7ff525a897c..cb365522a2b462e8c3e44eac68c64ad99c8738ff 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/server/players/PlayerList.java
+++ b/src/main/java/net/minecraft/server/players/PlayerList.java
@@ -253,7 +253,7 @@ public abstract class PlayerList {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
boolean flag1 = gamerules.getBoolean(GameRules.RULE_REDUCEDDEBUGINFO);
// Spigot - view distance
2022-12-07 19:32:25 +01:00
- playerconnection.send(new ClientboundLoginPacket(player.getId(), worlddata.isHardcore(), player.gameMode.getGameModeForPlayer(), player.gameMode.getPreviousGameModeForPlayer(), this.server.levelKeys(), this.synchronizedRegistries, worldserver1.dimensionTypeId(), worldserver1.dimension(), BiomeManager.obfuscateSeed(worldserver1.getSeed()), this.getMaxPlayers(), worldserver1.spigotConfig.viewDistance, worldserver1.spigotConfig.simulationDistance, flag1, !flag, worldserver1.isDebug(), worldserver1.isFlat(), player.getLastDeathLocation()));
+ playerconnection.send(new ClientboundLoginPacket(player.getId(), worlddata.isHardcore(), player.gameMode.getGameModeForPlayer(), player.gameMode.getPreviousGameModeForPlayer(), this.server.levelKeys(), this.synchronizedRegistries, worldserver1.dimensionTypeId(), worldserver1.dimension(), BiomeManager.obfuscateSeed(worldserver1.getSeed()), this.getMaxPlayers(), worldserver1.getChunkSource().chunkMap.playerChunkManager.getTargetSendDistance(), worldserver1.getChunkSource().chunkMap.playerChunkManager.getTargetTickViewDistance(), flag1, !flag, worldserver1.isDebug(), worldserver1.isFlat(), player.getLastDeathLocation())); // Paper - replace old player chunk management
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
player.getBukkitEntity().sendSupportedChannels(); // CraftBukkit
2022-12-07 19:32:25 +01:00
playerconnection.send(new ClientboundUpdateEnabledFeaturesPacket(FeatureFlags.REGISTRY.toNames(worldserver1.enabledFeatures())));
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
playerconnection.send(new ClientboundCustomPayloadPacket(ClientboundCustomPayloadPacket.BRAND, (new FriendlyByteBuf(Unpooled.buffer())).writeUtf(this.getServer().getServerModName())));
@@ -791,8 +791,8 @@ public abstract class PlayerList {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// CraftBukkit start
LevelData worlddata = worldserver1.getLevelData();
2022-12-07 19:32:25 +01:00
entityplayer1.connection.send(new ClientboundRespawnPacket(worldserver1.dimensionTypeId(), worldserver1.dimension(), BiomeManager.obfuscateSeed(worldserver1.getSeed()), entityplayer1.gameMode.getGameModeForPlayer(), entityplayer1.gameMode.getPreviousGameModeForPlayer(), worldserver1.isDebug(), worldserver1.isFlat(), (byte) i, entityplayer1.getLastDeathLocation()));
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- entityplayer1.connection.send(new ClientboundSetChunkCacheRadiusPacket(worldserver1.spigotConfig.viewDistance)); // Spigot
- entityplayer1.connection.send(new ClientboundSetSimulationDistancePacket(worldserver1.spigotConfig.simulationDistance)); // Spigot
+ entityplayer1.connection.send(new ClientboundSetChunkCacheRadiusPacket(worldserver1.getChunkSource().chunkMap.playerChunkManager.getTargetSendDistance())); // Spigot // Paper - replace old player chunk management
+ entityplayer1.connection.send(new ClientboundSetSimulationDistancePacket(worldserver1.getChunkSource().chunkMap.playerChunkManager.getTargetTickViewDistance())); // Spigot // Paper - replace old player chunk management
entityplayer1.spawnIn(worldserver1);
entityplayer1.unsetRemoved();
entityplayer1.connection.teleport(new Location(worldserver1.getWorld(), entityplayer1.getX(), entityplayer1.getY(), entityplayer1.getZ(), entityplayer1.getYRot(), entityplayer1.getXRot()));
@@ -1288,7 +1288,7 @@ public abstract class PlayerList {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public void setViewDistance(int viewDistance) {
this.viewDistance = viewDistance;
- this.broadcastAll(new ClientboundSetChunkCacheRadiusPacket(viewDistance));
+ //this.broadcastAll(new ClientboundSetChunkCacheRadiusPacket(viewDistance)); // Paper - move into setViewDistance
Iterator iterator = this.server.getAllLevels().iterator();
while (iterator.hasNext()) {
@@ -1303,7 +1303,7 @@ public abstract class PlayerList {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public void setSimulationDistance(int simulationDistance) {
this.simulationDistance = simulationDistance;
- this.broadcastAll(new ClientboundSetSimulationDistancePacket(simulationDistance));
+ //this.broadcastAll(new ClientboundSetSimulationDistancePacket(simulationDistance)); // Paper - handled by playerchunkloader
Iterator iterator = this.server.getAllLevels().iterator();
while (iterator.hasNext()) {
diff --git a/src/main/java/net/minecraft/util/SortedArraySet.java b/src/main/java/net/minecraft/util/SortedArraySet.java
2022-12-07 19:32:25 +01:00
index ca788f0dcec4a117b410fe8348969e056b138b1e..4f5f2c25e12ee6d977bc98d9118650cfe91e6c0e 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/util/SortedArraySet.java
+++ b/src/main/java/net/minecraft/util/SortedArraySet.java
@@ -22,6 +22,41 @@ public class SortedArraySet<T> extends AbstractSet<T> {
this.contents = (T[])castRawArray(new Object[initialCapacity]);
}
}
+ // Paper start - optimise removeIf
+ @Override
+ public boolean removeIf(java.util.function.Predicate<? super T> filter) {
+ // prev. impl used an iterator, which could be n^2 and creates garbage
+ int i = 0, len = this.size;
+ T[] backingArray = this.contents;
+
+ for (;;) {
+ if (i >= len) {
+ return false;
+ }
+ if (!filter.test(backingArray[i])) {
+ ++i;
+ continue;
+ }
+ break;
+ }
+
+ // we only want to write back to backingArray if we really need to
+
+ int lastIndex = i; // this is where new elements are shifted to
+
+ for (; i < len; ++i) {
+ T curr = backingArray[i];
+ if (!filter.test(curr)) { // if test throws we're screwed
+ backingArray[lastIndex++] = curr;
+ }
+ }
+
+ // cleanup end
+ Arrays.fill(backingArray, lastIndex, len, null);
+ this.size = lastIndex;
+ return true;
+ }
+ // Paper end - optimise removeIf
public static <T extends Comparable<T>> SortedArraySet<T> create() {
return create(10);
@@ -110,6 +145,31 @@ public class SortedArraySet<T> extends AbstractSet<T> {
}
}
+ // Paper start - rewrite chunk system
+ public T replace(T object) {
+ int i = this.findIndex(object);
+ if (i >= 0) {
+ T old = this.contents[i];
+ this.contents[i] = object;
+ return old;
+ } else {
+ this.addInternal(object, getInsertionPosition(i));
+ return object;
+ }
+ }
+
+ public T removeAndGet(T object) {
+ int i = this.findIndex(object);
+ if (i >= 0) {
+ final T ret = this.contents[i];
+ this.removeInternal(i);
+ return ret;
+ } else {
+ return null;
+ }
+ }
+ // Paper end - rewrite chunk system
+
@Override
public boolean remove(Object object) {
int i = this.findIndex((T)object);
diff --git a/src/main/java/net/minecraft/util/worldupdate/WorldUpgrader.java b/src/main/java/net/minecraft/util/worldupdate/WorldUpgrader.java
2023-03-14 19:05:23 +01:00
index 12e72ad737b1219fcdf88d344d41621d9fd5feec..e0bfeebeaac1aaea64bc07cdfdf7790e3e43ca7b 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/util/worldupdate/WorldUpgrader.java
+++ b/src/main/java/net/minecraft/util/worldupdate/WorldUpgrader.java
2022-12-07 19:32:25 +01:00
@@ -186,7 +186,11 @@ public class WorldUpgrader {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
WorldUpgrader.LOGGER.error("Error upgrading chunk {}", chunkcoordintpair, throwable);
+ // Paper start
+ } catch (IOException e) {
+ WorldUpgrader.LOGGER.error("Error upgrading chunk {}", chunkcoordintpair, e);
}
+ // Paper end
if (flag1) {
++this.converted;
diff --git a/src/main/java/net/minecraft/world/entity/Entity.java b/src/main/java/net/minecraft/world/entity/Entity.java
index 9171dfb7443d42a5f6d854b2abea55d2671c1177..8a9700d8bbfe2f813259543ed4167495d13e1b4f 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/entity/Entity.java
+++ b/src/main/java/net/minecraft/world/entity/Entity.java
2023-03-14 19:05:23 +01:00
@@ -322,6 +322,58 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
// Paper end
+ // Paper start
+ /**
+ * Overriding this field will cause memory leaks.
+ */
+ private final boolean hardCollides;
+
+ private static final java.util.Map<Class<? extends Entity>, Boolean> cachedOverrides = java.util.Collections.synchronizedMap(new java.util.WeakHashMap<>());
+ {
+ /* // Goodbye, broken on reobf...
+ Boolean hardCollides = cachedOverrides.get(this.getClass());
+ if (hardCollides == null) {
+ try {
+ java.lang.reflect.Method getHardCollisionBoxEntityMethod = Entity.class.getMethod("canCollideWith", Entity.class);
+ java.lang.reflect.Method hasHardCollisionBoxMethod = Entity.class.getMethod("canBeCollidedWith");
+ if (!this.getClass().getMethod(hasHardCollisionBoxMethod.getName(), hasHardCollisionBoxMethod.getParameterTypes()).equals(hasHardCollisionBoxMethod)
+ || !this.getClass().getMethod(getHardCollisionBoxEntityMethod.getName(), getHardCollisionBoxEntityMethod.getParameterTypes()).equals(getHardCollisionBoxEntityMethod)) {
+ hardCollides = Boolean.TRUE;
+ } else {
+ hardCollides = Boolean.FALSE;
+ }
+ cachedOverrides.put(this.getClass(), hardCollides);
+ }
+ catch (ThreadDeath thr) { throw thr; }
+ catch (Throwable thr) {
+ // shouldn't happen, just explode
+ throw new RuntimeException(thr);
+ }
+ } */
+ this.hardCollides = this instanceof Boat
+ || this instanceof net.minecraft.world.entity.monster.Shulker
+ || this instanceof net.minecraft.world.entity.vehicle.AbstractMinecart
+ || this.shouldHardCollide();
+ }
+
+ // plugins can override
+ protected boolean shouldHardCollide() {
+ return false;
+ }
+
+ public final boolean hardCollides() {
+ return this.hardCollides;
+ }
+
+ public net.minecraft.server.level.ChunkHolder.FullChunkStatus chunkStatus;
+
+ public int sectionX = Integer.MIN_VALUE;
+ public int sectionY = Integer.MIN_VALUE;
+ public int sectionZ = Integer.MIN_VALUE;
+
+ public boolean updatingSectionStatus = false;
+ // Paper end
+
public Entity(EntityType<?> type, Level world) {
this.id = Entity.ENTITY_COUNTER.incrementAndGet();
this.passengers = ImmutableList.of();
2023-03-14 19:05:23 +01:00
@@ -2136,11 +2188,11 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return InteractionResult.PASS;
}
- public boolean canCollideWith(Entity other) {
+ public boolean canCollideWith(Entity other) { // Paper - diff on change, hard colliding entities override this - TODO CHECK ON UPDATE - AbstractMinecart/Boat override
return other.canBeCollidedWith() && !this.isPassengerOfSameVehicle(other);
}
- public boolean canBeCollidedWith() {
+ public boolean canBeCollidedWith() { // Paper - diff on change, hard colliding entities override this TODO CHECK ON UPDATE - Boat/Shulker override
return false;
}
@@ -3442,6 +3494,16 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
};
}
+ // Paper start - rewrite chunk system
+ public boolean hasAnyPlayerPassengers() {
+ // copied from below
+ if (this.passengers.isEmpty()) { return false; }
+ return this.getIndirectPassengersStream().anyMatch((entity) -> {
+ return entity instanceof Player;
+ });
+ }
+ // Paper end - rewrite chunk system
+
public boolean hasExactlyOnePlayerPassenger() {
return this.getIndirectPassengersStream().filter((entity) -> {
return entity instanceof Player;
@@ -3769,6 +3831,12 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public final void setPosRaw(double x, double y, double z) {
+ // Paper start - rewrite chunk system
+ if (this.updatingSectionStatus) {
+ LOGGER.error("Refusing to update position for entity " + this + " to position " + new Vec3(x, y, z) + " since it is processing a section status update", new Throwable());
+ return;
+ }
+ // Paper end - rewrite chunk system
if (this.position.x != x || this.position.y != y || this.position.z != z) {
this.position = new Vec3(x, y, z);
int i = Mth.floor(x);
@@ -3876,6 +3944,13 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public final void setRemoved(Entity.RemovalReason reason) {
+ // Paper start - rewrite chunk system
+ io.papermc.paper.util.TickThread.ensureTickThread(this, "Cannot remove entity off-main");
+ if (this.updatingSectionStatus) {
+ LOGGER.warn("Entity " + this + " is currently prevented from being added/removed to world since it is processing section status updates", new Throwable());
+ return;
+ }
+ // Paper end - rewrite chunk system
if (this.removalReason == null) {
this.removalReason = reason;
}
@@ -3884,7 +3959,7 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.stopRiding();
}
- this.getPassengers().forEach(Entity::stopRiding);
+ if (reason != RemovalReason.UNLOADED_TO_CHUNK) this.getPassengers().forEach(Entity::stopRiding); // Paper - chunk system - don't adjust passenger state when unloading, it's just not safe (and messes with our logic in entity chunk unload)
this.levelCallback.onRemove(reason);
}
@@ -3899,7 +3974,7 @@ public abstract class Entity implements Nameable, EntityAccess, CommandSource {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public boolean shouldBeSaved() {
- return this.removalReason != null && !this.removalReason.shouldSave() ? false : (this.isPassenger() ? false : !this.isVehicle() || !this.hasExactlyOnePlayerPassenger());
+ return this.removalReason != null && !this.removalReason.shouldSave() ? false : (this.isPassenger() ? false : !this.isVehicle() || !this.hasAnyPlayerPassengers()); // Paper - rewrite chunk system - it should check if the entity has ANY player passengers
}
@Override
diff --git a/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiManager.java b/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiManager.java
2022-12-07 19:32:25 +01:00
index 4569f21643812e73897c9812be684a371df60f74..440e4f35e38d38e5407a6fecf09ab9511a44a670 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiManager.java
+++ b/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiManager.java
@@ -38,12 +38,28 @@ import net.minecraft.world.level.chunk.storage.SectionStorage;
public class PoiManager extends SectionStorage<PoiSection> {
public static final int MAX_VILLAGE_DISTANCE = 6;
public static final int VILLAGE_SECTION_SIZE = 1;
- private final PoiManager.DistanceTracker distanceTracker;
- private final LongSet loadedChunks = new LongOpenHashSet();
+ // Paper start - rewrite chunk system
+ // the vanilla tracker needs to be replaced because it does not support level removes
+ public final net.minecraft.server.level.ServerLevel world;
+ private final io.papermc.paper.util.misc.Delayed26WayDistancePropagator3D villageDistanceTracker = new io.papermc.paper.util.misc.Delayed26WayDistancePropagator3D();
+ static final int POI_DATA_SOURCE = 7;
+ public static int convertBetweenLevels(final int level) {
+ return POI_DATA_SOURCE - level;
+ }
+
+ protected void updateDistanceTracking(long section) {
+ if (this.isVillageCenter(section)) {
+ this.villageDistanceTracker.setSource(section, POI_DATA_SOURCE);
+ } else {
+ this.villageDistanceTracker.removeSource(section);
+ }
+ }
+ // Paper end - rewrite chunk system
+
public PoiManager(Path path, DataFixer dataFixer, boolean dsync, RegistryAccess registryManager, LevelHeightAccessor world) {
super(path, PoiSection::codec, PoiSection::new, dataFixer, DataFixTypes.POI_CHUNK, dsync, registryManager, world);
- this.distanceTracker = new PoiManager.DistanceTracker();
+ this.world = (net.minecraft.server.level.ServerLevel)world; // Paper - rewrite chunk system
}
public void add(BlockPos pos, Holder<PoiType> type) {
@@ -180,8 +196,8 @@ public class PoiManager extends SectionStorage<PoiSection> {
}
public int sectionsToVillage(SectionPos pos) {
- this.distanceTracker.runAllUpdates();
- return this.distanceTracker.getLevel(pos.asLong());
+ this.villageDistanceTracker.propagateUpdates(); // Paper - replace distance tracking util
+ return convertBetweenLevels(this.villageDistanceTracker.getLevel(io.papermc.paper.util.CoordinateUtils.getChunkSectionKey(pos))); // Paper - replace distance tracking util
}
boolean isVillageCenter(long pos) {
@@ -195,21 +211,106 @@ public class PoiManager extends SectionStorage<PoiSection> {
@Override
public void tick(BooleanSupplier shouldKeepTicking) {
- super.tick(shouldKeepTicking);
- this.distanceTracker.runAllUpdates();
+ this.villageDistanceTracker.propagateUpdates(); // Paper - rewrite chunk system
}
@Override
- protected void setDirty(long pos) {
- super.setDirty(pos);
- this.distanceTracker.update(pos, this.distanceTracker.getLevelFromSource(pos), false);
+ public void setDirty(long pos) {
+ // Paper start - rewrite chunk system
+ int chunkX = io.papermc.paper.util.CoordinateUtils.getChunkSectionX(pos);
+ int chunkZ = io.papermc.paper.util.CoordinateUtils.getChunkSectionZ(pos);
+ io.papermc.paper.chunk.system.scheduling.ChunkHolderManager manager = this.world.chunkTaskScheduler.chunkHolderManager;
+ io.papermc.paper.chunk.system.poi.PoiChunk chunk = manager.getPoiChunkIfLoaded(chunkX, chunkZ, false);
+ if (chunk != null) {
+ chunk.setDirty(true);
+ }
+ this.updateDistanceTracking(pos);
+ // Paper end - rewrite chunk system
}
@Override
protected void onSectionLoad(long pos) {
- this.distanceTracker.update(pos, this.distanceTracker.getLevelFromSource(pos), false);
+ this.updateDistanceTracking(pos); // Paper - move to new distance tracking util
}
+ @Override
+ public Optional<PoiSection> get(long pos) {
+ int chunkX = io.papermc.paper.util.CoordinateUtils.getChunkSectionX(pos);
+ int chunkY = io.papermc.paper.util.CoordinateUtils.getChunkSectionY(pos);
+ int chunkZ = io.papermc.paper.util.CoordinateUtils.getChunkSectionZ(pos);
+
+ io.papermc.paper.util.TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Accessing poi chunk off-main");
+
+ io.papermc.paper.chunk.system.scheduling.ChunkHolderManager manager = this.world.chunkTaskScheduler.chunkHolderManager;
+ io.papermc.paper.chunk.system.poi.PoiChunk ret = manager.getPoiChunkIfLoaded(chunkX, chunkZ, true);
+
+ return ret == null ? Optional.empty() : ret.getSectionForVanilla(chunkY);
+ }
+
+ @Override
+ public Optional<PoiSection> getOrLoad(long pos) {
+ int chunkX = io.papermc.paper.util.CoordinateUtils.getChunkSectionX(pos);
+ int chunkY = io.papermc.paper.util.CoordinateUtils.getChunkSectionY(pos);
+ int chunkZ = io.papermc.paper.util.CoordinateUtils.getChunkSectionZ(pos);
+
+ io.papermc.paper.util.TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Accessing poi chunk off-main");
+
+ io.papermc.paper.chunk.system.scheduling.ChunkHolderManager manager = this.world.chunkTaskScheduler.chunkHolderManager;
+
+ if (chunkY >= io.papermc.paper.util.WorldUtil.getMinSection(this.world) &&
+ chunkY <= io.papermc.paper.util.WorldUtil.getMaxSection(this.world)) {
+ io.papermc.paper.chunk.system.poi.PoiChunk ret = manager.getPoiChunkIfLoaded(chunkX, chunkZ, true);
+ if (ret != null) {
+ return ret.getSectionForVanilla(chunkY);
+ } else {
+ return manager.loadPoiChunk(chunkX, chunkZ).getSectionForVanilla(chunkY);
+ }
+ }
+ // retain vanilla behavior: do not load section if out of bounds!
+ return Optional.empty();
+ }
+
+ @Override
+ protected PoiSection getOrCreate(long pos) {
+ int chunkX = io.papermc.paper.util.CoordinateUtils.getChunkSectionX(pos);
+ int chunkY = io.papermc.paper.util.CoordinateUtils.getChunkSectionY(pos);
+ int chunkZ = io.papermc.paper.util.CoordinateUtils.getChunkSectionZ(pos);
+
+ io.papermc.paper.util.TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Accessing poi chunk off-main");
+
+ io.papermc.paper.chunk.system.scheduling.ChunkHolderManager manager = this.world.chunkTaskScheduler.chunkHolderManager;
+
+ io.papermc.paper.chunk.system.poi.PoiChunk ret = manager.getPoiChunkIfLoaded(chunkX, chunkZ, true);
+ if (ret != null) {
+ return ret.getOrCreateSection(chunkY);
+ } else {
+ return manager.loadPoiChunk(chunkX, chunkZ).getOrCreateSection(chunkY);
+ }
+ }
+
+ public void onUnload(long coordinate) { // Paper - rewrite chunk system
+ int chunkX = io.papermc.paper.util.MCUtil.getCoordinateX(coordinate);
+ int chunkZ = io.papermc.paper.util.MCUtil.getCoordinateZ(coordinate);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ io.papermc.paper.util.TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Unloading poi chunk off-main");
+ for (int section = this.levelHeightAccessor.getMinSection(); section < this.levelHeightAccessor.getMaxSection(); ++section) {
+ long sectionPos = SectionPos.asLong(chunkX, section, chunkZ);
+ this.updateDistanceTracking(sectionPos);
+ }
+ }
+
+ public void loadInPoiChunk(io.papermc.paper.chunk.system.poi.PoiChunk poiChunk) {
+ int chunkX = poiChunk.chunkX;
+ int chunkZ = poiChunk.chunkZ;
+ io.papermc.paper.util.TickThread.ensureTickThread(this.world, chunkX, chunkZ, "Loading poi chunk off-main");
+ for (int sectionY = this.levelHeightAccessor.getMinSection(); sectionY < this.levelHeightAccessor.getMaxSection(); ++sectionY) {
+ PoiSection section = poiChunk.getSection(sectionY);
+ if (section != null && !section.isEmpty()) {
+ this.onSectionLoad(SectionPos.asLong(chunkX, sectionY, chunkZ));
+ }
+ }
+ }
+ // Paper end - rewrite chunk system
+
public void checkConsistencyWithBlocks(ChunkPos chunkPos, LevelChunkSection chunkSection) {
SectionPos sectionPos = SectionPos.of(chunkPos, SectionPos.blockToSectionCoord(chunkSection.bottomBlockY()));
Util.ifElse(this.getOrLoad(sectionPos.asLong()), (poiSet) -> {
@@ -249,7 +350,7 @@ public class PoiManager extends SectionStorage<PoiSection> {
}).map((pair) -> {
return pair.getFirst().chunk();
}).filter((chunkPos) -> {
- return this.loadedChunks.add(chunkPos.toLong());
+ return true; // Paper - rewrite chunk system
}).forEach((chunkPos) -> {
world.getChunk(chunkPos.x, chunkPos.z, ChunkStatus.EMPTY);
});
@@ -265,7 +366,7 @@ public class PoiManager extends SectionStorage<PoiSection> {
@Override
protected int getLevelFromSource(long id) {
- return PoiManager.this.isVillageCenter(id) ? 0 : 7;
+ return PoiManager.this.isVillageCenter(id) ? 0 : 7; // Paper - rewrite chunk system - diff on change, this specifies the source level to use for distance tracking
}
@Override
@@ -288,6 +389,35 @@ public class PoiManager extends SectionStorage<PoiSection> {
}
}
+ // Paper start - Asynchronous chunk io
+ @javax.annotation.Nullable
+ @Override
+ public net.minecraft.nbt.CompoundTag read(ChunkPos chunkcoordintpair) throws java.io.IOException {
+ // Paper start - rewrite chunk system
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ return io.papermc.paper.chunk.system.io.RegionFileIOThread.loadData(
+ this.world, chunkcoordintpair.x, chunkcoordintpair.z, io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.getIOBlockingPriorityForCurrentThread()
+ );
+ }
+ // Paper end - rewrite chunk system
+ return super.read(chunkcoordintpair);
+ }
+
+ @Override
+ public void write(ChunkPos chunkcoordintpair, net.minecraft.nbt.CompoundTag nbttagcompound) throws java.io.IOException {
+ // Paper start - rewrite chunk system
+ if (!io.papermc.paper.chunk.system.io.RegionFileIOThread.isRegionFileThread()) {
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.scheduleSave(
+ this.world, chunkcoordintpair.x, chunkcoordintpair.z, nbttagcompound,
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.RegionFileType.POI_DATA);
+ return;
+ }
+ // Paper end - rewrite chunk system
+ super.write(chunkcoordintpair, nbttagcompound);
+ }
+ // Paper end
+
public static enum Occupancy {
HAS_SPACE(PoiRecord::hasSpace),
IS_OCCUPIED(PoiRecord::isOccupied),
diff --git a/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiSection.java b/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiSection.java
2022-12-07 19:32:25 +01:00
index 795a02941d7cecb58ec45b5e79c8d510ff21163a..3fc17817906876e83f040f908b8b1ba6cfa37b8b 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiSection.java
+++ b/src/main/java/net/minecraft/world/entity/ai/village/poi/PoiSection.java
@@ -29,6 +29,7 @@ public class PoiSection {
private final Map<Holder<PoiType>, Set<PoiRecord>> byType = Maps.newHashMap();
private final Runnable setDirty;
private boolean isValid;
+ public final Optional<PoiSection> noAllocateOptional = Optional.of(this); // Paper - rewrite chunk system
public static Codec<PoiSection> codec(Runnable updateListener) {
2022-12-07 19:32:25 +01:00
return RecordCodecBuilder.<PoiSection>create((instance) -> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@@ -46,6 +47,12 @@ public class PoiSection {
this(updateListener, true, ImmutableList.of());
}
+ // Paper start - isEmpty
+ public boolean isEmpty() {
+ return this.isValid && this.records.isEmpty() && this.byType.isEmpty();
+ }
+ // Paper end
+
private PoiSection(Runnable updateListener, boolean valid, List<PoiRecord> pois) {
this.setDirty = updateListener;
this.isValid = valid;
diff --git a/src/main/java/net/minecraft/world/level/EntityGetter.java b/src/main/java/net/minecraft/world/level/EntityGetter.java
2022-12-07 19:32:25 +01:00
index 127c4ebedb94631ceac92dbdcd465e904217d715..be6e3e21ad62da01e5e2dd78e300cbc8efdbeb42 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/EntityGetter.java
+++ b/src/main/java/net/minecraft/world/level/EntityGetter.java
@@ -18,6 +18,18 @@ import net.minecraft.world.phys.shapes.Shapes;
import net.minecraft.world.phys.shapes.VoxelShape;
public interface EntityGetter {
+
+ // Paper start
+ List<Entity> getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate);
+
+ void getEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into);
+
+ void getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into);
+
+ <T> void getEntitiesByClass(Class<? extends T> clazz, Entity except, final AABB box, List<? super T> into,
+ Predicate<? super T> predicate);
+ // Paper end
+
List<Entity> getEntities(@Nullable Entity except, AABB box, Predicate<? super Entity> predicate);
<T extends Entity> List<T> getEntities(EntityTypeTest<Entity, T> filter, AABB box, Predicate<? super T> predicate);
diff --git a/src/main/java/net/minecraft/world/level/Level.java b/src/main/java/net/minecraft/world/level/Level.java
2023-03-14 19:05:23 +01:00
index 69f55f7207b8c03ca886947c57c9d13a8e8eb576..0517c0932b79db1a0e27673654ae8bb5fd740425 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/Level.java
+++ b/src/main/java/net/minecraft/world/level/Level.java
2023-03-14 19:05:23 +01:00
@@ -460,6 +460,11 @@ public abstract class Level implements LevelAccessor, AutoCloseable {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
if ((i & 2) != 0 && (!this.isClientSide || (i & 4) == 0) && (this.isClientSide || chunk == null || (chunk.getFullStatus() != null && chunk.getFullStatus().isOrAfter(ChunkHolder.FullChunkStatus.TICKING)))) { // allow chunk to be null here as chunk.isReady() is false when we send our notification during block placement
this.sendBlockUpdated(blockposition, iblockdata1, iblockdata, i);
+ // Paper start - per player view distance - allow block updates for non-ticking chunks in player view distance
+ // if copied from above
2022-12-07 19:32:25 +01:00
+ } else if ((i & 2) != 0 && (!this.isClientSide || (i & 4) == 0) && (this.isClientSide || chunk == null || ((ServerLevel)this).getChunkSource().chunkMap.playerChunkManager.broadcastMap.getObjectsInRange(io.papermc.paper.util.MCUtil.getCoordinateKey(blockposition)) != null)) { // Paper - replace old player chunk management
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ((ServerLevel)this).getChunkSource().blockChanged(blockposition);
+ // Paper end - per player view distance
}
if ((i & 1) != 0) {
2023-03-14 19:05:23 +01:00
@@ -812,7 +817,7 @@ public abstract class Level implements LevelAccessor, AutoCloseable {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return this.capturedTileEntities.get(blockposition);
}
// CraftBukkit end
- return this.isOutsideBuildHeight(blockposition) ? null : (!this.isClientSide && Thread.currentThread() != this.thread ? null : this.getChunkAt(blockposition).getBlockEntity(blockposition, LevelChunk.EntityCreationType.IMMEDIATE));
+ return this.isOutsideBuildHeight(blockposition) ? null : (!this.isClientSide && !io.papermc.paper.util.TickThread.isTickThread() ? null : this.getChunkAt(blockposition).getBlockEntity(blockposition, LevelChunk.EntityCreationType.IMMEDIATE)); // Paper - rewrite chunk system
}
public void setBlockEntity(BlockEntity blockEntity) {
2023-03-14 19:05:23 +01:00
@@ -903,26 +908,7 @@ public abstract class Level implements LevelAccessor, AutoCloseable {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public List<Entity> getEntities(@Nullable Entity except, AABB box, Predicate<? super Entity> predicate) {
this.getProfiler().incrementCounter("getEntities");
List<Entity> list = Lists.newArrayList();
-
- this.getEntities().get(box, (entity1) -> {
- if (entity1 != except && predicate.test(entity1)) {
- list.add(entity1);
- }
-
- if (entity1 instanceof EnderDragon) {
- EnderDragonPart[] aentitycomplexpart = ((EnderDragon) entity1).getSubEntities();
- int i = aentitycomplexpart.length;
-
- for (int j = 0; j < i; ++j) {
- EnderDragonPart entitycomplexpart = aentitycomplexpart[j];
-
- if (entity1 != except && predicate.test(entitycomplexpart)) {
- list.add(entitycomplexpart);
- }
- }
- }
-
- });
+ ((ServerLevel)this).getEntityLookup().getEntities(except, box, list, predicate); // Paper - optimise this call
return list;
}
2023-03-14 19:05:23 +01:00
@@ -940,34 +926,23 @@ public abstract class Level implements LevelAccessor, AutoCloseable {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
2022-12-07 19:32:25 +01:00
public <T extends Entity> void getEntities(EntityTypeTest<Entity, T> filter, AABB box, Predicate<? super T> predicate, List<? super T> result, int limit) {
this.getProfiler().incrementCounter("getEntities");
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- this.getEntities().get(filter, box, (entity) -> {
- if (predicate.test(entity)) {
2022-12-07 19:32:25 +01:00
- result.add(entity);
- if (result.size() >= limit) {
- return AbortableIterationConsumer.Continuation.ABORT;
- }
2023-03-23 22:57:03 +01:00
- }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
- if (entity instanceof EnderDragon) {
- EnderDragon entityenderdragon = (EnderDragon) entity;
- EnderDragonPart[] aentitycomplexpart = entityenderdragon.getSubEntities();
2022-12-07 19:32:25 +01:00
- int j = aentitycomplexpart.length;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
2022-12-07 19:32:25 +01:00
- for (int k = 0; k < j; ++k) {
- EnderDragonPart entitycomplexpart = aentitycomplexpart[k];
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- T t0 = filter.tryCast(entitycomplexpart); // CraftBukkit - decompile error
-
- if (t0 != null && predicate.test(t0)) {
2022-12-07 19:32:25 +01:00
- result.add(t0);
- if (result.size() >= limit) {
- return AbortableIterationConsumer.Continuation.ABORT;
- }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- }
- }
2023-03-23 22:57:03 +01:00
+ // Paper start - optimise this call
+ //TODO use limit
+ if (filter instanceof net.minecraft.world.entity.EntityType entityTypeTest) {
+ ((ServerLevel) this).getEntityLookup().getEntities(entityTypeTest, box, result, predicate);
+ } else {
+ Predicate<? super T> test = (obj) -> {
+ return filter.tryCast(obj) != null;
+ };
+ predicate = predicate == null ? test : test.and((Predicate) predicate);
+ Class base;
+ if (filter == null || (base = filter.getBaseClass()) == null || base == Entity.class) {
+ ((ServerLevel) this).getEntityLookup().getEntities((Entity) null, box, (List) result, (Predicate)predicate);
+ } else {
+ ((ServerLevel) this).getEntityLookup().getEntities(base, null, box, (List) result, (Predicate)predicate); // Paper - optimise this call
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
-
2022-12-07 19:32:25 +01:00
- return AbortableIterationConsumer.Continuation.CONTINUE;
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- });
+ }
+ // Paper end - optimise this call
}
2022-12-07 19:32:25 +01:00
@Nullable
2023-03-14 19:05:23 +01:00
@@ -1305,4 +1280,45 @@ public abstract class Level implements LevelAccessor, AutoCloseable {
2022-12-07 19:32:25 +01:00
private ExplosionInteraction() {}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
+ // Paper start
+ //protected final io.papermc.paper.world.EntitySliceManager entitySliceManager; // Paper - rewrite chunk system
+
+ public org.bukkit.entity.Entity[] getChunkEntities(int chunkX, int chunkZ) {
+ io.papermc.paper.world.ChunkEntitySlices slices = ((ServerLevel)this).getEntityLookup().getChunk(chunkX, chunkZ);
+ if (slices == null) {
+ return new org.bukkit.entity.Entity[0];
+ }
+ return slices.getChunkEntities();
+ }
+
+ @Override
+ public List<Entity> getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate) {
+ List<Entity> ret = new java.util.ArrayList<>();
+ ((ServerLevel)this).getEntityLookup().getHardCollidingEntities(except, box, ret, predicate);
+ return ret;
+ }
+
+ @Override
+ public void getEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {
+ ((ServerLevel)this).getEntityLookup().getEntities(except, box, into, predicate);
+ }
+
+ @Override
+ public void getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {
+ ((ServerLevel)this).getEntityLookup().getHardCollidingEntities(except, box, into, predicate);
+ }
+
+ @Override
+ public <T> void getEntitiesByClass(Class<? extends T> clazz, Entity except, final AABB box, List<? super T> into,
+ Predicate<? super T> predicate) {
+ ((ServerLevel)this).getEntityLookup().getEntities((Class)clazz, except, box, (List)into, (Predicate)predicate);
+ }
+
+ @Override
+ public <T extends Entity> List<T> getEntitiesOfClass(Class<T> entityClass, AABB box, Predicate<? super T> predicate) {
+ List<T> ret = new java.util.ArrayList<>();
+ ((ServerLevel)this).getEntityLookup().getEntities(entityClass, null, box, ret, predicate);
+ return ret;
+ }
+ // Paper end
}
diff --git a/src/main/java/net/minecraft/world/level/chunk/ChunkGenerator.java b/src/main/java/net/minecraft/world/level/chunk/ChunkGenerator.java
2023-03-14 19:05:23 +01:00
index 8bb14230beb4b012f38997eec70934b96bae4db5..5ca3987683f7cecbce24bac434dc387bb5e9bf08 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/ChunkGenerator.java
+++ b/src/main/java/net/minecraft/world/level/chunk/ChunkGenerator.java
@@ -114,7 +114,7 @@ public abstract class ChunkGenerator {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return CompletableFuture.supplyAsync(Util.wrapThreadWithTaskName("init_biomes", () -> {
chunk.fillBiomesFromNoise(this.biomeSource, noiseConfig.sampler());
return chunk;
- }), Util.backgroundExecutor());
+ }), executor); // Paper - run with supplied executor
}
public abstract void applyCarvers(WorldGenRegion chunkRegion, long seed, RandomState noiseConfig, BiomeManager biomeAccess, StructureManager structureAccessor, ChunkAccess chunk, GenerationStep.Carving carverStep);
diff --git a/src/main/java/net/minecraft/world/level/chunk/ChunkStatus.java b/src/main/java/net/minecraft/world/level/chunk/ChunkStatus.java
2022-12-08 05:24:59 +01:00
index e2278ed457a7342d0d1b1a5fc1b5bdef6358816b..71df7c590e31932f2b8fc26a2afaaa54f52674ac 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/ChunkStatus.java
+++ b/src/main/java/net/minecraft/world/level/chunk/ChunkStatus.java
2022-12-07 19:32:25 +01:00
@@ -30,6 +30,30 @@ import net.minecraft.world.level.levelgen.structure.templatesystem.StructureTemp
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public class ChunkStatus {
+ // Paper start - rewrite chunk system
+ public boolean isParallelCapable; // Paper
+ public int writeRadius = -1;
+ public int loadRange = 0;
+
+ protected static final java.util.List<ChunkStatus> statuses = new java.util.ArrayList<>();
+
+ private ChunkStatus nextStatus;
+
+ public final ChunkStatus getNextStatus() {
+ return this.nextStatus;
+ }
+
+ public final boolean isEmptyLoadStatus() {
+ return this.loadingTask == PASSTHROUGH_LOAD_TASK;
+ }
+
+ public final boolean isEmptyGenStatus() {
+ return this == ChunkStatus.EMPTY || this == ChunkStatus.HEIGHTMAPS || this == ChunkStatus.LIQUID_CARVERS;
+ }
+
+ public final java.util.concurrent.atomic.AtomicBoolean warnedAboutNoImmediateComplete = new java.util.concurrent.atomic.AtomicBoolean();
+ // Paper end - rewrite chunk system
+
public static final int MAX_STRUCTURE_DISTANCE = 8;
private static final EnumSet<Heightmap.Types> PRE_FEATURES = EnumSet.of(Heightmap.Types.OCEAN_FLOOR_WG, Heightmap.Types.WORLD_SURFACE_WG);
public static final EnumSet<Heightmap.Types> POST_FEATURES = EnumSet.of(Heightmap.Types.OCEAN_FLOOR, Heightmap.Types.WORLD_SURFACE, Heightmap.Types.MOTION_BLOCKING, Heightmap.Types.MOTION_BLOCKING_NO_LEAVES);
2022-12-07 19:32:25 +01:00
@@ -151,10 +175,8 @@ public class ChunkStatus {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
protochunk.setStatus(chunkstatus);
}
- return lightenginethreaded.retainData(ichunkaccess).thenApply(Either::left);
- }, (chunkstatus, worldserver, structuretemplatemanager, lightenginethreaded, function, ichunkaccess) -> {
- return lightenginethreaded.retainData(ichunkaccess).thenApply(Either::left);
- });
+ return CompletableFuture.completedFuture(Either.left(ichunkaccess)); // Paper - rewrite chunk system
+ }); // Paper - rewrite chunk system
public static final ChunkStatus LIGHT = ChunkStatus.register("light", ChunkStatus.FEATURES, 1, ChunkStatus.POST_FEATURES, ChunkStatus.ChunkType.PROTOCHUNK, (chunkstatus, executor, worldserver, chunkgenerator, structuretemplatemanager, lightenginethreaded, function, list, ichunkaccess, flag) -> {
return ChunkStatus.lightChunk(chunkstatus, lightenginethreaded, ichunkaccess);
}, (chunkstatus, worldserver, structuretemplatemanager, lightenginethreaded, function, ichunkaccess) -> {
2022-12-07 19:32:25 +01:00
@@ -256,6 +278,13 @@ public class ChunkStatus {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.chunkType = chunkType;
this.heightmapsAfter = heightMapTypes;
this.index = previous == null ? 0 : previous.getIndex() + 1;
+ // Paper start
+ this.nextStatus = this;
+ if (statuses.size() > 0) {
+ statuses.get(statuses.size() - 1).nextStatus = this;
+ }
+ statuses.add(this);
+ // Paper end
}
public int getIndex() {
diff --git a/src/main/java/net/minecraft/world/level/chunk/LevelChunk.java b/src/main/java/net/minecraft/world/level/chunk/LevelChunk.java
index 975973e858ceb8340d2fdad54f636627d16612fd..a17ea775afb51dd31871a66a1b6f9f623dd7751a 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/LevelChunk.java
+++ b/src/main/java/net/minecraft/world/level/chunk/LevelChunk.java
@@ -176,6 +176,43 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
protected void onNeighbourChange(final long bitsetBefore, final long bitsetAfter) {
+ // Paper start - no-tick view distance
+ ServerChunkCache chunkProviderServer = ((ServerLevel)this.level).getChunkSource();
+ net.minecraft.server.level.ChunkMap chunkMap = chunkProviderServer.chunkMap;
+ // this code handles the addition of ticking tickets - the distance map handles the removal
+ if (!areNeighboursLoaded(bitsetBefore, 2) && areNeighboursLoaded(bitsetAfter, 2)) {
+ if (chunkMap.playerChunkManager.tickMap.getObjectsInRange(this.coordinateKey) != null) { // Paper - replace old player chunk loading system
+ // now we're ready for entity ticking
+ chunkProviderServer.mainThreadProcessor.execute(() -> {
+ // double check that this condition still holds.
+ if (LevelChunk.this.areNeighboursLoaded(2) && chunkMap.playerChunkManager.tickMap.getObjectsInRange(LevelChunk.this.coordinateKey) != null) { // Paper - replace old player chunk loading system
+ chunkMap.playerChunkManager.onChunkPlayerTickReady(this.chunkPos.x, this.chunkPos.z); // Paper - replace old player chunk
+ chunkProviderServer.addTicketAtLevel(net.minecraft.server.level.TicketType.PLAYER, LevelChunk.this.chunkPos, 31, LevelChunk.this.chunkPos); // 31 -> entity ticking, TODO check on update
+ }
+ });
+ }
+ }
+
+ // this code handles the chunk sending
+ if (!areNeighboursLoaded(bitsetBefore, 1) && areNeighboursLoaded(bitsetAfter, 1)) {
+ // Paper start - replace old player chunk loading system
+ if (chunkMap.playerChunkManager.isChunkNearPlayers(this.chunkPos.x, this.chunkPos.z)) {
+ // the post processing is expensive, so we don't want to run it unless we're actually near
+ // a player.
+ chunkProviderServer.mainThreadProcessor.execute(() -> {
+ if (!LevelChunk.this.areNeighboursLoaded(1)) {
+ return;
+ }
+ LevelChunk.this.postProcessGeneration();
+ if (!LevelChunk.this.areNeighboursLoaded(1)) {
+ return;
+ }
+ chunkMap.playerChunkManager.onChunkSendReady(this.chunkPos.x, this.chunkPos.z);
+ });
+ }
+ // Paper end - replace old player chunk loading system
+ }
+ // Paper end - no-tick view distance
}
public final boolean isAnyNeighborsLoaded() {
@@ -653,9 +690,26 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
- // CraftBukkit start
- public void loadCallback() {
- // Paper start - neighbour cache
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - new load callbacks
+ private io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder;
+ public io.papermc.paper.chunk.system.scheduling.NewChunkHolder getChunkHolder() {
+ return this.chunkHolder;
+ }
+
+ public void setChunkHolder(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ if (chunkHolder == null) {
+ throw new NullPointerException("Chunkholder cannot be null");
+ }
+ if (this.chunkHolder != null) {
+ throw new IllegalStateException("Already have chunkholder: " + this.chunkHolder + ", cannot replace with " + chunkHolder);
+ }
+ this.chunkHolder = chunkHolder;
+ this.playerChunk = chunkHolder.vanillaChunkHolder;
+ }
+
+ /* Note: We skip the light neighbour chunk loading done for the vanilla full chunk */
+ /* Starlight does not need these chunks for lighting purposes because of edge checks */
+ public void pushChunkIntoLoadedMap() {
int chunkX = this.chunkPos.x;
int chunkZ = this.chunkPos.z;
ServerChunkCache chunkProvider = this.level.getChunkSource();
@@ -670,10 +724,56 @@ public class LevelChunk extends ChunkAccess {
}
}
this.setNeighbourLoaded(0, 0, this);
+ this.level.getChunkSource().addLoadedChunk(this);
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ public void onChunkLoad(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ // figure out how this should interface with:
+ // the entity chunk load event // -> moved to the FULL status
+ // the chunk load event // -> stays here
+ // any entity add to world events // -> in FULL status
+ this.loadCallback();
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkBorder(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ public void onChunkUnload(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ // figure out how this should interface with:
+ // the entity chunk load event // -> moved to chunk unload to disk (not written yet)
+ // the chunk load event // -> stays here
+ // any entity add to world events // -> goes into the unload logic, it will completely explode
+ // etc later
+ this.unloadCallback();
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkNotBorder(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ public void onChunkTicking(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ this.postProcessGeneration();
+ this.level.startTickingChunk(this);
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkTicking(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ public void onChunkNotTicking(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkNotTicking(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ public void onChunkEntityTicking(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkEntityTicking(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+
+ public void onChunkNotEntityTicking(io.papermc.paper.chunk.system.scheduling.NewChunkHolder chunkHolder) {
+ io.papermc.paper.chunk.system.ChunkSystem.onChunkNotEntityTicking(this, chunkHolder.vanillaChunkHolder);
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ }
+ // Paper end - new load callbacks
+
+ // CraftBukkit start
+ public void loadCallback() {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ if (this.loadedTicketLevel) { LOGGER.error("Double calling chunk load!", new Throwable()); } // Paper
+ // Paper - rewrite chunk system - move into separate callback
this.loadedTicketLevel = true;
- // Paper end - neighbour cache
+ // Paper - rewrite chunk system - move into separate callback
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
org.bukkit.Server server = this.level.getCraftServer();
- this.level.getChunkSource().addLoadedChunk(this); // Paper
+ // Paper - rewrite chunk system - move into separate callback
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ((ServerLevel)this.level).getChunkSource().chunkMap.playerChunkManager.onChunkLoad(this.chunkPos.x, this.chunkPos.z); // Paper - rewrite player chunk management
if (server != null) {
/*
* If it's a new world, the first few chunks are generated inside
@@ -682,6 +782,7 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
*/
org.bukkit.Chunk bukkitChunk = new org.bukkit.craftbukkit.CraftChunk(this);
server.getPluginManager().callEvent(new org.bukkit.event.world.ChunkLoadEvent(bukkitChunk, this.needsDecoration));
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ this.chunkHolder.getEntityChunk().callEntitiesLoadEvent(); // Paper - rewrite chunk system
if (this.needsDecoration) {
try (co.aikar.timings.Timing ignored = this.level.timings.chunkLoadPopulate.startTiming()) { // Paper
@@ -710,9 +811,11 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public void unloadCallback() {
+ if (!this.loadedTicketLevel) { LOGGER.error("Double calling chunk unload!", new Throwable()); } // Paper
org.bukkit.Server server = this.level.getCraftServer();
+ this.chunkHolder.getEntityChunk().callEntitiesUnloadEvent(); // Paper - rewrite chunk system
org.bukkit.Chunk bukkitChunk = new org.bukkit.craftbukkit.CraftChunk(this);
- org.bukkit.event.world.ChunkUnloadEvent unloadEvent = new org.bukkit.event.world.ChunkUnloadEvent(bukkitChunk, this.isUnsaved());
+ org.bukkit.event.world.ChunkUnloadEvent unloadEvent = new org.bukkit.event.world.ChunkUnloadEvent(bukkitChunk, true); // Paper - rewrite chunk system - force save to true so that mustNotSave is correctly set below
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
server.getPluginManager().callEvent(unloadEvent);
// note: saving can be prevented, but not forced if no saving is actually required
this.mustNotSave = !unloadEvent.isSaveChunk();
@@ -734,9 +837,26 @@ public class LevelChunk extends ChunkAccess {
// Paper end
}
+ // Paper start - add dirty system to tick lists
+ @Override
+ public void setUnsaved(boolean needsSaving) {
+ if (!needsSaving) {
+ this.blockTicks.clearDirty();
+ this.fluidTicks.clearDirty();
+ }
+ super.setUnsaved(needsSaving);
+ }
+ // Paper end - add dirty system to tick lists
+
@Override
public boolean isUnsaved() {
- return super.isUnsaved() && !this.mustNotSave;
+ // Paper start - add dirty system to tick lists
+ long gameTime = this.level.getLevelData().getGameTime();
+ if (this.blockTicks.isDirty(gameTime) || this.fluidTicks.isDirty(gameTime)) {
+ return true;
+ }
+ // Paper end - add dirty system to tick lists
+ return super.isUnsaved(); // Paper - rewrite chunk system - do NOT clobber the dirty flag
}
// CraftBukkit end
@@ -811,7 +931,10 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
});
}
+ public boolean isPostProcessingDone; // Paper - replace chunk loader system
+
public void postProcessGeneration() {
+ try { // Paper - replace chunk loader system
ChunkPos chunkcoordintpair = this.getPos();
for (int i = 0; i < this.postProcessing.length; ++i) {
@@ -849,6 +972,11 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.pendingBlockEntities.clear();
this.upgradeData.upgrade(this);
+ } finally { // Paper start - replace chunk loader system
+ this.isPostProcessingDone = true;
+ this.level.getChunkSource().chunkMap.playerChunkManager.onChunkPostProcessing(this.chunkPos.x, this.chunkPos.z);
+ }
+ // Paper end - replace chunk loader system
}
@Nullable
@@ -898,7 +1026,7 @@ public class LevelChunk extends ChunkAccess {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
public ChunkHolder.FullChunkStatus getFullStatus() {
- return this.fullStatus == null ? ChunkHolder.FullChunkStatus.BORDER : (ChunkHolder.FullChunkStatus) this.fullStatus.get();
+ return this.chunkHolder == null ? ChunkHolder.FullChunkStatus.INACCESSIBLE : this.chunkHolder.getChunkStatus(); // Paper - rewrite chunk system
}
public void setFullStatus(Supplier<ChunkHolder.FullChunkStatus> levelTypeProvider) {
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/ChunkSerializer.java b/src/main/java/net/minecraft/world/level/chunk/storage/ChunkSerializer.java
2023-03-14 19:05:23 +01:00
index bca8084f7e76d036378705aa802217cc1bfdbc7d..f9dd2d9312297a727344d43150c5cab74128cbca 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/ChunkSerializer.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/ChunkSerializer.java
2022-12-07 19:32:25 +01:00
@@ -94,7 +94,31 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public ChunkSerializer() {}
+ // Paper start
+ public static final class InProgressChunkHolder {
+
+ public final ProtoChunk protoChunk;
+ public final java.util.ArrayDeque<Runnable> tasks;
+
+ public CompoundTag poiData;
+
+ public InProgressChunkHolder(final ProtoChunk protoChunk, final java.util.ArrayDeque<Runnable> tasks) {
+ this.protoChunk = protoChunk;
+ this.tasks = tasks;
+ }
+ }
+ // Paper end
+
public static ProtoChunk read(ServerLevel world, PoiManager poiStorage, ChunkPos chunkPos, CompoundTag nbt) {
+ // Paper start - add variant for async calls
+ InProgressChunkHolder holder = loadChunk(world, poiStorage, chunkPos, nbt, true);
+ holder.tasks.forEach(Runnable::run);
+ return holder.protoChunk;
+ }
+
+ public static InProgressChunkHolder loadChunk(ServerLevel world, PoiManager poiStorage, ChunkPos chunkPos, CompoundTag nbt, boolean distinguish) {
+ java.util.ArrayDeque<Runnable> tasksToExecuteOnMain = new java.util.ArrayDeque<>();
+ // Paper end
ChunkPos chunkcoordintpair1 = new ChunkPos(nbt.getInt("xPos"), nbt.getInt("zPos"));
if (!Objects.equals(chunkPos, chunkcoordintpair1)) {
2022-12-07 19:32:25 +01:00
@@ -158,7 +182,9 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
LevelChunkSection chunksection = new LevelChunkSection(b0, datapaletteblock, (PalettedContainer) object); // CraftBukkit - read/write
achunksection[k] = chunksection;
+ tasksToExecuteOnMain.add(() -> { // Paper - delay this task since we're executing off-main
poiStorage.checkConsistencyWithBlocks(chunkPos, chunksection);
+ }); // Paper - delay this task since we're executing off-main
}
boolean flag3 = nbttagcompound1.contains("BlockLight", 7);
2022-12-07 19:32:25 +01:00
@@ -319,7 +345,7 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
if (chunkstatus_type == ChunkStatus.ChunkType.LEVELCHUNK) {
- return new ImposterProtoChunk((LevelChunk) object1, false);
+ return new InProgressChunkHolder(new ImposterProtoChunk((LevelChunk) object1, false), tasksToExecuteOnMain); // Paper - Async chunk loading
} else {
ProtoChunk protochunk1 = (ProtoChunk) object1;
2023-03-23 22:57:03 +01:00
@@ -362,9 +388,41 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
protochunk1.setCarvingMask(worldgenstage_features, new CarvingMask(nbttagcompound4.getLongArray(s1), ((ChunkAccess) object1).getMinBuildHeight()));
}
- return protochunk1;
+ return new InProgressChunkHolder(protochunk1, tasksToExecuteOnMain); // Paper - Async chunk loading
2023-03-23 22:57:03 +01:00
+ }
+ }
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - async chunk save for unload
+ public record AsyncSaveData(
+ Tag blockTickList, // non-null if we had to go to the server's tick list
+ Tag fluidTickList, // non-null if we had to go to the server's tick list
+ ListTag blockEntities,
+ long worldTime
+ ) {}
+
+ // must be called sync
+ public static AsyncSaveData getAsyncSaveData(ServerLevel world, ChunkAccess chunk) {
+ org.spigotmc.AsyncCatcher.catchOp("preparation of chunk data for async save");
+
+ final CompoundTag tickLists = new CompoundTag();
+ ChunkSerializer.saveTicks(world, tickLists, chunk.getTicksForSerialization());
+
+ ListTag blockEntitiesSerialized = new ListTag();
+ for (final BlockPos blockPos : chunk.getBlockEntitiesPos()) {
+ final CompoundTag blockEntityNbt = chunk.getBlockEntityNbtForSaving(blockPos);
+ if (blockEntityNbt != null) {
+ blockEntitiesSerialized.add(blockEntityNbt);
+ }
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ return new AsyncSaveData(
+ tickLists.get(BLOCK_TICKS_TAG),
+ tickLists.get(FLUID_TICKS_TAG),
+ blockEntitiesSerialized,
+ world.getGameTime()
+ );
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper end
2023-03-23 22:57:03 +01:00
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private static void logErrors(ChunkPos chunkPos, int y, String message) {
ChunkSerializer.LOGGER.error("Recoverable errors when loading section [" + chunkPos.x + ", " + y + ", " + chunkPos.z + "]: " + message);
2022-12-07 19:32:25 +01:00
@@ -381,6 +439,11 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// CraftBukkit end
public static CompoundTag write(ServerLevel world, ChunkAccess chunk) {
+ // Paper start
+ return saveChunk(world, chunk, null);
+ }
+ public static CompoundTag saveChunk(ServerLevel world, ChunkAccess chunk, @org.checkerframework.checker.nullness.qual.Nullable AsyncSaveData asyncsavedata) {
+ // Paper end
// Paper start - rewrite light impl
final int minSection = io.papermc.paper.util.WorldUtil.getMinLightSection(world);
final int maxSection = io.papermc.paper.util.WorldUtil.getMaxLightSection(world);
2023-03-14 19:05:23 +01:00
@@ -393,7 +456,7 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
nbttagcompound.putInt("xPos", chunkcoordintpair.x);
nbttagcompound.putInt("yPos", chunk.getMinSection());
nbttagcompound.putInt("zPos", chunkcoordintpair.z);
- nbttagcompound.putLong("LastUpdate", world.getGameTime());
+ nbttagcompound.putLong("LastUpdate", asyncsavedata != null ? asyncsavedata.worldTime : world.getGameTime()); // Paper - async chunk unloading
nbttagcompound.putLong("InhabitedTime", chunk.getInhabitedTime());
nbttagcompound.putString("Status", chunk.getStatus().getName());
BlendingData blendingdata = chunk.getBlendingData();
2023-03-14 19:05:23 +01:00
@@ -493,8 +556,17 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
nbttagcompound.putBoolean("isLightOn", false); // Paper - set to false but still store, this allows us to detect --eraseCache (as eraseCache _removes_)
}
- ListTag nbttaglist1 = new ListTag();
- Iterator iterator = chunk.getBlockEntitiesPos().iterator();
+ // Paper start
+ ListTag nbttaglist1;
+ Iterator<BlockPos> iterator;
+ if (asyncsavedata != null) {
+ nbttaglist1 = asyncsavedata.blockEntities;
+ iterator = java.util.Collections.emptyIterator();
+ } else {
+ nbttaglist1 = new ListTag();
+ iterator = chunk.getBlockEntitiesPos().iterator();
+ }
+ // Paper end
CompoundTag nbttagcompound2;
2023-03-14 19:05:23 +01:00
@@ -531,7 +603,14 @@ public class ChunkSerializer {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
nbttagcompound.put("CarvingMasks", nbttagcompound2);
}
+ // Paper start
+ if (asyncsavedata != null) {
+ nbttagcompound.put(BLOCK_TICKS_TAG, asyncsavedata.blockTickList);
+ nbttagcompound.put(FLUID_TICKS_TAG, asyncsavedata.fluidTickList);
+ } else {
ChunkSerializer.saveTicks(world, nbttagcompound, chunk.getTicksForSerialization());
+ }
+ // Paper end
nbttagcompound.put("PostProcessing", ChunkSerializer.packOffsets(chunk.getPostProcessing()));
CompoundTag nbttagcompound3 = new CompoundTag();
Iterator iterator1 = chunk.getHeightmaps().iterator();
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/ChunkStorage.java b/src/main/java/net/minecraft/world/level/chunk/storage/ChunkStorage.java
2023-03-15 20:17:42 +01:00
index ba267f3082607669282a61f013991e1c391a3f09..0747d2d411a85a26bfea08cc1dd29df660620f67 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/ChunkStorage.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/ChunkStorage.java
@@ -28,26 +28,33 @@ import net.minecraft.world.level.storage.DimensionDataStorage;
public class ChunkStorage implements AutoCloseable {
public static final int LAST_MONOLYTH_STRUCTURE_DATA_VERSION = 1493;
- private final IOWorker worker;
+ // Paper - nuke IO worker
protected final DataFixer fixerUpper;
@Nullable
private volatile LegacyStructureDataHandler legacyStructureHandler;
+ // Paper start - async chunk loading
+ private final Object persistentDataLock = new Object(); // Paper
+ public final RegionFileStorage regionFileCache;
+ // Paper end - async chunk loading
public ChunkStorage(Path directory, DataFixer dataFixer, boolean dsync) {
this.fixerUpper = dataFixer;
- this.worker = new IOWorker(directory, dsync, "chunk");
+ // Paper start - async chunk io
+ // remove IO worker
+ this.regionFileCache = new RegionFileStorage(directory, dsync); // Paper - nuke IOWorker
+ // Paper end - async chunk io
}
public boolean isOldChunkAround(ChunkPos chunkPos, int checkRadius) {
- return this.worker.isOldChunkAround(chunkPos, checkRadius);
+ return true; // Paper - (for now, old unoptimised behavior) TODO implement later? the chunk status that blender uses SHOULD already have this radius loaded, no need to go back for it...
}
// CraftBukkit start
private boolean check(ServerChunkCache cps, int x, int z) {
ChunkPos pos = new ChunkPos(x, z);
if (cps != null) {
- com.google.common.base.Preconditions.checkState(org.bukkit.Bukkit.isPrimaryThread(), "primary thread");
- if (cps.hasChunk(x, z)) {
+ //com.google.common.base.Preconditions.checkState(org.bukkit.Bukkit.isPrimaryThread(), "primary thread"); // Paper - this function is now MT-Safe
+ if (cps.getChunkAtIfCachedImmediately(x, z) != null) { // Paper - isLoaded is a ticket level check, not a chunk loaded check!
return true;
}
}
@@ -75,6 +82,7 @@ public class ChunkStorage implements AutoCloseable {
public CompoundTag upgradeChunkTag(ResourceKey<LevelStem> resourcekey, Supplier<DimensionDataStorage> supplier, CompoundTag nbttagcompound, Optional<ResourceKey<Codec<? extends ChunkGenerator>>> optional, ChunkPos pos, @Nullable LevelAccessor generatoraccess) {
// CraftBukkit end
+ nbttagcompound = nbttagcompound.copy(); // Paper - defensive copy, another thread might modify this
int i = ChunkStorage.getVersion(nbttagcompound);
// CraftBukkit start
@@ -92,9 +100,11 @@ public class ChunkStorage implements AutoCloseable {
if (i < 1493) {
ca.spottedleaf.dataconverter.minecraft.MCDataConverter.convertTag(ca.spottedleaf.dataconverter.minecraft.datatypes.MCTypeRegistry.CHUNK, nbttagcompound, i, 1493); // Paper - replace chunk converter
if (nbttagcompound.getCompound("Level").getBoolean("hasLegacyStructureData")) {
+ synchronized (this.persistentDataLock) { // Paper - Async chunk loading
LegacyStructureDataHandler persistentstructurelegacy = this.getLegacyStructureHandler(resourcekey, supplier);
nbttagcompound = persistentstructurelegacy.updateFromLegacy(nbttagcompound);
+ } // Paper - Async chunk loading
}
}
@@ -127,7 +137,7 @@ public class ChunkStorage implements AutoCloseable {
LegacyStructureDataHandler persistentstructurelegacy = this.legacyStructureHandler;
if (persistentstructurelegacy == null) {
- synchronized (this) {
+ synchronized (this.persistentDataLock) { // Paper - async chunk loading
persistentstructurelegacy = this.legacyStructureHandler;
if (persistentstructurelegacy == null) {
2022-11-12 21:57:41 +01:00
this.legacyStructureHandler = persistentstructurelegacy = LegacyStructureDataHandler.getLegacyStructureHandler(worldKey, (DimensionDataStorage) stateManagerGetter.get());
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@@ -153,26 +163,49 @@ public class ChunkStorage implements AutoCloseable {
}
public CompletableFuture<Optional<CompoundTag>> read(ChunkPos chunkPos) {
- return this.worker.loadAsync(chunkPos);
+ // Paper start - async chunk io
+ try {
+ return CompletableFuture.completedFuture(Optional.ofNullable(this.readSync(chunkPos)));
+ } catch (Throwable thr) {
+ return CompletableFuture.failedFuture(thr);
+ }
2023-03-23 22:57:03 +01:00
+ }
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ @Nullable
+ public CompoundTag readSync(ChunkPos chunkPos) throws IOException {
+ return this.regionFileCache.read(chunkPos);
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper end - async chunk io
- public void write(ChunkPos chunkPos, CompoundTag nbt) {
- this.worker.store(chunkPos, nbt);
+ // Paper start - async chunk io
+ public void write(ChunkPos chunkPos, CompoundTag nbt) throws IOException {
+ this.regionFileCache.write(chunkPos, nbt);
+ // Paper end - Async chunk loading
if (this.legacyStructureHandler != null) {
+ synchronized (this.persistentDataLock) { // Paper - Async chunk loading
this.legacyStructureHandler.removeIndex(chunkPos.toLong());
+ } // Paper - Async chunk loading
}
}
public void flushWorker() {
- this.worker.synchronize(true).join();
+ io.papermc.paper.chunk.system.io.RegionFileIOThread.flush(); // Paper - rewrite chunk system
}
public void close() throws IOException {
- this.worker.close();
+ this.regionFileCache.close(); // Paper - nuke IO worker
}
public ChunkScanAccess chunkScanner() {
- return this.worker;
+ // Paper start - nuke IO worker
+ return ((chunkPos, streamTagVisitor) -> {
+ try {
+ this.regionFileCache.scanChunk(chunkPos, streamTagVisitor);
+ return java.util.concurrent.CompletableFuture.completedFuture(null);
+ } catch (IOException e) {
+ throw new RuntimeException(e);
+ }
+ });
+ // Paper end
}
}
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/EntityStorage.java b/src/main/java/net/minecraft/world/level/chunk/storage/EntityStorage.java
2023-03-15 20:17:42 +01:00
index 98b3909b536f11eda9c481ffd74066ad0cdb0ebc..0ec0be22f7292d57c40da6f1f4575bdebf8dbd09 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/EntityStorage.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/EntityStorage.java
2023-03-14 19:05:23 +01:00
@@ -30,43 +30,31 @@ public class EntityStorage implements EntityPersistentStorage<Entity> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private static final String ENTITIES_TAG = "Entities";
private static final String POSITION_TAG = "Position";
public final ServerLevel level;
- private final IOWorker worker;
+ // Paper - rewrite chunk system
private final LongSet emptyChunks = new LongOpenHashSet();
- public final ProcessorMailbox<Runnable> entityDeserializerQueue;
+ // Paper - rewrite chunk system
protected final DataFixer fixerUpper;
public EntityStorage(ServerLevel world, Path path, DataFixer dataFixer, boolean dsync, Executor executor) {
this.level = world;
this.fixerUpper = dataFixer;
- this.entityDeserializerQueue = ProcessorMailbox.create(executor, "entity-deserializer");
- this.worker = new IOWorker(path, dsync, "entities");
+ // Paper - rewrite chunk system
}
@Override
public CompletableFuture<ChunkEntities<Entity>> loadEntities(ChunkPos pos) {
- return this.emptyChunks.contains(pos.toLong()) ? CompletableFuture.completedFuture(emptyChunk(pos)) : this.worker.loadAsync(pos).thenApplyAsync((nbt) -> {
- if (nbt.isEmpty()) {
- this.emptyChunks.add(pos.toLong());
- return emptyChunk(pos);
- } else {
- try {
- ChunkPos chunkPos2 = readChunkPos(nbt.get());
- if (!Objects.equals(pos, chunkPos2)) {
- LOGGER.error("Chunk file at {} is in the wrong location. (Expected {}, got {})", pos, pos, chunkPos2);
- }
- } catch (Exception var6) {
- LOGGER.warn("Failed to parse chunk {} position info", pos, var6);
- }
-
- CompoundTag compoundTag = this.upgradeChunkTag(nbt.get());
- ListTag listTag = compoundTag.getList("Entities", 10);
- List<Entity> list = EntityType.loadEntitiesRecursive(listTag, this.level).collect(ImmutableList.toImmutableList());
- return new ChunkEntities<>(pos, list);
- }
- }, this.entityDeserializerQueue::tell);
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system - copy out read logic into readEntities
2023-03-23 22:57:03 +01:00
+ }
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper start - rewrite chunk system
+ public static List<Entity> readEntities(ServerLevel level, CompoundTag compoundTag) {
+ ListTag listTag = compoundTag.getList("Entities", 10);
+ List<Entity> list = EntityType.loadEntitiesRecursive(listTag, level).collect(ImmutableList.toImmutableList());
+ return list;
2023-03-23 22:57:03 +01:00
}
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ // Paper end - rewrite chunk system
2023-03-23 22:57:03 +01:00
- private static ChunkPos readChunkPos(CompoundTag chunkNbt) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ public static ChunkPos readChunkPos(CompoundTag chunkNbt) { // Paper - public
int[] is = chunkNbt.getIntArray("Position");
return new ChunkPos(is[0], is[1]);
}
2023-03-14 19:05:23 +01:00
@@ -81,45 +69,75 @@ public class EntityStorage implements EntityPersistentStorage<Entity> {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public void storeEntities(ChunkEntities<Entity> dataList) {
+ // Paper start - rewrite chunk system
+ if (true) {
+ throw new UnsupportedOperationException();
+ }
+ // Paper end - rewrite chunk system
ChunkPos chunkPos = dataList.getPos();
if (dataList.isEmpty()) {
if (this.emptyChunks.add(chunkPos.toLong())) {
- this.worker.store(chunkPos, (CompoundTag)null);
+ // Paper - rewrite chunk system
}
} else {
- ListTag listTag = new ListTag();
- dataList.getEntities().forEach((entity) -> {
- CompoundTag compoundTag = new CompoundTag();
- if (entity.save(compoundTag)) {
- listTag.add(compoundTag);
- }
-
- });
2023-03-14 19:05:23 +01:00
- CompoundTag compoundTag = NbtUtils.addCurrentDataVersion(new CompoundTag());
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
- compoundTag.put("Entities", listTag);
- writeChunkPos(compoundTag, chunkPos);
- this.worker.store(chunkPos, compoundTag).exceptionally((ex) -> {
- LOGGER.error("Failed to store chunk {}", chunkPos, ex);
- return null;
- });
+ // Paper - move into saveEntityChunk0
this.emptyChunks.remove(chunkPos.toLong());
}
}
+ // Paper start - rewrite chunk system
+ public static void copyEntities(final CompoundTag from, final CompoundTag into) {
+ if (from == null) {
+ return;
+ }
+ final ListTag entitiesFrom = from.getList("Entities", net.minecraft.nbt.Tag.TAG_COMPOUND);
+ if (entitiesFrom == null || entitiesFrom.isEmpty()) {
+ return;
+ }
+
+ final ListTag entitiesInto = into.getList("Entities", net.minecraft.nbt.Tag.TAG_COMPOUND);
+ into.put("Entities", entitiesInto); // this is in case into doesn't have any entities
+ entitiesInto.addAll(0, entitiesFrom.copy()); // need to copy, this is coming from the save thread
2023-03-23 22:57:03 +01:00
+ }
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ public static CompoundTag saveEntityChunk(List<Entity> entities, ChunkPos chunkPos, ServerLevel level) {
+ return saveEntityChunk0(entities, chunkPos, level, false);
+ }
+ private static CompoundTag saveEntityChunk0(List<Entity> entities, ChunkPos chunkPos, ServerLevel level, boolean force) {
+ if (!force && entities.isEmpty()) {
+ return null;
+ }
2023-03-14 19:05:23 +01:00
+
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ ListTag listTag = new ListTag();
+ entities.forEach((entity) -> { // diff here: use entities parameter
+ CompoundTag compoundTag = new CompoundTag();
+ if (entity.save(compoundTag)) {
+ listTag.add(compoundTag);
+ }
+
+ });
2023-03-14 19:05:23 +01:00
+ CompoundTag compoundTag = NbtUtils.addCurrentDataVersion(new CompoundTag());
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ compoundTag.put("Entities", listTag);
+ writeChunkPos(compoundTag, chunkPos);
2023-03-14 19:05:23 +01:00
+ // Paper - remove worker usage
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+
+ return !force && listTag.isEmpty() ? null : compoundTag;
+ }
+ // Paper end - rewrite chunk system
+
2023-03-23 22:57:03 +01:00
@Override
public void flush(boolean sync) {
- this.worker.synchronize(sync).join();
- this.entityDeserializerQueue.runAll();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
2023-03-23 22:57:03 +01:00
}
- private CompoundTag upgradeChunkTag(CompoundTag chunkNbt) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
+ public static CompoundTag upgradeChunkTag(CompoundTag chunkNbt) { // Paper - public and static
2023-03-14 19:05:23 +01:00
int i = NbtUtils.getDataVersion(chunkNbt, -1);
2023-03-15 20:17:42 +01:00
return ca.spottedleaf.dataconverter.minecraft.MCDataConverter.convertTag(ca.spottedleaf.dataconverter.minecraft.datatypes.MCTypeRegistry.ENTITY_CHUNK, chunkNbt, i, net.minecraft.SharedConstants.getCurrentVersion().getDataVersion().getVersion()); // Paper - route to new converter system
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
@Override
public void close() throws IOException {
- this.worker.close();
+ throw new UnsupportedOperationException(); // Paper - rewrite chunk system
}
}
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/RegionFile.java b/src/main/java/net/minecraft/world/level/chunk/storage/RegionFile.java
2022-11-12 21:57:41 +01:00
index d9daf07132c46548964a75588b69d7a74680e917..e68205fe7169c7c5b7c6fdada2ee97d86107ca97 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/RegionFile.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/RegionFile.java
@@ -44,6 +44,7 @@ public class RegionFile implements AutoCloseable {
private final IntBuffer timestamps;
@VisibleForTesting
protected final RegionBitmap usedSectors;
+ public final java.util.concurrent.locks.ReentrantLock fileLock = new java.util.concurrent.locks.ReentrantLock(true); // Paper
public RegionFile(Path file, Path directory, boolean dsync) throws IOException {
this(file, directory, RegionFileVersion.VERSION_DEFLATE, dsync);
@@ -228,7 +229,7 @@ public class RegionFile implements AutoCloseable {
return (byteCount + 4096 - 1) / 4096;
}
- public boolean doesChunkExist(ChunkPos pos) {
+ public synchronized boolean doesChunkExist(ChunkPos pos) { // Paper - synchronized
int i = this.getOffset(pos);
if (i == 0) {
@@ -393,6 +394,11 @@ public class RegionFile implements AutoCloseable {
}
public void close() throws IOException {
+ // Paper start - Prevent regionfiles from being closed during use
+ this.fileLock.lock();
+ synchronized (this) {
+ try {
+ // Paper end
try {
this.padToFullSector();
} finally {
@@ -402,6 +408,10 @@ public class RegionFile implements AutoCloseable {
this.file.close();
}
}
+ } finally { // Paper start - Prevent regionfiles from being closed during use
+ this.fileLock.unlock();
+ }
+ } // Paper end
}
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/RegionFileStorage.java b/src/main/java/net/minecraft/world/level/chunk/storage/RegionFileStorage.java
2022-12-07 19:32:25 +01:00
index b9b50c56e79297bb824a92355f437a5d4d7e6760..18ef7025f7f4dc2a4aff85ca65ff5a2d35a1ef06 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/RegionFileStorage.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/RegionFileStorage.java
@@ -24,16 +24,37 @@ public class RegionFileStorage implements AutoCloseable {
private final Path folder;
private final boolean sync;
- RegionFileStorage(Path directory, boolean dsync) {
+ protected RegionFileStorage(Path directory, boolean dsync) { // Paper - protected constructor
this.folder = directory;
this.sync = dsync;
}
- private RegionFile getRegionFile(ChunkPos chunkcoordintpair, boolean existingOnly) throws IOException { // CraftBukkit
+ // Paper start
+ public synchronized RegionFile getRegionFileIfLoaded(ChunkPos chunkcoordintpair) {
+ return this.regionCache.getAndMoveToFirst(ChunkPos.asLong(chunkcoordintpair.getRegionX(), chunkcoordintpair.getRegionZ()));
+ }
+
+ public synchronized boolean chunkExists(ChunkPos pos) throws IOException {
+ RegionFile regionfile = getRegionFile(pos, true);
+
+ return regionfile != null ? regionfile.hasChunk(pos) : false;
+ }
+
+ public synchronized RegionFile getRegionFile(ChunkPos chunkcoordintpair, boolean existingOnly) throws IOException { // CraftBukkit
+ return this.getRegionFile(chunkcoordintpair, existingOnly, false);
+ }
+ public synchronized RegionFile getRegionFile(ChunkPos chunkcoordintpair, boolean existingOnly, boolean lock) throws IOException {
+ // Paper end
long i = ChunkPos.asLong(chunkcoordintpair.getRegionX(), chunkcoordintpair.getRegionZ());
RegionFile regionfile = (RegionFile) this.regionCache.getAndMoveToFirst(i);
if (regionfile != null) {
+ // Paper start
+ if (lock) {
+ // must be in this synchronized block
+ regionfile.fileLock.lock();
+ }
+ // Paper end
return regionfile;
} else {
if (this.regionCache.size() >= 256) {
@@ -48,6 +69,12 @@ public class RegionFileStorage implements AutoCloseable {
RegionFile regionfile1 = new RegionFile(path1, this.folder, this.sync);
this.regionCache.putAndMoveToFirst(i, regionfile1);
+ // Paper start
+ if (lock) {
+ // must be in this synchronized block
+ regionfile1.fileLock.lock();
+ }
+ // Paper end
return regionfile1;
}
}
@@ -55,11 +82,12 @@ public class RegionFileStorage implements AutoCloseable {
@Nullable
public CompoundTag read(ChunkPos pos) throws IOException {
// CraftBukkit start - SPIGOT-5680: There's no good reason to preemptively create files on read, save that for writing
- RegionFile regionfile = this.getRegionFile(pos, true);
+ RegionFile regionfile = this.getRegionFile(pos, true, true); // Paper
if (regionfile == null) {
return null;
}
// CraftBukkit end
+ try { // Paper
DataInputStream datainputstream = regionfile.getChunkDataInputStream(pos);
CompoundTag nbttagcompound;
@@ -96,6 +124,9 @@ public class RegionFileStorage implements AutoCloseable {
}
return nbttagcompound;
+ } finally { // Paper start
+ regionfile.fileLock.unlock();
+ } // Paper end
}
2022-12-07 19:32:25 +01:00
public void scanChunk(ChunkPos chunkPos, StreamTagVisitor scanner) throws IOException {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@@ -130,7 +161,12 @@ public class RegionFileStorage implements AutoCloseable {
}
protected void write(ChunkPos pos, @Nullable CompoundTag nbt) throws IOException {
- RegionFile regionfile = this.getRegionFile(pos, false); // CraftBukkit
+ RegionFile regionfile = this.getRegionFile(pos, nbt == null, true); // CraftBukkit // Paper // Paper start - rewrite chunk system
+ if (nbt == null && regionfile == null) {
+ return;
+ }
+ // Paper end - rewrite chunk system
+ try { // Paper
if (nbt == null) {
regionfile.clear(pos);
@@ -156,9 +192,12 @@ public class RegionFileStorage implements AutoCloseable {
}
}
+ } finally { // Paper start
+ regionfile.fileLock.unlock();
+ } // Paper end
}
- public void close() throws IOException {
+ public synchronized void close() throws IOException { // Paper -> synchronized
ExceptionCollector<IOException> exceptionsuppressor = new ExceptionCollector<>();
ObjectIterator objectiterator = this.regionCache.values().iterator();
@@ -175,7 +214,7 @@ public class RegionFileStorage implements AutoCloseable {
exceptionsuppressor.throwIfPresent();
}
- public void flush() throws IOException {
+ public synchronized void flush() throws IOException { // Paper - synchronize
ObjectIterator objectiterator = this.regionCache.values().iterator();
while (objectiterator.hasNext()) {
diff --git a/src/main/java/net/minecraft/world/level/chunk/storage/SectionStorage.java b/src/main/java/net/minecraft/world/level/chunk/storage/SectionStorage.java
index b0c8a0e64c7a5d41c1b4cc1e39c4399c142b56af..0887cba39bfc4279abec21c6c316abab28beb0a3 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/chunk/storage/SectionStorage.java
+++ b/src/main/java/net/minecraft/world/level/chunk/storage/SectionStorage.java
@@ -34,27 +34,28 @@ import net.minecraft.world.level.ChunkPos;
import net.minecraft.world.level.LevelHeightAccessor;
import org.slf4j.Logger;
-public class SectionStorage<R> implements AutoCloseable {
+public class SectionStorage<R> extends RegionFileStorage implements AutoCloseable { // Paper - nuke IOWorker
private static final Logger LOGGER = LogUtils.getLogger();
private static final String SECTIONS_TAG = "Sections";
- private final IOWorker worker;
+ // Paper - remove mojang I/O thread
private final Long2ObjectMap<Optional<R>> storage = new Long2ObjectOpenHashMap<>();
private final LongLinkedOpenHashSet dirty = new LongLinkedOpenHashSet();
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
private final Function<Runnable, Codec<R>> codec;
private final Function<Runnable, R> factory;
private final DataFixer fixerUpper;
private final DataFixTypes type;
- private final RegistryAccess registryAccess;
+ public final RegistryAccess registryAccess; // Paper - rewrite chunk system
protected final LevelHeightAccessor levelHeightAccessor;
public SectionStorage(Path path, Function<Runnable, Codec<R>> codecFactory, Function<Runnable, R> factory, DataFixer dataFixer, DataFixTypes dataFixTypes, boolean dsync, RegistryAccess dynamicRegistryManager, LevelHeightAccessor world) {
+ super(path, dsync); // Paper - remove mojang I/O thread
this.codec = codecFactory;
this.factory = factory;
this.fixerUpper = dataFixer;
this.type = dataFixTypes;
this.registryAccess = dynamicRegistryManager;
this.levelHeightAccessor = world;
- this.worker = new IOWorker(path, dsync, path.getFileName().toString());
+ // Paper - remove mojang I/O thread
}
protected void tick(BooleanSupplier shouldKeepTicking) {
@@ -116,23 +117,21 @@ public class SectionStorage<R> implements AutoCloseable {
}
private void readColumn(ChunkPos pos) {
- Optional<CompoundTag> optional = this.tryRead(pos).join();
- RegistryOps<Tag> registryOps = RegistryOps.create(NbtOps.INSTANCE, this.registryAccess);
- this.readColumn(pos, registryOps, optional.orElse((CompoundTag)null));
+ throw new IllegalStateException("Only chunk system can load in state, offending class:" + this.getClass().getName()); // Paper - rewrite chunk system
}
private CompletableFuture<Optional<CompoundTag>> tryRead(ChunkPos pos) {
- return this.worker.loadAsync(pos).exceptionally((throwable) -> {
- if (throwable instanceof IOException iOException) {
- LOGGER.error("Error reading chunk {} data from disk", pos, iOException);
- return Optional.empty();
- } else {
- throw new CompletionException(throwable);
- }
- });
+ // Paper start - rewrite chunk system
+ try {
+ return CompletableFuture.completedFuture(Optional.ofNullable(this.read(pos)));
+ } catch (Throwable thr) {
+ return CompletableFuture.failedFuture(thr);
+ }
+ // Paper end - rewrite chunk system
}
private <T> void readColumn(ChunkPos pos, DynamicOps<T> ops, @Nullable T data) {
+ if (true) throw new IllegalStateException("Only chunk system can load in state, offending class:" + this.getClass().getName()); // Paper - rewrite chunk system
if (data == null) {
for(int i = this.levelHeightAccessor.getMinSection(); i < this.levelHeightAccessor.getMaxSection(); ++i) {
this.storage.put(getKey(pos, i), Optional.empty());
@@ -177,7 +176,7 @@ public class SectionStorage<R> implements AutoCloseable {
Dynamic<Tag> dynamic = this.writeColumn(pos, registryOps);
Tag tag = dynamic.getValue();
if (tag instanceof CompoundTag) {
- this.worker.store(pos, (CompoundTag)tag);
+ try { this.write(pos, (CompoundTag)tag); } catch (IOException ioexception) { SectionStorage.LOGGER.error("Error writing data to disk", ioexception); } // Paper - nuke IOWorker
} else {
LOGGER.error("Expected compound tag, got {}", (Object)tag);
}
@@ -222,7 +221,7 @@ public class SectionStorage<R> implements AutoCloseable {
}
private static int getVersion(Dynamic<?> dynamic) {
- return dynamic.get("DataVersion").asInt(1945);
+ return dynamic.get("DataVersion").asInt(1945); // Paper - diff on change, constant used in ChunkLoadTask
}
public void flush(ChunkPos pos) {
@@ -240,6 +239,9 @@ public class SectionStorage<R> implements AutoCloseable {
@Override
public void close() throws IOException {
- this.worker.close();
+ //this.worker.close(); // Paper - nuke I/O worker - don't call the worker
+ super.close(); // Paper - nuke I/O worker - call super.close method which is responsible for closing used files.
}
+
+ // Paper - rewrite chunk system
}
diff --git a/src/main/java/net/minecraft/world/level/entity/EntityTickList.java b/src/main/java/net/minecraft/world/level/entity/EntityTickList.java
index 2830d32bba3dc85847e3a5d9b4d98f822e34b606..4cdfc433df67afcd455422e9baf56f167dd712ae 100644
--- a/src/main/java/net/minecraft/world/level/entity/EntityTickList.java
+++ b/src/main/java/net/minecraft/world/level/entity/EntityTickList.java
@@ -8,54 +8,42 @@ import javax.annotation.Nullable;
import net.minecraft.world.entity.Entity;
public class EntityTickList {
- private Int2ObjectMap<Entity> active = new Int2ObjectLinkedOpenHashMap<>();
- private Int2ObjectMap<Entity> passive = new Int2ObjectLinkedOpenHashMap<>();
- @Nullable
- private Int2ObjectMap<Entity> iterated;
+ private final io.papermc.paper.util.maplist.IteratorSafeOrderedReferenceSet<Entity> entities = new io.papermc.paper.util.maplist.IteratorSafeOrderedReferenceSet<>(true); // Paper - rewrite this, always keep this updated - why would we EVER tick an entity that's not ticking?
private void ensureActiveIsNotIterated() {
- if (this.iterated == this.active) {
- this.passive.clear();
-
- for(Int2ObjectMap.Entry<Entity> entry : Int2ObjectMaps.fastIterable(this.active)) {
- this.passive.put(entry.getIntKey(), entry.getValue());
- }
-
- Int2ObjectMap<Entity> int2ObjectMap = this.active;
- this.active = this.passive;
- this.passive = int2ObjectMap;
- }
+ // Paper - replace with better logic, do not delay removals
}
public void add(Entity entity) {
+ io.papermc.paper.util.TickThread.ensureTickThread("Asynchronous entity ticklist addition"); // Paper
this.ensureActiveIsNotIterated();
- this.active.put(entity.getId(), entity);
+ this.entities.add(entity); // Paper - replace with better logic, do not delay removals/additions
}
public void remove(Entity entity) {
+ io.papermc.paper.util.TickThread.ensureTickThread("Asynchronous entity ticklist removal"); // Paper
this.ensureActiveIsNotIterated();
- this.active.remove(entity.getId());
+ this.entities.remove(entity); // Paper - replace with better logic, do not delay removals/additions
}
public boolean contains(Entity entity) {
- return this.active.containsKey(entity.getId());
+ return this.entities.contains(entity); // Paper - replace with better logic, do not delay removals/additions
}
public void forEach(Consumer<Entity> action) {
- if (this.iterated != null) {
- throw new UnsupportedOperationException("Only one concurrent iteration supported");
- } else {
- this.iterated = this.active;
-
- try {
- for(Entity entity : this.active.values()) {
- action.accept(entity);
- }
- } finally {
- this.iterated = null;
+ io.papermc.paper.util.TickThread.ensureTickThread("Asynchronous entity ticklist iteration"); // Paper
+ // Paper start - replace with better logic, do not delay removals/additions
+ // To ensure nothing weird happens with dimension travelling, do not iterate over new entries...
+ // (by dfl iterator() is configured to not iterate over new entries)
+ io.papermc.paper.util.maplist.IteratorSafeOrderedReferenceSet.Iterator<Entity> iterator = this.entities.iterator();
+ try {
+ while (iterator.hasNext()) {
+ action.accept(iterator.next());
}
-
+ } finally {
+ iterator.finishedIterating();
}
+ // Paper end - replace with better logic, do not delay removals/additions
}
}
diff --git a/src/main/java/net/minecraft/world/level/levelgen/NoiseBasedChunkGenerator.java b/src/main/java/net/minecraft/world/level/levelgen/NoiseBasedChunkGenerator.java
2023-03-14 19:05:23 +01:00
index ea4e005e4d394490f45ebf626d2229f7b4624294..4f3387d1b128bc98614cfabfb5306781bdcd93be 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/net/minecraft/world/level/levelgen/NoiseBasedChunkGenerator.java
+++ b/src/main/java/net/minecraft/world/level/levelgen/NoiseBasedChunkGenerator.java
2022-12-07 19:32:25 +01:00
@@ -87,7 +87,7 @@ public final class NoiseBasedChunkGenerator extends ChunkGenerator {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return CompletableFuture.supplyAsync(Util.wrapThreadWithTaskName("init_biomes", () -> {
this.doCreateBiomes(blender, noiseConfig, structureAccessor, chunk);
return chunk;
- }), Util.backgroundExecutor());
+ }), executor); // Paper - run with supplied executor
}
private void doCreateBiomes(Blender blender, RandomState noiseConfig, StructureManager structureAccessor, ChunkAccess chunk) {
2022-12-07 19:32:25 +01:00
@@ -286,7 +286,7 @@ public final class NoiseBasedChunkGenerator extends ChunkGenerator {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return CompletableFuture.supplyAsync(Util.wrapThreadWithTaskName("wgen_fill_noise", () -> {
return this.doFill(blender, structureAccessor, noiseConfig, chunk, j, k);
- }), Util.backgroundExecutor()).whenCompleteAsync((ichunkaccess1, throwable) -> {
+ }), executor).whenCompleteAsync((ichunkaccess1, throwable) -> { // Paper - run with supplied executor
Iterator iterator = set.iterator();
while (iterator.hasNext()) {
diff --git a/src/main/java/net/minecraft/world/ticks/LevelChunkTicks.java b/src/main/java/net/minecraft/world/ticks/LevelChunkTicks.java
index 9f6c2e5b5d9e8d714a47c770e255d06c0ef7c190..ac807277a6b26d140ea9873d17c7aa4fb5fe37b2 100644
--- a/src/main/java/net/minecraft/world/ticks/LevelChunkTicks.java
+++ b/src/main/java/net/minecraft/world/ticks/LevelChunkTicks.java
@@ -25,6 +25,19 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
@Nullable
private BiConsumer<LevelChunkTicks<T>, ScheduledTick<T>> onTickAdded;
+ // Paper start - add dirty flag
+ private boolean dirty;
+ private long lastSaved = Long.MIN_VALUE;
+
+ public boolean isDirty(final long tick) {
+ return this.dirty || (!this.tickQueue.isEmpty() && tick != this.lastSaved);
+ }
+
+ public void clearDirty() {
+ this.dirty = false;
+ }
+ // Paper end - add dirty flag
+
public LevelChunkTicks() {
}
@@ -50,6 +63,7 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
public ScheduledTick<T> poll() {
ScheduledTick<T> scheduledTick = this.tickQueue.poll();
if (scheduledTick != null) {
+ this.dirty = true; // Paper - add dirty flag
this.ticksPerPosition.remove(scheduledTick);
}
@@ -59,6 +73,7 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
@Override
public void schedule(ScheduledTick<T> orderedTick) {
if (this.ticksPerPosition.add(orderedTick)) {
+ this.dirty = true; // Paper - add dirty flag
this.scheduleUnchecked(orderedTick);
}
@@ -83,7 +98,7 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
while(iterator.hasNext()) {
ScheduledTick<T> scheduledTick = iterator.next();
if (predicate.test(scheduledTick)) {
- iterator.remove();
+ iterator.remove(); this.dirty = true; // Paper - add dirty flag
this.ticksPerPosition.remove(scheduledTick);
}
}
@@ -101,6 +116,7 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
@Override
public ListTag save(long l, Function<T, String> function) {
+ this.lastSaved = l; // Paper - add dirty system to level ticks
ListTag listTag = new ListTag();
if (this.pendingTicks != null) {
for(SavedTick<T> savedTick : this.pendingTicks) {
@@ -117,6 +133,11 @@ public class LevelChunkTicks<T> implements SerializableTickContainer<T>, TickCon
public void unpack(long time) {
if (this.pendingTicks != null) {
+ // Paper start - add dirty system to level chunk ticks
+ if (this.tickQueue.isEmpty()) {
+ this.lastSaved = time;
+ }
+ // Paper end - add dirty system to level chunk ticks
int i = -this.pendingTicks.size();
for(SavedTick<T> savedTick : this.pendingTicks) {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
diff --git a/src/main/java/org/bukkit/craftbukkit/CraftChunk.java b/src/main/java/org/bukkit/craftbukkit/CraftChunk.java
index 60907cc435004398c46381b7b32bf47fe8938736..317bf725767e7dc8b916eeb8cd5bcd30ac219aaa 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/CraftChunk.java
+++ b/src/main/java/org/bukkit/craftbukkit/CraftChunk.java
@@ -111,7 +111,7 @@ public class CraftChunk implements Chunk {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public boolean isEntitiesLoaded() {
- return this.getCraftWorld().getHandle().entityManager.areEntitiesLoaded(ChunkPos.asLong(x, z));
+ return this.getCraftWorld().getHandle().areEntitiesLoaded(io.papermc.paper.util.CoordinateUtils.getChunkKey(this.x, this.z)); // Paper - rewrite chunk system
}
@Override
@@ -120,51 +120,7 @@ public class CraftChunk implements Chunk {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.getWorld().getChunkAt(x, z); // Transient load for this tick
}
- PersistentEntitySectionManager<net.minecraft.world.entity.Entity> entityManager = this.getCraftWorld().getHandle().entityManager;
- long pair = ChunkPos.asLong(x, z);
-
- if (entityManager.areEntitiesLoaded(pair)) {
- return entityManager.getEntities(new ChunkPos(this.x, this.z)).stream()
- .map(net.minecraft.world.entity.Entity::getBukkitEntity)
- .filter(Objects::nonNull).toArray(Entity[]::new);
- }
-
- entityManager.ensureChunkQueuedForLoad(pair); // Start entity loading
-
- // SPIGOT-6772: Use entity mailbox and re-schedule entities if they get unloaded
- ProcessorMailbox<Runnable> mailbox = ((EntityStorage) entityManager.permanentStorage).entityDeserializerQueue;
- BooleanSupplier supplier = () -> {
- // only execute inbox if our entities are not present
- if (entityManager.areEntitiesLoaded(pair)) {
- return true;
- }
-
- if (!entityManager.isPending(pair)) {
- // Our entities got unloaded, this should normally not happen.
- entityManager.ensureChunkQueuedForLoad(pair); // Re-start entity loading
- }
-
- // tick loading inbox, which loads the created entities to the world
- // (if present)
- entityManager.tick();
- // check if our entities are loaded
- return entityManager.areEntitiesLoaded(pair);
- };
-
- // now we wait until the entities are loaded,
- // the converting from NBT to entity object is done on the main Thread which is why we wait
- while (!supplier.getAsBoolean()) {
- if (mailbox.size() != 0) {
- mailbox.run();
- } else {
- Thread.yield();
- LockSupport.parkNanos("waiting for entity loading", 100000L);
- }
- }
-
- return entityManager.getEntities(new ChunkPos(this.x, this.z)).stream()
- .map(net.minecraft.world.entity.Entity::getBukkitEntity)
- .filter(Objects::nonNull).toArray(Entity[]::new);
+ return getCraftWorld().getHandle().getChunkEntities(this.x, this.z); // Paper - rewrite chunk system
}
@Override
diff --git a/src/main/java/org/bukkit/craftbukkit/CraftServer.java b/src/main/java/org/bukkit/craftbukkit/CraftServer.java
index 394d8631cb0a143cc7b5d71c919a8298652d5678..cd667805a4ace3edc88a8ff11369fd169e5614e3 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/CraftServer.java
+++ b/src/main/java/org/bukkit/craftbukkit/CraftServer.java
2023-03-14 19:05:23 +01:00
@@ -1124,7 +1124,7 @@ public final class CraftServer implements Server {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.console.addLevel(internal);
this.getServer().prepareLevels(internal.getChunkSource().chunkMap.progressListener, internal);
- internal.entityManager.tick(); // SPIGOT-6526: Load pending entities so they are available to the API
+ //internal.entityManager.tick(); // SPIGOT-6526: Load pending entities so they are available to the API // Paper - rewrite chunk system
this.pluginManager.callEvent(new WorldLoadEvent(internal.getWorld()));
return internal.getWorld();
2023-03-14 19:05:23 +01:00
@@ -1168,7 +1168,7 @@ public final class CraftServer implements Server {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
}
handle.getChunkSource().close(save);
- handle.entityManager.close(save); // SPIGOT-6722: close entityManager
+ // handle.entityManager.close(save); // SPIGOT-6722: close entityManager // Paper - rewrite chunk system
handle.convertable.close();
} catch (Exception ex) {
this.getLogger().log(Level.SEVERE, null, ex);
2023-03-14 19:05:23 +01:00
@@ -1987,7 +1987,7 @@ public final class CraftServer implements Server {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public boolean isPrimaryThread() {
- return Thread.currentThread().equals(console.serverThread) || this.console.hasStopped() || !org.spigotmc.AsyncCatcher.enabled; // All bets are off if we have shut down (e.g. due to watchdog)
+ return io.papermc.paper.util.TickThread.isTickThread(); // Paper - rewrite chunk system
}
// Paper start
diff --git a/src/main/java/org/bukkit/craftbukkit/CraftWorld.java b/src/main/java/org/bukkit/craftbukkit/CraftWorld.java
index 4186943ba9e94758d579477e8fadaf8e05ef93f3..1684b3ee1b51597e799486bbe129b344a9fe3098 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/CraftWorld.java
+++ b/src/main/java/org/bukkit/craftbukkit/CraftWorld.java
@@ -325,10 +325,14 @@ public class CraftWorld extends CraftRegionAccessor implements World {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
ChunkHolder playerChunk = this.world.getChunkSource().chunkMap.getVisibleChunkIfPresent(ChunkPos.asLong(x, z));
if (playerChunk == null) return false;
- playerChunk.getTickingChunkFuture().thenAccept(either -> {
- either.left().ifPresent(chunk -> {
+ // Paper start - rewrite player chunk loader
+ net.minecraft.world.level.chunk.LevelChunk chunk = playerChunk.getSendingChunk();
+ if (chunk == null) {
+ return false;
+ }
+ // Paper end - rewrite player chunk loader
List<ServerPlayer> playersInRange = playerChunk.playerProvider.getPlayers(playerChunk.getPos(), false);
- if (playersInRange.isEmpty()) return;
+ if (playersInRange.isEmpty()) return true; // Paper - rewrite player chunk loader
ClientboundLevelChunkWithLightPacket refreshPacket = new ClientboundLevelChunkWithLightPacket(chunk, this.world.getLightEngine(), null, null, true);
for (ServerPlayer player : playersInRange) {
@@ -336,8 +340,7 @@ public class CraftWorld extends CraftRegionAccessor implements World {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
player.connection.send(refreshPacket);
}
- });
- });
+ // Paper - rewrite player chunk loader
return true;
}
@@ -414,20 +417,7 @@ public class CraftWorld extends CraftRegionAccessor implements World {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
@Override
public Collection<Plugin> getPluginChunkTickets(int x, int z) {
DistanceManager chunkDistanceManager = this.world.getChunkSource().chunkMap.distanceManager;
- SortedArraySet<Ticket<?>> tickets = chunkDistanceManager.tickets.get(ChunkPos.asLong(x, z));
-
- if (tickets == null) {
- return Collections.emptyList();
- }
-
- ImmutableList.Builder<Plugin> ret = ImmutableList.builder();
- for (Ticket<?> ticket : tickets) {
- if (ticket.getType() == TicketType.PLUGIN_TICKET) {
- ret.add((Plugin) ticket.key);
- }
- }
-
- return ret.build();
+ return chunkDistanceManager.getChunkHolderManager().getPluginChunkTickets(x, z); // Paper - rewrite chunk system
}
@Override
@@ -435,7 +425,7 @@ public class CraftWorld extends CraftRegionAccessor implements World {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
Map<Plugin, ImmutableList.Builder<Chunk>> ret = new HashMap<>();
DistanceManager chunkDistanceManager = this.world.getChunkSource().chunkMap.distanceManager;
- for (Long2ObjectMap.Entry<SortedArraySet<Ticket<?>>> chunkTickets : chunkDistanceManager.tickets.long2ObjectEntrySet()) {
+ for (Long2ObjectMap.Entry<SortedArraySet<Ticket<?>>> chunkTickets : chunkDistanceManager.getChunkHolderManager().getTicketsCopy().long2ObjectEntrySet()) { // Paper - rewrite chunk system
long chunkKey = chunkTickets.getLongKey();
SortedArraySet<Ticket<?>> tickets = chunkTickets.getValue();
@@ -1931,14 +1921,53 @@ public class CraftWorld extends CraftRegionAccessor implements World {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
// Spigot start
@Override
public int getViewDistance() {
- return world.spigotConfig.viewDistance;
+ return getHandle().getChunkSource().chunkMap.playerChunkManager.getTargetNoTickViewDistance(); // Paper - replace old player chunk management
}
@Override
public int getSimulationDistance() {
- return world.spigotConfig.simulationDistance;
+ return getHandle().getChunkSource().chunkMap.playerChunkManager.getTargetTickViewDistance(); // Paper - replace old player chunk management
}
// Spigot end
+ // Paper start - view distance api
+ @Override
+ public void setViewDistance(int viewDistance) {
+ if (viewDistance < 2 || viewDistance > 32) {
+ throw new IllegalArgumentException("View distance " + viewDistance + " is out of range of [2, 32]");
+ }
+ net.minecraft.server.level.ChunkMap chunkMap = getHandle().getChunkSource().chunkMap;
+ chunkMap.setViewDistance(viewDistance);
+ }
+
+ @Override
+ public void setSimulationDistance(int simulationDistance) {
+ if (simulationDistance < 2 || simulationDistance > 32) {
+ throw new IllegalArgumentException("Simulation distance " + simulationDistance + " is out of range of [2, 32]");
+ }
+ net.minecraft.server.level.ChunkMap chunkMap = getHandle().getChunkSource().chunkMap;
+ chunkMap.setTickViewDistance(simulationDistance);
+ }
+
+ @Override
+ public int getNoTickViewDistance() {
+ return this.getViewDistance();
+ }
+
+ @Override
+ public void setNoTickViewDistance(int viewDistance) {
+ this.setViewDistance(viewDistance);
+ }
+
+ @Override
+ public int getSendViewDistance() {
+ return getHandle().getChunkSource().chunkMap.playerChunkManager.getTargetSendDistance();
+ }
+
+ @Override
+ public void setSendViewDistance(int viewDistance) {
+ getHandle().getChunkSource().chunkMap.playerChunkManager.setSendDistance(viewDistance);
+ }
+ // Paper end - view distance api
// Spigot start
private final org.bukkit.World.Spigot spigot = new org.bukkit.World.Spigot()
diff --git a/src/main/java/org/bukkit/craftbukkit/entity/CraftPlayer.java b/src/main/java/org/bukkit/craftbukkit/entity/CraftPlayer.java
index 97a8c43ff3f32d929de117ff608a69221b1ab571..6d14414d7ee86dcb176ad46c564d6823b0bd16a3 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/entity/CraftPlayer.java
+++ b/src/main/java/org/bukkit/craftbukkit/entity/CraftPlayer.java
@@ -179,6 +179,81 @@ public class CraftPlayer extends CraftHumanEntity implements Player {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
this.firstPlayed = System.currentTimeMillis();
}
+ // Paper start - implement view distances
+ @Override
+ public int getViewDistance() {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ return chunkMap.playerChunkManager.getTargetNoTickViewDistance();
+ }
+ return data.getTargetNoTickViewDistance();
+ }
+
+ @Override
+ public void setViewDistance(int viewDistance) {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ throw new IllegalStateException("Player is not attached to world");
+ }
+
+ data.setTargetNoTickViewDistance(viewDistance);
+ }
+
+ @Override
+ public int getSimulationDistance() {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ return chunkMap.playerChunkManager.getTargetTickViewDistance();
+ }
+ return data.getTargetTickViewDistance();
+ }
+
+ @Override
+ public void setSimulationDistance(int simulationDistance) {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ throw new IllegalStateException("Player is not attached to world");
+ }
+
+ data.setTargetTickViewDistance(simulationDistance);
+ }
+
+ @Override
+ public int getNoTickViewDistance() {
+ return this.getViewDistance();
+ }
+
+ @Override
+ public void setNoTickViewDistance(int viewDistance) {
+ this.setViewDistance(viewDistance);
+ }
+
+ @Override
+ public int getSendViewDistance() {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ return chunkMap.playerChunkManager.getTargetSendDistance();
+ }
+ return data.getTargetSendViewDistance();
+ }
+
+ @Override
+ public void setSendViewDistance(int viewDistance) {
+ net.minecraft.server.level.ChunkMap chunkMap = this.getHandle().getLevel().getChunkSource().chunkMap;
+ io.papermc.paper.chunk.PlayerChunkLoader.PlayerLoaderData data = chunkMap.playerChunkManager.getData(this.getHandle());
+ if (data == null) {
+ throw new IllegalStateException("Player is not attached to world");
+ }
+
+ data.setTargetSendViewDistance(viewDistance);
+ }
+ // Paper end - implement view distances
+
public GameProfile getProfile() {
return this.getHandle().getGameProfile();
}
diff --git a/src/main/java/org/bukkit/craftbukkit/generator/CustomChunkGenerator.java b/src/main/java/org/bukkit/craftbukkit/generator/CustomChunkGenerator.java
2022-12-07 19:32:25 +01:00
index 61767be202f95df16155e7a4c36701b21619f317..a9673a804d597599c35c83f4f245510c83005328 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/generator/CustomChunkGenerator.java
+++ b/src/main/java/org/bukkit/craftbukkit/generator/CustomChunkGenerator.java
2022-12-07 19:32:25 +01:00
@@ -270,7 +270,7 @@ public class CustomChunkGenerator extends InternalChunkGenerator {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
return ichunkaccess1;
};
- return future == null ? CompletableFuture.supplyAsync(() -> function.apply(chunk), net.minecraft.Util.backgroundExecutor()) : future.thenApply(function);
+ return future == null ? CompletableFuture.supplyAsync(() -> function.apply(chunk), executor) : future.thenApply(function); // Paper - run with supplied executor
}
@Override
diff --git a/src/main/java/org/bukkit/craftbukkit/util/DummyGeneratorAccess.java b/src/main/java/org/bukkit/craftbukkit/util/DummyGeneratorAccess.java
index 629178347039893fb9de710810fe8112499bb91c..d5861dfa771c4eb217e082e3c832c3a6c603710d 100644
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
--- a/src/main/java/org/bukkit/craftbukkit/util/DummyGeneratorAccess.java
+++ b/src/main/java/org/bukkit/craftbukkit/util/DummyGeneratorAccess.java
@@ -258,4 +258,20 @@ public class DummyGeneratorAccess implements WorldGenLevel {
Rewrite chunk system (#8177) Patch documentation to come Issues with the old system that are fixed now: - World generation does not scale with cpu cores effectively. - Relies on the main thread for scheduling and maintaining chunk state, dropping chunk load/generate rates at lower tps. - Unreliable prioritisation of chunk gen/load calls that block the main thread. - Shutdown logic is utterly unreliable, as it has to wait for all chunks to unload - is it guaranteed that the chunk system is in a state on shutdown that it can reliably do this? Watchdog shutdown also typically failed due to thread checks, which is now resolved. - Saving of data is not unified (i.e can save chunk data without saving entity data, poses problems for desync if shutdown is really abnormal. - Entities are not loaded with chunks. This caused quite a bit of headache for Chunk#getEntities API, but now the new chunk system loads entities with chunks so that they are ready whenever the chunk loads in. Effectively brings the behavior back to 1.16 era, but still storing entities in their own separate regionfiles. The above list is not complete. The patch documentation will complete it. New chunk system hard relies on starlight and dataconverter, and most importantly the new concurrent utilities in ConcurrentUtil. Some of the old async chunk i/o interface (i.e the old file io thread reroutes _some_ calls to the new file io thread) is kept for plugin compat reasons. It will be removed in the next major version of minecraft. The old legacy chunk system patches have been moved to the removed folder in case we need them again.
2022-09-26 10:02:51 +02:00
public boolean destroyBlock(BlockPos pos, boolean drop, Entity breakingEntity, int maxUpdateDepth) {
return false; // SPIGOT-6515
}
+
+ // Paper start
+ @Override
+ public List<Entity> getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate) {
+ return java.util.Collections.emptyList();
+ }
+
+ @Override
+ public void getEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {}
+
+ @Override
+ public void getHardCollidingEntities(Entity except, AABB box, Predicate<? super Entity> predicate, List<Entity> into) {}
+
+ @Override
+ public <T> void getEntitiesByClass(Class<? extends T> clazz, Entity except, AABB box, List<? super T> into, Predicate<? super T> predicate) {}
+ // Paper end
}
diff --git a/src/main/java/org/spigotmc/ActivationRange.java b/src/main/java/org/spigotmc/ActivationRange.java
index 38cf408899cef72bc9d2888109a7ac7ce0aec638..d3639643cda7d8ccf3c1208502605120590a2d30 100644
--- a/src/main/java/org/spigotmc/ActivationRange.java
+++ b/src/main/java/org/spigotmc/ActivationRange.java
@@ -132,7 +132,13 @@ public class ActivationRange
ActivationType.ANIMAL.boundingBox = player.getBoundingBox().inflate( animalActivationRange, 256, animalActivationRange );
ActivationType.MONSTER.boundingBox = player.getBoundingBox().inflate( monsterActivationRange, 256, monsterActivationRange );
- world.getEntities().get(maxBB, ActivationRange::activateEntity);
+ // Paper start
+ java.util.List<Entity> entities = world.getEntities((Entity)null, maxBB, null);
+ for (int i = 0; i < entities.size(); i++) {
+ Entity entity = entities.get(i);
+ ActivationRange.activateEntity(entity);
+ }
+ // Paper end
}
MinecraftTimings.entityActivationCheckTimer.stopTiming();
}
diff --git a/src/main/java/org/spigotmc/AsyncCatcher.java b/src/main/java/org/spigotmc/AsyncCatcher.java
index 78669fa035b7537ff7e533cf32aaf2995625424f..05e94702e42b8f5c35d2a112c486d57948a3acba 100644
--- a/src/main/java/org/spigotmc/AsyncCatcher.java
+++ b/src/main/java/org/spigotmc/AsyncCatcher.java
@@ -9,7 +9,7 @@ public class AsyncCatcher
public static void catchOp(String reason)
{
- if ( (AsyncCatcher.enabled || io.papermc.paper.util.TickThread.STRICT_THREAD_CHECKS) && Thread.currentThread() != MinecraftServer.getServer().serverThread ) // Paper
+ if ( !io.papermc.paper.util.TickThread.isTickThread() ) // Paper // Paper - rewrite chunk system
{
throw new IllegalStateException( "Asynchronous " + reason + "!" );
}
diff --git a/src/main/java/org/spigotmc/WatchdogThread.java b/src/main/java/org/spigotmc/WatchdogThread.java
index 335120afc88a8fc1543c2e6df516fd728e3ab032..f1194eb6fdfba60959e00080d0562f2820d13b27 100644
--- a/src/main/java/org/spigotmc/WatchdogThread.java
+++ b/src/main/java/org/spigotmc/WatchdogThread.java
@@ -8,7 +8,7 @@ import java.util.logging.Logger;
import net.minecraft.server.MinecraftServer;
import org.bukkit.Bukkit;
-public class WatchdogThread extends Thread
+public final class WatchdogThread extends io.papermc.paper.util.TickThread // Paper - rewrite chunk system
{
private static WatchdogThread instance;
@@ -83,6 +83,7 @@ public class WatchdogThread extends Thread
//
log.log( Level.SEVERE, "------------------------------" );
log.log( Level.SEVERE, "Server thread dump (Look for plugins here before reporting to Spigot!):" );
+ io.papermc.paper.chunk.system.scheduling.ChunkTaskScheduler.dumpAllChunkLoadInfo(isLongTimeout); // Paper // Paper - rewrite chunk system
WatchdogThread.dumpThread( ManagementFactory.getThreadMXBean().getThreadInfo( MinecraftServer.getServer().serverThread.getId(), Integer.MAX_VALUE ), log );
log.log( Level.SEVERE, "------------------------------" );
//