Paper/patches/server/0488-Improve-Chunk-Status-Transition-Speed.patch

82 lines
4.4 KiB
Diff
Raw Normal View History

2021-06-11 14:02:28 +02:00
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Aikar <aikar@aikar.co>
Date: Fri, 29 May 2020 23:32:14 -0400
Subject: [PATCH] Improve Chunk Status Transition Speed
When a chunk is loaded from disk that has already been generated,
the server has to promote the chunk through the system to reach
it's current desired status level.
This results in every single status transition going from the main thread
to the world gen threads, only to discover it has no work it actually
needs to do.... and then it returns back to main.
This back and forth costs a lot of time and can really delay chunk loads
when the server is under high TPS due to their being a lot of time in
between chunk load times, as well as hogs up the chunk threads from doing
actual generation and light work.
Additionally, the whole task system uses a lot of CPU on the server threads anyways.
So by optimizing status transitions for status's that are already complete,
we can run them to the desired level while on main thread (where it has
to happen anyways) instead of ever jumping to world gen thread.
This will improve chunk loading effeciency to be reduced down to the following
scenario / path:
1) MAIN: Chunk Requested, Load Request sent to ChunkTaskManager / IO Queue
2) IO: Once position in queue comes, submit read IO data and schedule to chunk task thread
3) CHUNK: Once IO is loaded and position in queue comes, deserialize the chunk data, process conversions, submit to main queue
4) MAIN: next Chunk Task process (Mid Tick or End Of Tick), load chunk data into world (POI, main thread tasks)
5) MAIN: process status transitions all the way to LIGHT, light schedules Threaded task
6) SERVER: Light tasks register light enablement for chunk and any lighting needing to be done
7) MAIN: Task returns to main, finish processing to FULL/TICKING status
Previously would have hopped to SERVER around 12+ times there extra.
diff --git a/src/main/java/net/minecraft/server/level/ChunkHolder.java b/src/main/java/net/minecraft/server/level/ChunkHolder.java
index e317c73b430105052cbf28502916adc3bd60846b..96759f349e10c8ef634a11a6b48f9affae77d65a 100644
2021-06-11 14:02:28 +02:00
--- a/src/main/java/net/minecraft/server/level/ChunkHolder.java
+++ b/src/main/java/net/minecraft/server/level/ChunkHolder.java
@@ -244,6 +244,13 @@ public class ChunkHolder {
2021-06-11 14:02:28 +02:00
// Paper end - optimise isOutsideOfRange
2021-06-17 10:37:27 +02:00
long lastAutoSaveTime; // Paper - incremental autosave
long inactiveTimeStart; // Paper - incremental autosave
2021-06-11 14:02:28 +02:00
+ // Paper start - optimize chunk status progression without jumping through thread pool
+ public boolean canAdvanceStatus() {
+ ChunkStatus status = getChunkHolderStatus();
+ ChunkAccess chunk = getAvailableChunkNow();
+ return chunk != null && (status == null || chunk.getStatus().isOrAfter(getNextStatus(status)));
2021-06-11 14:02:28 +02:00
+ }
+ // 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());
2021-06-11 14:02:28 +02:00
diff --git a/src/main/java/net/minecraft/server/level/ChunkMap.java b/src/main/java/net/minecraft/server/level/ChunkMap.java
index ac848504fd7105506d6d733709ad05dc8b37c9d9..2a85deeab49f32a897b53707d453d1a5f2972902 100644
2021-06-11 14:02:28 +02:00
--- a/src/main/java/net/minecraft/server/level/ChunkMap.java
+++ b/src/main/java/net/minecraft/server/level/ChunkMap.java
Merge tuinity (#6413) This PR contains all of Tuinity's patches. Very notable ones are: - Highly optimised collisions - Optimised entity lookups by bounding box (Mojang made regressions in 1.17, this brings it back to 1.16) - Starlight https://github.com/PaperMC/Starlight - Rewritten dataconverter system https://github.com/PaperMC/DataConverter - Random block ticking optimisation (wrongly dropped from Paper 1.17) - Chunk ticking optimisations - Anything else I've forgotten in the 60 or so patches If you are a previous Tuinity user, your config will not migrate. You must do it yourself. The config options have simply been moved into paper.yml, so it will be an easy migration. However, please note that the chunk loading options in tuinity.yml are NOT compatible with the options in paper.yml. * Port tuinity, initial patchset * Update gradle to 7.2 jmp said it fixes rebuildpatches not working for me. it fucking better * Completely clean apply * Remove tuinity config, add per player api patch * Remove paper reobf mappings patch * Properly update gradlew * Force clean rebuild * Mark fixups Comments and ATs still need to be done * grep -r "Tuinity" * Fixup * Ensure gameprofile lastaccess is written only under the state lock * update URL for dataconverter * Only clean rebuild tuinity patches might fix merge conflicts * Use UTF-8 for gradlew * Clean rb patches again * Convert block ids used as item ids Neither the converters of pre 1.13 nor DFU handled these cases, as by the time they were written the game at the time didn't consider these ids valid - they would be air. Because of this, some worlds have logspam since only DataConverter (not DFU or legacy converters) will warn when an invalid id has been seen. While quite a few do need to now be considered as air, quite a lot do not. So it makes sense to add conversion for these items, instead of simply suppressing or ignoring the logs. I've now added id -> string conversion for all block ids that could be used as items that existed in the game before 1.7.10 (I have no interest in tracking down the exact version block ids stopped working) that were on https://minecraft-ids.grahamedgecombe.com/ Items that did not directly convert to new items will be instead converted to air: stems, wheat crops, piston head, tripwire wire block * Fix LightPopulated parsing in V1466 The DFU code was checking if the number existed, not if it didn't exist. I misread the original code. * Always parse protochunk light sources unless it is marked as non-lit Chunks not marked as lit will always go through the light engine, so they should always have their block sources parsed. * Update custom names to JSON for players Missed this fix from CB, as it was inside the DataFixers class. I decided to double check all of the CB changes again: DataFixers.java was the only area I missed, as I had inspected all datafixer diffs and implemented them all into DataConverter. I also checked Bootstrap.java again, and re-evaluated their changes. I had previously done this, but determined that they were all bad. The changes to make standing_sign block map to oak_sign block in V1450 is bad, because that's not the item id V1450 accepts. Only in 1.14 did oak_sign even exist, and as expected there is a converter to rename all existing sign items/blocks. The fix to register the portal block under id 1440 is useless, as the flattenning logic will default to the lowest registered id - which is the exact blockstate that CB registers into 1440. So it just doesn't do anything. The extra item ids in the id -> string converter are already added, but I found this from EMC originally. The change for the spawn egg id 23 -> Arrow is just wrong, that id DOES correspond to TippedArrow, NOT Arrow. As expected, the spawn egg already has a dedicated mapping for Arrow, which is id 10 - which was Arrow's entity id. I also ported a fix for the cooked_fished id update. This doesn't really matter since there is already a dataconverter to fix this, but the game didn't accept cooked_fished at the time. So I see no harm. * Review all converters and walkers - Refactor V99 to have helper methods for defining entity/tile entity types - Automatically namespace all ids that should be namespaced. While vanilla never saved non-namespaced data for things that are namespaced, plugins/users might have. - Synchronised the identity ensure map in HelperBlockFlatteningV1450 - Code style consistency - Add missing log warning in V102 for ITEM_NAME type conversion - Use getBoolean instead of getByte - Use ConverterAbstractEntityRename for V143 TippedArrow -> Arrow rename, as it will affect ENTITY_NAME type - Always set isVillager to false in V502 for Zombie - Register V808's converter under subversion 1 like DFU - Register a breakpoint for V1.17.1. In the future, all final versions of major releases will have a breakpoint so that the work required to determine if a converter needs a breakpoint is minimal - Validate that a dataconverter is only registered for a version that is registered - ConverterFlattenTileEntity is actually ConverterFlattenEntity It even registered the converters under TILE_ENTITY, instead of ENTITY. - Fix id comparison in V1492 STRUCTURE_FEATURE renamer - Use ConverterAbstractStatsRename for V1510 stats renamer At the time I had written that class, the abstract renamer didn't exist. - Ensure OwnerUUID is at least set to empty string in V1904 if the ocelot is converted to a cat (this is likely so that it retains a collar) - Use generic read/write for Records in V1946 Records is actually a list, not a map. So reading map was invalid. * Always set light to zero when propagating decrease This fixes an almost infinite loop where light values would be spam queued on a very small subset on blocks. This also likely fixes the memory issues people were seeing. * re-organize patches * Apply and fix conflicts * Revert some patches getChunkAt retains chunks so that plugins don't spam loads revert mc-4 fix will remain unless issues pop up * Shuffle iterated chunks if per player is not enabled Can help with some mob spawning stacking up at locations * Make per player default, migrate all configs * Adjust comments in fixups * Rework config for player chunk loader Old config is not compatible. Move all configs to be under `settings` in paper.yml The player chunk loader has been modified to less aggressively load chunks, but to send chunks at higher rates compared to tuinity. There are new config entries to tune this behavior. * Add back old constructor to CompressionEncoder/Decoder (fixes Tuinity #358) * Raise chunk loading default limits * Reduce worldgen thread workers for lower core count cpus * Raise limits for chunk loading config Also place it under `chunk-loading` * Disable max chunk send rate by default * Fix conflicts and rebuild patches * Drop default send rate again Appears to be still causing problems for no known reason * Raise chunk send limits to 100 per player While a low limit fixes ping issues for some people, most people do not suffer from this issue and thus should not suffer from an extremely slow load-in rate. * Rebase part 1 Autosquash the fixups * Move not implemented up * Fixup mc-dev fixes Missed this one * Rebase per player viewdistance api into the original api patch * Remove old light engine patch part 1 The prioritisation must be kept from it, so that part has been rebased into the priority patch. Part 2 will deal with rebasing all of the patches _after_ * Rebase remaining patches for old light patch removal * Remove other mid tick patch * Remove Optimize-PlayerChunkMap-memory-use-for-visibleChunks.patch Replaced by `Do not copy visible chunks` * Revert AT for Vec3i setX/Y/Z The class is immutable. set should not be exposed * Remove old IntegerUtil class * Replace old CraftChunk#getEntities patch * Remove import for SWMRNibbleArray in ChunkAccess * Finished merge checklist * Remove ensureTickThread impl in urgency patch Co-authored-by: Spottedleaf <Spottedleaf@users.noreply.github.com> Co-authored-by: Jason Penilla <11360596+jpenilla@users.noreply.github.com>
2021-08-31 13:02:11 +02:00
@@ -786,7 +786,7 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
2021-06-11 14:02:28 +02:00
return either.mapLeft((list) -> {
return (LevelChunk) list.get(list.size() / 2);
});
- }, this.mainThreadExecutor);
+ }, this.mainInvokingExecutor); // Paper
}
@Nullable
Merge tuinity (#6413) This PR contains all of Tuinity's patches. Very notable ones are: - Highly optimised collisions - Optimised entity lookups by bounding box (Mojang made regressions in 1.17, this brings it back to 1.16) - Starlight https://github.com/PaperMC/Starlight - Rewritten dataconverter system https://github.com/PaperMC/DataConverter - Random block ticking optimisation (wrongly dropped from Paper 1.17) - Chunk ticking optimisations - Anything else I've forgotten in the 60 or so patches If you are a previous Tuinity user, your config will not migrate. You must do it yourself. The config options have simply been moved into paper.yml, so it will be an easy migration. However, please note that the chunk loading options in tuinity.yml are NOT compatible with the options in paper.yml. * Port tuinity, initial patchset * Update gradle to 7.2 jmp said it fixes rebuildpatches not working for me. it fucking better * Completely clean apply * Remove tuinity config, add per player api patch * Remove paper reobf mappings patch * Properly update gradlew * Force clean rebuild * Mark fixups Comments and ATs still need to be done * grep -r "Tuinity" * Fixup * Ensure gameprofile lastaccess is written only under the state lock * update URL for dataconverter * Only clean rebuild tuinity patches might fix merge conflicts * Use UTF-8 for gradlew * Clean rb patches again * Convert block ids used as item ids Neither the converters of pre 1.13 nor DFU handled these cases, as by the time they were written the game at the time didn't consider these ids valid - they would be air. Because of this, some worlds have logspam since only DataConverter (not DFU or legacy converters) will warn when an invalid id has been seen. While quite a few do need to now be considered as air, quite a lot do not. So it makes sense to add conversion for these items, instead of simply suppressing or ignoring the logs. I've now added id -> string conversion for all block ids that could be used as items that existed in the game before 1.7.10 (I have no interest in tracking down the exact version block ids stopped working) that were on https://minecraft-ids.grahamedgecombe.com/ Items that did not directly convert to new items will be instead converted to air: stems, wheat crops, piston head, tripwire wire block * Fix LightPopulated parsing in V1466 The DFU code was checking if the number existed, not if it didn't exist. I misread the original code. * Always parse protochunk light sources unless it is marked as non-lit Chunks not marked as lit will always go through the light engine, so they should always have their block sources parsed. * Update custom names to JSON for players Missed this fix from CB, as it was inside the DataFixers class. I decided to double check all of the CB changes again: DataFixers.java was the only area I missed, as I had inspected all datafixer diffs and implemented them all into DataConverter. I also checked Bootstrap.java again, and re-evaluated their changes. I had previously done this, but determined that they were all bad. The changes to make standing_sign block map to oak_sign block in V1450 is bad, because that's not the item id V1450 accepts. Only in 1.14 did oak_sign even exist, and as expected there is a converter to rename all existing sign items/blocks. The fix to register the portal block under id 1440 is useless, as the flattenning logic will default to the lowest registered id - which is the exact blockstate that CB registers into 1440. So it just doesn't do anything. The extra item ids in the id -> string converter are already added, but I found this from EMC originally. The change for the spawn egg id 23 -> Arrow is just wrong, that id DOES correspond to TippedArrow, NOT Arrow. As expected, the spawn egg already has a dedicated mapping for Arrow, which is id 10 - which was Arrow's entity id. I also ported a fix for the cooked_fished id update. This doesn't really matter since there is already a dataconverter to fix this, but the game didn't accept cooked_fished at the time. So I see no harm. * Review all converters and walkers - Refactor V99 to have helper methods for defining entity/tile entity types - Automatically namespace all ids that should be namespaced. While vanilla never saved non-namespaced data for things that are namespaced, plugins/users might have. - Synchronised the identity ensure map in HelperBlockFlatteningV1450 - Code style consistency - Add missing log warning in V102 for ITEM_NAME type conversion - Use getBoolean instead of getByte - Use ConverterAbstractEntityRename for V143 TippedArrow -> Arrow rename, as it will affect ENTITY_NAME type - Always set isVillager to false in V502 for Zombie - Register V808's converter under subversion 1 like DFU - Register a breakpoint for V1.17.1. In the future, all final versions of major releases will have a breakpoint so that the work required to determine if a converter needs a breakpoint is minimal - Validate that a dataconverter is only registered for a version that is registered - ConverterFlattenTileEntity is actually ConverterFlattenEntity It even registered the converters under TILE_ENTITY, instead of ENTITY. - Fix id comparison in V1492 STRUCTURE_FEATURE renamer - Use ConverterAbstractStatsRename for V1510 stats renamer At the time I had written that class, the abstract renamer didn't exist. - Ensure OwnerUUID is at least set to empty string in V1904 if the ocelot is converted to a cat (this is likely so that it retains a collar) - Use generic read/write for Records in V1946 Records is actually a list, not a map. So reading map was invalid. * Always set light to zero when propagating decrease This fixes an almost infinite loop where light values would be spam queued on a very small subset on blocks. This also likely fixes the memory issues people were seeing. * re-organize patches * Apply and fix conflicts * Revert some patches getChunkAt retains chunks so that plugins don't spam loads revert mc-4 fix will remain unless issues pop up * Shuffle iterated chunks if per player is not enabled Can help with some mob spawning stacking up at locations * Make per player default, migrate all configs * Adjust comments in fixups * Rework config for player chunk loader Old config is not compatible. Move all configs to be under `settings` in paper.yml The player chunk loader has been modified to less aggressively load chunks, but to send chunks at higher rates compared to tuinity. There are new config entries to tune this behavior. * Add back old constructor to CompressionEncoder/Decoder (fixes Tuinity #358) * Raise chunk loading default limits * Reduce worldgen thread workers for lower core count cpus * Raise limits for chunk loading config Also place it under `chunk-loading` * Disable max chunk send rate by default * Fix conflicts and rebuild patches * Drop default send rate again Appears to be still causing problems for no known reason * Raise chunk send limits to 100 per player While a low limit fixes ping issues for some people, most people do not suffer from this issue and thus should not suffer from an extremely slow load-in rate. * Rebase part 1 Autosquash the fixups * Move not implemented up * Fixup mc-dev fixes Missed this one * Rebase per player viewdistance api into the original api patch * Remove old light engine patch part 1 The prioritisation must be kept from it, so that part has been rebased into the priority patch. Part 2 will deal with rebasing all of the patches _after_ * Rebase remaining patches for old light patch removal * Remove other mid tick patch * Remove Optimize-PlayerChunkMap-memory-use-for-visibleChunks.patch Replaced by `Do not copy visible chunks` * Revert AT for Vec3i setX/Y/Z The class is immutable. set should not be exposed * Remove old IntegerUtil class * Replace old CraftChunk#getEntities patch * Remove import for SWMRNibbleArray in ChunkAccess * Finished merge checklist * Remove ensureTickThread impl in urgency patch Co-authored-by: Spottedleaf <Spottedleaf@users.noreply.github.com> Co-authored-by: Jason Penilla <11360596+jpenilla@users.noreply.github.com>
2021-08-31 13:02:11 +02:00
@@ -1237,6 +1237,12 @@ public class ChunkMap extends ChunkStorage implements ChunkHolder.PlayerProvider
return "chunkGenerate " + requiredStatus.getName();
});
Executor executor = (runnable) -> {
2021-06-11 14:02:28 +02:00
+ // Paper start - optimize chunk status progression without jumping through thread pool
+ if (holder.canAdvanceStatus()) {
+ this.mainInvokingExecutor.execute(runnable);
+ return;
+ }
+ // Paper end
this.worldgenMailbox.tell(ChunkTaskPriorityQueueSorter.message(holder, runnable));
};