mirror of
https://github.com/goharbor/harbor.git
synced 2024-11-23 10:45:45 +01:00
f21b1481bb
refactor: refactor the old goroutine execution sweep to global execution sweep job 1. Delete the old goroutine execution sweeper when create execution.(in the case of high concurrency can cause goroutine backlogs, affect the performance of core) 2. Introduce the new way to sweep executions, a global scheduled job will take the work. Signed-off-by: chlins <chenyuzh@vmware.com>
352 lines
16 KiB
Python
352 lines
16 KiB
Python
from __future__ import absolute_import
|
|
import time
|
|
|
|
import unittest
|
|
import v2_swagger_client
|
|
from library import base
|
|
from testutils import harbor_server, ADMIN_CLIENT, suppress_urllib3_warning
|
|
from library.jobservice import Jobservice
|
|
from library.gc import GC
|
|
from library.purge import Purge
|
|
from library.user import User
|
|
from library.project import Project
|
|
from library.retention import Retention
|
|
from library.preheat import Preheat
|
|
from library.replication import Replication
|
|
from library.registry import Registry
|
|
from library.scan_all import ScanAll
|
|
from library.schedule import Schedule
|
|
|
|
|
|
class TestJobServiceDashboard(unittest.TestCase, object):
|
|
|
|
@suppress_urllib3_warning
|
|
def setUp(self):
|
|
self.jobservice = Jobservice()
|
|
self.gc = GC()
|
|
self.purge = Purge()
|
|
self.user = User()
|
|
self.project = Project()
|
|
self.retention = Retention()
|
|
self.preheat = Preheat()
|
|
self.replication = Replication()
|
|
self.registry = Registry()
|
|
self.scan_all = ScanAll()
|
|
self.schedule = Schedule()
|
|
self.job_types = [ "GARBAGE_COLLECTION", "PURGE_AUDIT_LOG", "P2P_PREHEAT", "IMAGE_SCAN", "REPLICATION", "RETENTION", "SCAN_DATA_EXPORT", "SCHEDULER", "SLACK", "SYSTEM_ARTIFACT_CLEANUP", "WEBHOOK", "EXECUTION_SWEEP"]
|
|
self.cron_type = "Custom"
|
|
self.cron = "0 0 0 * * 0"
|
|
|
|
def testJobQueues(self):
|
|
"""
|
|
Test case:
|
|
Job Service Dashboard Job Queues
|
|
Test step and expected result:
|
|
1. List job queue;
|
|
2. Pause GC Job and purge audit Job;
|
|
3. Verify that the Job status is Paused;
|
|
4. Run GC and purge audit;
|
|
5. Verify pending jobs of job queues;
|
|
6. Resume GC Job and purge audit Job;
|
|
7. Verify pending jobs of job queues;
|
|
8. Pause GC Job and purge audit Job;
|
|
9. Verify that the Job status is Paused;
|
|
10. Run GC and purge audit;
|
|
11. Verify pending jobs of job queues;
|
|
12. Stop GC Job and purge audit Job;
|
|
13. Verify pending jobs of job queues;
|
|
14. Run GC and purge audit;
|
|
15. Verify pending jobs of job queues;
|
|
16. Stop all Job;
|
|
17. Verify pending jobs of job queues;
|
|
18. Resume all Job;
|
|
"""
|
|
# 1. List job queue
|
|
job_queues = self.jobservice.get_job_queues()
|
|
self.assertSetEqual(set(self.job_types), set(job_queues.keys()))
|
|
|
|
# 2. Pause GC Job and purge audit Job
|
|
self.jobservice.action_pending_jobs(self.job_types[0], "pause")
|
|
self.jobservice.action_pending_jobs(self.job_types[1], "pause")
|
|
|
|
# 3. Verify that the Job status is Paused
|
|
job_queues = self.jobservice.get_job_queues()
|
|
self.assertTrue(job_queues[self.job_types[0]].paused)
|
|
self.assertTrue(job_queues[self.job_types[1]].paused)
|
|
|
|
# 4. Run GC and purge audit
|
|
self.gc.gc_now()
|
|
self.purge.create_purge_schedule(type="Manual", cron=None, dry_run=False)
|
|
time.sleep(2)
|
|
|
|
# 5. Verify pending jobs of job queues
|
|
self.verifyPendingJobs([self.job_types[0], self.job_types[1]])
|
|
|
|
# 6. Resume GC Job and purge audit Job
|
|
self.jobservice.action_pending_jobs(self.job_types[0], "resume")
|
|
self.jobservice.action_pending_jobs(self.job_types[1], "resume")
|
|
|
|
# 7. Verify pending jobs of job queues
|
|
self.waitJobQueuesStopToComplete([self.job_types[0], self.job_types[1]])
|
|
|
|
# 8. Pause GC Job and purge audit Job;
|
|
self.jobservice.action_pending_jobs(self.job_types[0], "pause")
|
|
self.jobservice.action_pending_jobs(self.job_types[1], "pause")
|
|
|
|
# 9. Verify that the Job status is Paused
|
|
job_queues = self.jobservice.get_job_queues()
|
|
self.assertTrue(job_queues[self.job_types[0]].paused)
|
|
self.assertTrue(job_queues[self.job_types[1]].paused)
|
|
|
|
# 10. Run GC and purge audit
|
|
self.gc.gc_now()
|
|
self.purge.create_purge_schedule(type="Manual", cron=None, dry_run=False)
|
|
time.sleep(2)
|
|
|
|
# 11. Verify pending jobs of job queues
|
|
self.verifyPendingJobs([self.job_types[0], self.job_types[1]])
|
|
|
|
# 12. Stop GC Job and purge audit Job
|
|
self.jobservice.action_pending_jobs(self.job_types[0], "stop")
|
|
self.jobservice.action_pending_jobs(self.job_types[1], "stop")
|
|
|
|
# 13. Verify pending jobs of job queues
|
|
self.waitJobQueuesStopToComplete([self.job_types[0], self.job_types[1]])
|
|
|
|
# 14. Run GC and purge audit
|
|
self.gc.gc_now()
|
|
self.purge.create_purge_schedule(type="Manual", cron=None, dry_run=False)
|
|
time.sleep(2)
|
|
|
|
# 15. Verify pending jobs of job queues
|
|
self.verifyPendingJobs([self.job_types[0], self.job_types[1]])
|
|
|
|
# 16. Stop all Job
|
|
self.jobservice.action_pending_jobs("all", "stop")
|
|
|
|
# 17. Verify pending jobs of job queues
|
|
self.waitJobQueuesStopToComplete([self.job_types[0], self.job_types[1]])
|
|
|
|
# 18. Resume all Job
|
|
self.jobservice.action_pending_jobs("all", "resume")
|
|
job_queues = self.jobservice.get_job_queues()
|
|
self.assertFalse(job_queues[self.job_types[0]].paused)
|
|
self.assertFalse(job_queues[self.job_types[1]].paused)
|
|
|
|
|
|
def verifyPendingJobs(self, job_types):
|
|
job_queues = self.jobservice.get_job_queues()
|
|
for job_type in job_types:
|
|
self.assertTrue(job_queues[job_type].count > 0)
|
|
self.assertTrue(job_queues[job_type].latency > 0)
|
|
self.assertTrue(job_queues[job_type].count > 0)
|
|
self.assertTrue(job_queues[job_type].latency > 0)
|
|
|
|
|
|
def waitJobQueuesStopToComplete(self, job_types):
|
|
is_success = False
|
|
for i in range(10):
|
|
print("Wait for queues to be consumed:", i)
|
|
job_queues = self.jobservice.get_job_queues()
|
|
for job_type in job_types:
|
|
if job_queues[job_type].count is not None or job_queues[job_type].latency is not None:
|
|
is_success = False
|
|
break
|
|
else:
|
|
is_success = True
|
|
break
|
|
time.sleep(2)
|
|
self.assertTrue(is_success)
|
|
|
|
|
|
def testSchedules(self):
|
|
"""
|
|
Test case:
|
|
Job Service Dashboard Schedules
|
|
Test step and expected result:
|
|
1. Create a new project;
|
|
2. Create a retention policy triggered by schedule;
|
|
3. Create a new distribution;
|
|
4. Create a preheat policy triggered by schedule;
|
|
5. Create a new registry;
|
|
6. Create a replication policy triggered by schedule;
|
|
7. Set up a schedule to scan all;
|
|
8. Set up a schedule to GC;
|
|
9. Set up a schedule to log rotation;
|
|
10. Verify schedules;
|
|
11. Pause all schedules;
|
|
12. Verify schedules is Paused;
|
|
13. Resume all schedules;
|
|
14. Verify schedules is not Paused;
|
|
15. Reset the schedule for scan all, GC, and log rotation;
|
|
16. Verify schedules;
|
|
"""
|
|
# 1. Create a new project(PA) by user(UA)
|
|
project_id, project_name = self.project.create_project(metadata = {"public": "false"})
|
|
|
|
# 2. Create a retention policy
|
|
retention_id = self.retention.create_retention_policy(project_id, selector_repository="**", selector_tag="**")
|
|
self.retention.update_retention_policy(retention_id, project_id, cron=self.cron)
|
|
|
|
# 3. Create a new distribution
|
|
_, distribution_name = self.preheat.create_instance(endpoint_url=base._random_name("https://"))
|
|
|
|
# 4. Create a new preheat policy
|
|
distribution = self.preheat.get_instance(distribution_name)
|
|
_, preheat_policy_name = self.preheat.create_policy(project_name, project_id, distribution.id, trigger=r'{"type":"scheduled","trigger_setting":{"cron":"%s"}}' % (self.cron))
|
|
preheat_policy = self.preheat.get_policy(project_name, preheat_policy_name)
|
|
|
|
# 5. Create a new registry
|
|
registry_id, _ = self.registry.create_registry("https://" + harbor_server)
|
|
|
|
# 6. Create a replication policy triggered by schedule
|
|
replication_id, _ = self.replication.create_replication_policy(dest_registry=v2_swagger_client.Registry(id=int(registry_id)), trigger=v2_swagger_client.ReplicationTrigger(type="scheduled",trigger_settings=v2_swagger_client.ReplicationTriggerSettings(cron=self.cron)))
|
|
|
|
# 7. Set up a schedule to scan all
|
|
self.scan_all.create_scan_all_schedule(self.cron_type, cron=self.cron)
|
|
|
|
# 8. Set up a schedule to GC
|
|
self.gc.create_gc_schedule(self.cron_type, is_delete_untagged=True, cron=self.cron)
|
|
|
|
# 9. Set up a schedule to Log Rotation
|
|
self.purge.create_purge_schedule(self.cron_type, self.cron, True)
|
|
|
|
# 10. Verify schedules
|
|
schedules = self.schedule.list_schedules(page_size=50, page=1)
|
|
# 10.1 Verify retention schedule
|
|
retention_schedule = schedules["%s-%d" % (self.job_types[5], retention_id)]
|
|
self.assertEqual(retention_schedule.cron, self.cron)
|
|
# 10.2 Verify preheat schedule
|
|
preheat_schedule = schedules["%s-%d" % (self.job_types[2], preheat_policy.id)]
|
|
self.assertEqual(preheat_schedule.cron, self.cron)
|
|
# 10.3 Verify replication schedule
|
|
replication_schedule = schedules["%s-%d" % (self.job_types[4], replication_id)]
|
|
self.assertEqual(replication_schedule.vendor_type, self.job_types[4])
|
|
self.assertEqual(replication_schedule.cron, self.cron)
|
|
# 10.4 Verify scan all schedule
|
|
scan_all_schedule = schedules["SCAN_ALL"]
|
|
self.assertEqual(scan_all_schedule.cron, self.cron)
|
|
# 10.5 Verify GC schedule
|
|
gc_schedule = schedules[self.job_types[0]]
|
|
self.assertEqual(gc_schedule.cron, self.cron)
|
|
# 10.6 Verify log rotation
|
|
log_rotation_schedule = schedules["PURGE_AUDIT_LOG"]
|
|
self.assertEqual(log_rotation_schedule.cron, self.cron)
|
|
|
|
# 11. Pause all schedules
|
|
self.jobservice.action_pending_jobs("scheduler", "pause")
|
|
|
|
# 12. Verify schedules is Paused;
|
|
self.assertTrue(self.schedule.get_schedule_paused("all").paused)
|
|
|
|
# 13. Resume all schedules
|
|
self.jobservice.action_pending_jobs("scheduler", "resume")
|
|
|
|
# 14. Verify schedules is not Paused
|
|
self.assertFalse(self.schedule.get_schedule_paused("all").paused)
|
|
|
|
# 15. Reset the schedule for scan all, GC, and log rotation
|
|
self.scan_all.update_scan_all_schedule("None", cron="")
|
|
self.gc.update_gc_schedule("None", cron="")
|
|
self.purge.update_purge_schedule("None", "")
|
|
|
|
# 16. Verify schedules
|
|
schedules = self.schedule.list_schedules(page_size=50, page=1)
|
|
# 16.1 Verify retention schedule
|
|
retention_schedule = schedules["%s-%d" % (self.job_types[5], retention_id)]
|
|
self.assertEqual(retention_schedule.cron, self.cron)
|
|
# 16.2 Verify preheat schedule
|
|
preheat_schedule = schedules["%s-%d" % (self.job_types[2], preheat_policy.id)]
|
|
self.assertEqual(preheat_schedule.cron, self.cron)
|
|
# 16.3 Verify replication schedule
|
|
replication_schedule = schedules["%s-%d" % (self.job_types[4], replication_id)]
|
|
self.assertEqual(replication_schedule.vendor_type, self.job_types[4])
|
|
self.assertEqual(replication_schedule.cron, self.cron)
|
|
# 16.4 Verify scan all schedule
|
|
self.assertNotIn("SCAN_ALL", schedules)
|
|
# 16.5 Verify GC schedule
|
|
self.assertNotIn(self.job_types[0], schedules)
|
|
# 16.6 Verify log rotation
|
|
self.assertNotIn("PURGE_AUDIT_LOG", schedules)
|
|
|
|
|
|
def testWorkers(self):
|
|
"""
|
|
Test case:
|
|
Job Service Dashboard Workers
|
|
Test step and expected result:
|
|
1. Get worker pools;
|
|
2. Get workers in current pool;
|
|
3. Stop running job;
|
|
"""
|
|
# 1. Get worker pools
|
|
worker_pools = self.jobservice.get_worker_pools()
|
|
for worker_pool in worker_pools:
|
|
self.assertIsNotNone(worker_pool.pid)
|
|
self.assertIsNotNone(worker_pool.worker_pool_id)
|
|
self.assertIsNotNone(worker_pool.concurrency)
|
|
self.assertIsNotNone(worker_pool.start_at)
|
|
self.assertIsNotNone(worker_pool.heartbeat_at)
|
|
|
|
# 2. Get workers in current pool
|
|
workers = self.jobservice.get_workers(worker_pool.worker_pool_id)
|
|
self.assertEqual(len(workers), worker_pool.concurrency)
|
|
for worker in workers:
|
|
self.assertIsNotNone(worker.id)
|
|
self.assertEqual(worker_pool.worker_pool_id, worker.pool_id)
|
|
|
|
# 3. Stop running job
|
|
self.jobservice.stop_running_job("966a49aa2278b67d743f8aca")
|
|
|
|
|
|
def testJobServiceDashboardAPIPermission(self):
|
|
"""
|
|
Test case:
|
|
Log Rotaion Permission API
|
|
Test step and expected result:
|
|
1. Create a new user(UA);
|
|
2. User(UA) should not have permission to list job queue API;
|
|
3. User(UA) should not have permission to action_pending_jobs API;
|
|
4. User(UA) should not have permission to stop running job API;
|
|
5. User(UA) should not have permission to get worker pools API;
|
|
6. User(UA) should not have permission to get workers in current pool API;
|
|
7. User(UA) should not have permission to list schedules API;
|
|
8. User(UA) should have permission to get scheduler paused status API;
|
|
9. Verify that the get scheduler paused status API parameter job_type only support all;
|
|
"""
|
|
expect_status_code = 403
|
|
expect_response_body = "FORBIDDEN"
|
|
|
|
# 1. Create a new user(UA)
|
|
user_password = "Aa123456"
|
|
_, user_name = self.user.create_user(user_password = user_password)
|
|
USER_CLIENT = dict(endpoint = ADMIN_CLIENT["endpoint"], username = user_name, password = user_password)
|
|
|
|
# 2. User(UA) should not have permission to list job queue API
|
|
self.jobservice.get_job_queues(expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 3. User(UA) should not have permission to action_pending_jobs API
|
|
self.jobservice.action_pending_jobs(self.job_types[0], "pause", expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 4. User(UA) should not have permission to stop running job API
|
|
self.jobservice.stop_running_job("966a49aa2278b67d743f8aca", expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 5. User(UA) should not have permission to get worker pools API
|
|
self.jobservice.get_worker_pools(expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 6. User(UA) should not have permission to get workers in current pool API
|
|
self.jobservice.get_workers("13e0cfe999715102c47614ec", expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 7. User(UA) should not have permission to list schedules API
|
|
self.schedule.list_schedules(expect_status_code=expect_status_code, expect_response_body=expect_response_body, **USER_CLIENT)
|
|
|
|
# 8. User(UA) should have permission to get scheduler paused status API
|
|
self.schedule.get_schedule_paused("all")
|
|
|
|
# 9. Verify that the get scheduler paused status API parameter job_type only support all
|
|
self.schedule.get_schedule_paused(self.job_types[0], expect_status_code=400, expect_response_body="job_type can only be 'all'")
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|