[IGNITE-14140] Checkpointer thread holds write lock too long

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.11
• Component/s: persistence
• Release Note: Decreased an exclusive checkpoint lock time.
• Ignite Flags: Release Notes Required
• Assignee: Vladislav Pyatkov
• Reporter: Vladislav Pyatkov
• Created: 08/Feb/21 15:06
• Updated: 01/Jul/21 09:42
• Resolved: 11/Feb/21 13:25

Description

Free lists flushing optimization can block db-checkpoint-thread when it got Write lock. It might block all transactions for several hundreds milliseconds.

"db-checkpoint-thread-#334%DPL_GRID%DplGridNodeName%" #667 daemon prio=5 os_prio=0 tid=0x00007e765c123800 nid=0xee0b8 runnable [0x00007e767f535000]
   java.lang.Thread.State: RUNNABLE
	at sun.misc.Unsafe.getObjectVolatile(Native Method)
	at java.util.concurrent.atomic.AtomicReferenceArray.getRaw(AtomicReferenceArray.java:130)
	at java.util.concurrent.atomic.AtomicReferenceArray.get(AtomicReferenceArray.java:125)
	at org.apache.ignite.internal.processors.cache.persistence.freelist.AbstractFreeList.getBucketCache(AbstractFreeList.java:690)
	at org.apache.ignite.internal.processors.cache.persistence.freelist.PagesList.flushBucketsCache(PagesList.java:374)
	at org.apache.ignite.internal.processors.cache.persistence.freelist.PagesList.saveMetadata(PagesList.java:343)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheOffheapManager.saveStoreMetadata(GridCacheOffheapManager.java:373)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheOffheapManager.syncMetadata(GridCacheOffheapManager.java:336)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheOffheapManager.syncMetadata(GridCacheOffheapManager.java:322)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheOffheapManager.onMarkCheckpointBegin(GridCacheOffheapManager.java:247)
	at org.apache.ignite.internal.processors.cache.persistence.checkpoint.CheckpointWorkflow.markCheckpointBegin(CheckpointWorkflow.java:281)
	at org.apache.ignite.internal.processors.cache.persistence.checkpoint.Checkpointer.doCheckpoint(Checkpointer.java:388)
	at org.apache.ignite.internal.processors.cache.persistence.checkpoint.Checkpointer.body(Checkpointer.java:264)
	at org.apache.ignite.internal.util.worker.GridWorker.run(GridWorker.java:119)
	at java.lang.Thread.run(Thread.java:748)

We can to reduce time into Write lock if switch off optimization before the lock will be gotten and enable it after the lock will be left off.

Issue Links

• links to: GitHub Pull Request #8772