[HBASE-26675] Data race on Compactor.writer

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.5.0, 3.0.0-alpha-3
• Component/s: Compaction
• Hadoop Flags: Reviewed
• Assignee: Duo Zhang
• Reporter: Duo Zhang
• Created: 16/Jan/22 15:09
• Updated: 19/Mar/24 15:02
• Resolved: 24/Jan/22 11:56

Description

In Compactor.getCompactionTargets, we use synchronized on the writer instance but it does not help resolve the data race problem, because in other places we do not use synchronized...

And since it is not a big deal to the upper layer an old snapshot, we could just make the writer instance volatile and use it directly in the method, just make sure we always use the same instance in the method, i.e, assign it to a local var first.

Issue Links

• is broken by: HBASE-26271 Cleanup the broken store files under data directory
• links to: GitHub Pull Request #4035