[HBASE-28140] AbstractWALProvider may miss the WAL which is under creation in getWALs method

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0-beta-1
• Component/s: Replication, wal
• Hadoop Flags: Reviewed
• Assignee: Duo Zhang
• Reporter: Duo Zhang
• Created: 08/Oct/23 11:24
• Updated: 15/Oct/23 20:05
• Resolved: 13/Oct/23 14:30

Description

Found this problem when analysis HBASE-28114.

In the AbstractFSWALProvider implementation, if wal field is null, we will acquire the creation lock so we can get the WAL instance which is under creation, so we will not miss the WAL instance, and also will not get an uninitialized WAL instance.

We should try to do the same for remote WAL, otherwise the logic in WALEntryStream, where we rely on the return value of getWALs to test whether a WAL file is currently being written, will be broken.

Issue Links

• links to: GitHub Pull Request #5455