[CASSANDRA-11479] BatchlogManager unit tests failing on truncate race condition

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Cannot Reproduce
• Fix Version/s: 2.2.x, 3.0.26, 3.11.12
• Component/s: Local/Compaction
• Severity: Normal
• Assignee: Yuki Morishita
• Reporter: Joel Knighton

Authors

Yuki Morishita

Reviewers

Joel Knighton

• Created: 01/Apr/16 20:50
• Updated: 17/Feb/22 08:13
• Resolved: 08/Sep/21 08:42

Description

Example on CI here. This seems to have only started happening relatively recently (within the last month or two).

As far as I can tell, this is only showing up on BatchlogManagerTests purely because it is an aggressive user of truncate. The assertion is hit in the setUp method, so it can happen before any of the test methods. The assertion occurs because a compaction is happening when truncate wants to discard SSTables; trace level logs suggest that this compaction is submitted after the pause on the CompactionStrategyManager.

This should be reproducible by running BatchlogManagerTest in a loop - it takes up to half an hour in my experience. A trace-level log from such a run is attached - grep for my added log message "SSTABLES COMPACTING WHEN DISCARDING" to find when the assert is hit.