[CASSANDRA-19182] IR may leak SSTables with pending repair when coming from streaming

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0.13, 4.1.5, 5.0-rc1
• Component/s: Consistency/Repair
• Bug Category: Correctness - Recoverable Corruption / Loss
• Severity: Critical
• Complexity: Challenging
• Discovered By: User Report
• Platform: All
• Since Version: 4.0

Source Control Link

https://github.com/apache/cassandra/commit/c14abb40b2d0e2e1db121eac65a1264a287bcd18

Test and Documentation Plan

updated tests

• Assignee: David Capwell
• Reporter: David Capwell

Authors

David Capwell

Reviewers

Blake Eggleston

• Created: 07/Dec/23 23:11
• Updated: 01/May/24 21:56
• Resolved: 01/May/24 21:38

Description

There is a race condition where SSTables from streaming may race with pending repair cleanup in compaction causing us to cleanup the pending repair state in compaction while the SSTables are being added to it; this leads to IR failing in the future when those files get selected for repair.

This problem was hard to track down as the in-memory state was wiped, so we don’t have any details.  To better aid these types of investigation we should make sure the repair vtables get updated when IR session failures are submitted

Issue Links

• links to: GH: 4.0
• links to: GH: 4.1
• links to: GH: 5.0
• links to: GH: trunk
• links to: GitHub Pull Request #2976