[CASSANDRA-16446] Parent repair sessions leak may lead to node long pauses

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0-rc1, 4.0
• Component/s: Consistency/Repair
• Bug Category: Degradation
• Severity: Normal
• Complexity: Normal
• Discovered By: User Report
• Platform: All
• Since Version: 4.0-alpha1

Source Control Link

https://github.com/apache/cassandra/commit/23512cf3da5e8206d8797841f2238cdd86c13d96 https://github.com/apache/cassandra-dtest/commit/c89dea0e8c38ed35ed40d59c975a07585584a637

Test and Documentation Plan

See PR

• Assignee: Berenguer Blasi
• Reporter: Berenguer Blasi

Authors

Berenguer Blasi

Reviewers

Andres de la Peña, Ekaterina Dimitrova

• Created: 15/Feb/21 05:57
• Updated: 01/Dec/21 17:28
• Resolved: 24/Feb/21 18:56

Description

ActiveRepairService keeps a map `parentRepairSessions`. If these sessions leak, that map can grow to a size when a node restarts ActiveRepairService.onRestart() triggers a cleanup of sessions that can pause nodes in a cluster for a long time.

The proposed solution is for repairs to cleanup these sessions on all nodes on completion by sending a CLEANUP message to involved nodes. Tests rely on a new parentRepairSessionsCount() method on the parent repair sessions MBean to keep track of these.

Issue Links

• links to: Cassandra PR
• links to: dtests PR