[SOLR-13237] Not all types of index corruption garuntee a leader will "give up its leadership"

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Unassigned
• Reporter: Chris M. Hostetter
• Created: 09/Feb/19 00:44
• Updated: 07/Dec/20 22:21

Description

While investigating failures from LeaderTragicEventTest, I've found some reproducible situations where (externally introduced) index corruption can cause a leader to reject updates, but not automatically give up it's leadership.

See discussion in LUCENE-8692 – notably simon's comment on why/how some things are explicitly not treated as tragic today for a disucssion of the root cause.

We may need/want to rethink & improve the situations where a leader gives up leadership, above and beyond IW registering a tragic exception

 

Issue Links

• is part of: LUCENE-8692 IndexWriter.getTragicException() may not reflect all corrupting exceptions (notably: NoSuchFileException)
• is related to: SOLR-13027 Harden LeaderTragicEventTest.
• relates to: SOLR-15029 More gracefully allow Shard Leader to give up leadership