[CASSANDRA-17830] Read Unavailables due to GossipStage backlog with high number of pending tasks

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 5.0-alpha1, 5.0
• Component/s: Cluster/Gossip
• Change Category: Operability
• Complexity: Normal
• Platform: All

Source Control Link

https://gitbox.apache.org/repos/asf?p=cassandra.git;a=commit;h=902dcba740300797684ed3c8c355ecc84373210d

Test and Documentation Plan

No new testing required; perform existing tests.

• Assignee: Josh McKenzie
• Reporter: Josh McKenzie

Authors

Josh McKenzie

Reviewers

Marcus Eriksson

• Created: 17/Aug/22 20:04
• Updated: 12/Sep/23 13:02
• Resolved: 18/Aug/22 16:14

Description

In CASSANDRA-16759, like we changed from iterating over liveMembers + unreachableMembers to all keys in endpointStateMap here.

We can have an endpoint in quarantine (in justRemovedEndpoints) which has no release version information. In this case we don't memoize the value and instead recalculate the min version for every endpoint for every incoming gossip message... which can be bad.