[CASSANDRA-18430] When decommissioning should set Severity to limit traffic

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

Source Control Link

https://github.com/apache/cassandra/commit/8b0dc8ad6b5e657cf5946eaed17da0fdabaa53f4

Test and Documentation Plan

tests

• Assignee: David Capwell
• Reporter: David Capwell

Authors

David Capwell

Reviewers

Brandon Williams

• Created: 05/Apr/23 18:43
• Updated: 12/Sep/23 13:03
• Resolved: 25/Apr/23 16:27

Description

When we are decommissioning we first set LEAVING, then LEFT, then disable networking; timeouts start to follow at this last stage. LEFT nodes should not be seen as part of the ring, but that may not be seen before the network is disabled. To better mitigate timeouts we should set severity as part of decom during the LEAVING phase; by setting severity reads should deprioritize traffic to this node.

Remote DC writes do not leverage proximity or severity and instead use random for its select, writes may still timeout even though we know the node is leaving, and severity is set… to work in this model we should update remote DC writes to deprioritize nodes with severity set

Issue Links

• links to: CI: trunk
• links to: GH: trunk
• links to: GitHub Pull Request #2261