[FLINK-31498] DeclartiveSlotManager always request redundant task manager when resource is not enough

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Done
• Fix Version/s: 1.18.0
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: !huwh
• Reporter: !huwh
• Created: 17/Mar/23 10:09
• Updated: 22/May/23 05:56
• Resolved: 22/May/23 05:56

Description

Currently redundant task manager check in DeclarativeSlotManager only compare free slots with required redundant slots.

when there are no enough resources in YARN/Kubernetes, this mechanism will always try to request new task manager.

there are two way to address this.
1. maintain the state of redundant workers to avoid request twice
2. only try to request redundant workers when there is no pending worker

The first way will make the logic of redundant worker too complicated, I would like to choose the second way

Looking forward to any suggestion.

Issue Links

• links to: GitHub Pull Request #22246