[FLINK-30462] DefaultMultipleComponentLeaderElectionService saves wrong leader session ID

• Type: Sub-task
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.16.0, 1.17.0, 1.15.3
• Fix Version/s: 1.17.0, 1.15.4, 1.16.2
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: Matthias Pohl
• Reporter: Matthias Pohl
• Created: 20/Dec/22 11:56
• Updated: 02/Feb/23 10:26
• Resolved: 01/Feb/23 13:41

Description

We save a UUID in DefaultMultipleComponentLeaderElectionService:195 but use a different one for granting the leadership to each of the components (see DefaultMultipleComponentLeaderElectionService:199).

We reach an inconsistent state if the JM acquired the leadership before another component registers itself at the MultipleComponentLeaderElectionService. In that case, a different session ID is passed than the one that is set for components that have been registered before the leadership was acquired by the JobManager. That could be the case for JobMaster leadershipt.

Issue Links

• is related to: FLINK-30474 DefaultMultipleComponentLeaderElectionService triggers HA backend change even if it's not the leader
• links to: GitHub Pull Request #21537
• links to: GitHub Pull Request #21816
• links to: GitHub Pull Request #21817