[FLINK-30528] Job may be stuck in upgrade loop when last-state fallback is disabled and deployment is missing

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: kubernetes-operator-1.3.0
• Fix Version/s: kubernetes-operator-1.4.0, kubernetes-operator-1.3.1
• Component/s: Kubernetes Operator
• Labels: pull-request-available
• Assignee: Gyula Fora
• Reporter: Gyula Fora
• Created: 28/Dec/22 14:58
• Updated: 01/Jan/23 16:17
• Resolved: 01/Jan/23 16:17

Description

When last-state upgrade fallback is disabled (or we are switching flink versions) and the JM deployment is missing for some reason, we get stuck in the upgrade loop as the spec change /upgrade logic always takes precedence over the JM deployment recovery logic.

In this cases the JM deployment need to be recovered first so savepoint upgrade can be executed.

Issue Links

• links to: GitHub Pull Request #497