vsantwana commented on PR #1237: URL: https://github.com/apache/flink-kubernetes-operator/pull/1237#issuecomment-6488934682
We tested this on minikube with Flink 1.20's `StateMachineExample` in native mode, without HA and with `upgradeMode: savepoint`. We ran the same steps against an operator built from `main` and one built from this PR: suspend the job, then delete the JobManager pod if one is still there. On `main`, the resource went to `SUSPENDED` / `FINISHED`, but the JobManager deployment stayed. After we deleted the JobManager pod, the new one submitted the job again from empty state and started a new TaskManager: ``` Submitting Job with JobId=cf5706c1f4146d868fd86d0073ea6217. No checkpoint found during restore. Job State machine job (cf5706c1f4146d868fd86d0073ea6217) switched from state CREATED to RUNNING. ``` The resource kept showing `SUSPENDED` / `FINISHED` the whole time. With this PR, the JobManager deployment was gone as soon as the resource reached `SUSPENDED` (`jobManagerDeploymentStatus: MISSING`), so there was nothing left to restart. Setting `job.state: running` again brought the job back from the stop savepoint: ``` Restoring job 3e3c44d2ce1a7e51d3e821782d7f122e from Savepoint 22 ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
