Fix job recovery and savepoint bug

GoogleCloudPlatform / flink-on-k8s-operator

[DEPRECATED] Kubernetes operator for managing the lifecycle of Apache Flink and Beam applications.

Apache License 2.0

657 stars 265 forks source link

Fix job recovery and savepoint bug #401

Closed elanv closed 3 years ago

elanv commented 3 years ago

Job recovery bug When job manager fails and job state falls to Lost, job is not properly recovered. The job is resubmitted, but the previous job ID is not cleared, so it is considered an unexpected job and is canceled.
Savepoint bug Savepoint is not triggered properly.

Resolves #398

elanv commented 3 years ago

@shashken @functicons I have also fixed the issue https://github.com/GoogleCloudPlatform/flink-on-k8s-operator/pull/392#discussion_r568605162 because savepoint is required for this PR.

shashken commented 3 years ago

@elanv good job, @functicons this needs to be approved quickly, the version on master does not take SP. I haven't done a complete test after I pushed CR changes and that line: error != nil broken

morelina commented 3 years ago

I am using the commit in this PR and I am having issues when updating the job: https://github.com/GoogleCloudPlatform/flink-on-k8s-operator/issues/408

functicons commented 3 years ago

/gcbrun

pgandhijr commented 3 years ago

I am experiencing this bug where when the job manager fails the task is submitted but it fails because it is using a a old job id. I am using tag v1beta-9. Is there an older stable version that is not affected by this bug that we could use?