gyfora commented on PR #357: URL: https://github.com/apache/flink-kubernetes-operator/pull/357#issuecomment-1232535054
@Grypse I understand your concern regarding the JM. Let me try to address some of your points. Your biggest argument for using StatefulSets for JM is improved recovery time with a PVC. In my experience JM recovery time is almost never an issue, and having PVC vs remote storage have almost no difference in speed for recovering checkpoint/ha metadata as they are not that large. Furthermore I think even with the current Deployment + podTemplate architecture you can configure a PVC if you really need to. The Flink community also decided to use Deployment instead of StatefulSet for implementing the native integration and I don't remember people hitting any issues that (but of course I could be wrong here). Based on this I don't think we need to implement this right away. In any case if we cannot agree on this we should discuss further on the dev mailing list to include the community. I think we shouldn't downplay the cost of increased complexity in the operator. Instead of maintaining 2 parallel observer paths we could instead focus on making the Deployment observer logic more robust and better. -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
