[
https://issues.apache.org/jira/browse/FLINK-17273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17090578#comment-17090578
]
Till Rohrmann commented on FLINK-17273:
---------------------------------------
I think part of the problem why we missed to call this function is that
{{ResourceManager}} does not enforce a certain control flow. I think it would
be better if the {{ResourceManager}} offered some calls like
{{notifyWorkerFailed}} which will trigger the failover behaviour controlled by
the {{ResourceManager}} and not by the sub class. In order to make this work, I
guess we should take a look at the overall architecture and think about what
callbacks the {{ResourceManager}} would need in order to do its job. Then the
{{ResourceManager}} should be responsible for reacting to failures and other
signals and simply call the implementation specific callbacks (e.g. terminating
a pod). In contrast to that, our current {{ResourceManager}} implementations
handle most of the logic themselves which can lead to problems such as
forgetting to call a method in order to not violate the contract.
> Fix not calling ResourceManager#closeTaskManagerConnection in
> KubernetesResourceManager in case of registered TaskExecutor failure
> ----------------------------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-17273
> URL: https://issues.apache.org/jira/browse/FLINK-17273
> Project: Flink
> Issue Type: Bug
> Components: Deployment / Kubernetes, Runtime / Coordination
> Affects Versions: 1.10.0, 1.10.1
> Reporter: Canbin Zheng
> Assignee: Canbin Zheng
> Priority: Major
> Fix For: 1.11.0
>
>
> At the moment, the {{KubernetesResourceManager}} does not call the method of
> {{ResourceManager#closeTaskManagerConnection}} once it detects that a
> currently registered task executor has failed. This ticket propoeses to fix
> this problem.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
