[
https://issues.apache.org/jira/browse/FLINK-15297?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16998385#comment-16998385
]
Till Rohrmann commented on FLINK-15297:
---------------------------------------
This sounds like a bug to me. In particular in detached mode it could happen
that the deployed job terminates and does not wait for the client to connect.
Hence, I think we should correct this.
> Do not throw exception if YARN Application switched to FINISHED immediately
> after deployed in YarnClusterDescriptor#startAppMaster
> ----------------------------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-15297
> URL: https://issues.apache.org/jira/browse/FLINK-15297
> Project: Flink
> Issue Type: Bug
> Components: Deployment / YARN
> Reporter: Zili Chen
> Assignee: Zili Chen
> Priority: Major
>
> Currently we throw an exception in {{YarnClusterDescriptor#startAppMaster}}
> if we first detect {{FINISHED}} before {{RUNNING}}. However, it is possible a
> legal state that the application finished normally immediately.
> Right now we always try to connect the Dispatcher so it may be fine to throw
> the exception a bit earlier(otherwise when connect to a closed cluster an
> exception thrown also), but it is semantically wrong. Internally we have a
> code path that only required to report the ApplicationReport and it causes
> trouble.
> cc [~trohrmann] what do you think?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
