[
https://issues.apache.org/jira/browse/FLINK-10535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654890#comment-16654890
]
Till Rohrmann commented on FLINK-10535:
---------------------------------------
Hi [~yinhua], this is actually an artifact of how the {{ExecutionEnvironments}}
work currently. If you start a per-job mode in attached mode, Flink will
actually start a session cluster underneath because it could be that your job
consists of multiple parts (e.g. by using eager execution calls like
{{collect}} and {{print}} when using {{DataSet}} API). Therefore, we follow the
session cluster code path which first starts the cluster and then uploads the
user code jars.
In the future this will change once we update how the {{ExecutionEnvironments}}
work. Therefore, I will close this issue.
> User jar is present in the flink job manager's class path
> ---------------------------------------------------------
>
> Key: FLINK-10535
> URL: https://issues.apache.org/jira/browse/FLINK-10535
> Project: Flink
> Issue Type: Bug
> Components: YARN
> Affects Versions: 1.5.0, 1.6.1
> Reporter: yinhua.dai
> Priority: Major
>
> Run flink on yarn with attach and per-job mode, i.e.
> {code:java}
> flink run -m yarn-cluster -yn 1 *** user.jar
> {code}
> The user.jar will not be present in flink job manager's class path, but as we
> observed that in flink 1.3 it will by default put to the beginning of flink
> job manager's class path.
>
> cc [~GJL] [~twalthr]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
