[
https://issues.apache.org/jira/browse/FLINK-4273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15426414#comment-15426414
]
ASF GitHub Bot commented on FLINK-4273:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/2313#discussion_r75305261
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/client/JobClientActor.java
---
@@ -198,10 +211,47 @@ else if (message instanceof SubmitJobAndWait) {
decorateMessage(new Status.Failure(new
Exception(msg))), ActorRef.noSender());
}
}
+ else if (message instanceof AttachToJobAndWait) {
--- End diff --
At the moment the `JobClientActor` can be used to attach to a job and
submit a new job at the same time. Then depending on the order in which the
jobs terminate either the submitted job's or the attached job's result is
reported back but the other future will never be completed. It might make sense
to guard against this or to have two different JobClientActors based on the
same base class.
> Refactor JobClientActor to watch already submitted jobs
> --------------------------------------------------------
>
> Key: FLINK-4273
> URL: https://issues.apache.org/jira/browse/FLINK-4273
> Project: Flink
> Issue Type: Sub-task
> Components: Client
> Reporter: Maximilian Michels
> Assignee: Maximilian Michels
> Priority: Minor
> Fix For: 1.2.0
>
>
> The JobClientActor assumes that it receives a job, submits it, and waits for
> the result. This process should be broken up into a submission process and a
> waiting process which can both be entered independently. This leads to two
> different entry points:
> 1) submit(job) -> wait
> 2) retrieve(jobID) -> wait
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
