flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-4046) Failing a restarting job can get stuck in JobStatus.FAILING
Date Thu, 23 Jun 2016 11:55:16 GMT

    [ https://issues.apache.org/jira/browse/FLINK-4046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15346310#comment-15346310

ASF GitHub Bot commented on FLINK-4046:

Github user tillrohrmann commented on the issue:

    Thanks for the review @uce. I think you're right and we should also accept the `FAILED`
state when calling `restart` and simply print a log message instead of throwing the `IllegalStateException`.
But I'm not so sure for any terminal state because we should not have reached the state `FINISHED`.
This usually indicates a failure and should not go unnoticed.
    I will add the `FAILED` check and then merge the PR.

> Failing a restarting job can get stuck in JobStatus.FAILING
> -----------------------------------------------------------
>                 Key: FLINK-4046
>                 URL: https://issues.apache.org/jira/browse/FLINK-4046
>             Project: Flink
>          Issue Type: Bug
>          Components: Distributed Coordination
>    Affects Versions: 1.1.0
>            Reporter: Till Rohrmann
>             Fix For: 1.1.0
> When a job is in state {{RESTARTING}}, then it can happen that all of its {{ExecutionJobVertices}}
are in a final state (if they have not been reset). When calling {{fail}} on this {{ExecutionGraph}}
will transition the state to {{FAILING}} and call cancel on all {{ExecutionJobVertices}}.
The job state {{FAILING}} can only be left iff all {{ExecutionJobVertices}} have reached a
final state. The notification of this final state is only sent to the {{ExecutionGraph}} when
all subtasks of an {{ExecutionJobVertex}} have transitioned to a final state. However, this
won't happen because the {{ExeuctionJobVertices}} are already in a final state. The result
is that a job can get stuck in the state {{FAILING}} if {{fail}} is called on a {{RESTARTING}}
> I propose to add a direct transition from {{RESTARTING}} to {{FAILED}} as it is the case
for the {{cancel}} call (transition from {{RESTARTING}} to {{CANCELED}}).

This message was sent by Atlassian JIRA

View raw message