[
https://issues.apache.org/jira/browse/SLIDER-775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14304844#comment-14304844
]
Steve Loughran commented on SLIDER-775:
---------------------------------------
it looks like there's some specific logic to actually set the {{force}} flag if
you try to kill an app that hasn't yet started. It's likely therefore that we
have seen this problem.
{code}
// IPC request for a managed shutdown is only possible if the app is
running.
// so we need to force kill if the app is accepted or submitted
if (!forcekill
&& app.getYarnApplicationState().ordinal() <
YarnApplicationState.RUNNING.ordinal()) {
log.info("Cluster {} is in a pre-running state {}. Force killing it",
clustername,
app.getYarnApplicationState());
forcekill = true;
}
OK, SLIDER-426 was meant to address this. I don't see a test for it; we'll have
to think about how to test it. Setting up the minicluster to have less capacity
than the client asks for is the obvious strategy
{code}
> Can't kill an application that is stuck in ACCEPTED
> ---------------------------------------------------
>
> Key: SLIDER-775
> URL: https://issues.apache.org/jira/browse/SLIDER-775
> Project: Slider
> Issue Type: Bug
> Components: client
> Affects Versions: Slider 0.60
> Reporter: Steve Loughran
> Priority: Minor
>
> Report of slider not being able to stop, even with {{--force}} an application
> stuck at accepted
> bq. If the application is stuck in ACCEPT status, slider stop command doesn't
> stop the yarn application. I didn't setup the scheduler address correctly so
> the application stuck ACCEPT status and I couldn't use slider stop to stop
> the application
> This may be a simple state-check in the client, or it could be something in
> YARN
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
