[
https://issues.apache.org/jira/browse/YARN-2013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14054428#comment-14054428
]
Junping Du commented on YARN-2013:
----------------------------------
[~gtCarrera9], I reopen YARN-2242 as we agreed to address RM/NM side
separately. Let's do an improved patch on that jira.
[~ozawa], Thanks for the patch here which is in good direction. Do you think we
should do similar thing with LinuxContainerExecutor? If so, please add. Also, I
think it is better to add some unit test (i.e. add in TestContainerLaunch.java)
to verify messages.
> The diagnostics is always the ExitCodeException stack when the container
> crashes
> --------------------------------------------------------------------------------
>
> Key: YARN-2013
> URL: https://issues.apache.org/jira/browse/YARN-2013
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: nodemanager
> Reporter: Zhijie Shen
> Assignee: Tsuyoshi OZAWA
> Attachments: YARN-2013.1.patch, YARN-2013.2.patch,
> YARN-2013.3-2.patch, YARN-2013.3.patch
>
>
> When a container crashes, ExitCodeException will be thrown from Shell.
> Default/LinuxContainerExecutor captures the exception, put the exception
> stack into the diagnostic. Therefore, the exception stack is always the same.
> {code}
> String diagnostics = "Exception from container-launch: \n"
> + StringUtils.stringifyException(e) + "\n" + shExec.getOutput();
> container.handle(new ContainerDiagnosticsUpdateEvent(containerId,
> diagnostics));
> {code}
> In addition, it seems that the exception always has a empty message as
> there's no message from stderr. Hence the diagnostics is not of much use for
> users to analyze the reason of container crash.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
