Jason Lowe commented on YARN-3967:

Thanks for filing the followup.

bq. This was to keep the implementation consistent with what the YarnClient 
does. If we want to make the change, I would suggest changing both YarnClient 
and AppReportFetcher to behave in a similar manner.

That's unfortunate, since it's a convoluted way to handle the exception.  It's 
much clearer to catch the specific exception rather than the parent exception 
and rethrow it when it's not the specific one.  Even elsewhere in 
YarnClientImpl it does the right thing and catches ApplicationNotFoundException 
directly, see YarnClientImpl.submitApplication.

> Fetch the application report from the AHS if the RM does not know about it
> --------------------------------------------------------------------------
>                 Key: YARN-3967
>                 URL: https://issues.apache.org/jira/browse/YARN-3967
>             Project: Hadoop YARN
>          Issue Type: Bug
>    Affects Versions: 2.7.1
>            Reporter: Mit Desai
>            Assignee: Mit Desai
>             Fix For: 2.7.2
>         Attachments: YARN-3967.1.patch, YARN-3967.2.patch, YARN-3967.3.patch
> If the application history service has been enabled and RM has forgotten 
> anout an application, try and fetch the app report form the AHS.
> On larger clusters, the RM can forget about the applications in about 30 
> minutes. The proxy url generated during the job submission will try to fetch 
> the app report from the RM and will fail to get anything from there. If the 
> app is not found in the RM, we will need to get the application report from 
> the Application History Server  (if it is enabled) to see if we can get any 
> information on that application before throwing an exception.

This message was sent by Atlassian JIRA

Reply via email to