[jira] [Commented] (TEZ-3543) Diagnostics message doesn't show on YARN UI when DAGAppMaster dies before registration.

["Jason Lowe (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TEZ-3543?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15674607#comment-15674607 ] 

Jason Lowe commented on TEZ-3543:
---------------------------------

Unfortunately we cannot provide a diagnostic to the RM without also terminating the application.  If this is the last AM attempt then I think it makes sense to register with the RM and unregister with the final diagnostics.  If this isn't the last attempt then we cannot unregister to provide diagnostics or we won't get any further AM attempts.  To fix this in the general case I think we need YARN-4142 implemented.

> Diagnostics message doesn't show on YARN UI when DAGAppMaster dies before registration.
> ---------------------------------------------------------------------------------------
>
>                 Key: TEZ-3543
>                 URL: https://issues.apache.org/jira/browse/TEZ-3543
>             Project: Apache Tez
>          Issue Type: Improvement
>    Affects Versions: 0.7.1
>            Reporter: Eric Payne
>            Priority: Minor
>
> For example, if the DAGAppMaster dies while trying to set up the server socket due to bind exception, this information does not get back to the app's diagnostic section of the application history server's UI:
> {noformat}
> 2016-11-12 03:39:13,577 [ERROR] [ServiceThread:DAGClientRPCServer] |client.DAGClientServer|: Failed to start DAGClientServer: 
> java.net.BindException: Problem binding to [0.0.0.0:0] java.net.BindException: Could not find a free port in 50500-51000;
> {noformat}
> Is it possible to refactor the DAGAppMaster to capture errors of this type and report them to YARN?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)