hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Subru Krishnan (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (YARN-5711) AM cannot reconnect to RM after failover when using RequestHedgingRMFailoverProxyProvider
Date Fri, 21 Oct 2016 01:42:58 GMT

    [ https://issues.apache.org/jira/browse/YARN-5711?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15593646#comment-15593646
] 

Subru Krishnan edited comment on YARN-5711 at 10/21/16 1:42 AM:
----------------------------------------------------------------

Attaching a patch that returns any exception encountered with the active RM as discussed offline
with [~jianhe]. Thanks to [~ellenfkh] for extensively testing this out in our cluster.

FYI, there are some formatting fixes in *RequestHedgingRMFailoverProxyProvider* as it seems
to follow intellij formatter rather than standard hadoop. 


was (Author: subru):
Attaching a patch that returns any exception encountered with the active RM as discussed offline
with [~jianhe]. Thanks to [~ellenfkh] for extensively testing this out in our cluster.

> AM cannot reconnect to RM after failover when using RequestHedgingRMFailoverProxyProvider
> -----------------------------------------------------------------------------------------
>
>                 Key: YARN-5711
>                 URL: https://issues.apache.org/jira/browse/YARN-5711
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: applications, resourcemanager
>    Affects Versions: 2.9.0, 3.0.0-alpha1
>            Reporter: Subru Krishnan
>            Assignee: Subru Krishnan
>            Priority: Critical
>         Attachments: YARN-5711-v1.patch
>
>
> When RM failsover, it does _not_ auto re-register running apps and so they need to re-register
when reconnecting to new primary. This is done by catching {{ApplicationMasterNotRegisteredException}}
in *allocate* calls and re-registering. But *RequestHedgingRMFailoverProxyProvider* does _not_
propagate {{YarnException}} as the actual invocation is done asynchronously using seperate
threads, so AMs cannot reconnect to RM after failover. 
> This JIRA proposes that the *RequestHedgingRMFailoverProxyProvider* propagate any {{YarnException}}
that it encounters.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org


Mime
View raw message