[
https://issues.apache.org/jira/browse/YARN-5464?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16157579#comment-16157579
]
Robert Kanter commented on YARN-5464:
-------------------------------------
One of the reasons that the patch is so complicated is the way that the state
store classes are written, and that we have 4 implementations.
I'm not sure there's any other existing state info that we could use instead,
regardless of how we keep track of the decom timeout. They can be arbitrary
based on what the user entered. The only alternative I can think of would be
if the RM re-read the excludes file, but then the user would have to sync the
excludes file between the RMs and we'd have to use the absolute timeouts. Is
that what you were thinking [~mingma]? Though that won't work if the timeout
is specified on the CLI.
> Server-Side NM Graceful Decommissioning with RM HA
> --------------------------------------------------
>
> Key: YARN-5464
> URL: https://issues.apache.org/jira/browse/YARN-5464
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: graceful
> Reporter: Robert Kanter
> Attachments: YARN-5464.wip.patch
>
>
> Make sure to remove the note added by YARN-7094 about RM HA failover not
> working right.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
