[
https://issues.apache.org/jira/browse/YARN-5464?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16157232#comment-16157232
]
Ming Ma commented on YARN-5464:
-------------------------------
I want to bring up the support for relative timeout value at the command line
and at RPC level. For ease of use, seems relative timeout is better than
absolute timestampe. However at RPC {{RefreshNodesRequest}} level we can
consider using absolute timestamp and have admin CLI convert relative timeout
to absolute timestamp before sending to RM. The reason is as follows:
* The initial patch from [~rkanter] is quite complicated. Seems that is due to
the relative timestamp semantics. If we were to use absolute timestamp, I image
RM can reconstruct decommission status from other persistent state and
eliminate the need for additional state.
* If we agree specifying absolute timestamp in JSON file is better, it might be
more consistent to have use absolute timestamp in {{RefreshNodesRequest}} as
the JSON functionality might end up using {{RefreshNodesRequest}}.
Thoughts?
> 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
