[
https://issues.apache.org/jira/browse/YARN-6156?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15863577#comment-15863577
]
Sunil G commented on YARN-6156:
-------------------------------
bq.We should only add this node(i.e. count it towards NM count) and not remove
it when AM containers fail then.
Yes.
bq.Maybe update it when RMContainer is created
I think we can try to do this when APP is LAUNCHED also. I am not so sure.
bq.And a similar thing will have to be done in YARN-6148 if we do consider non
exclusive labels this way.
I think YARN-6148 is more complex than I initial thought. We have to consider
all such cases from label and give a clean api for MR. Then MR has to take a
call on to select per-label node count to blacklist/remove map/reducer from
same label. I guess we need more discussion on that.
> AM blacklistdisableThreshold consider node partition count
> -----------------------------------------------------------
>
> Key: YARN-6156
> URL: https://issues.apache.org/jira/browse/YARN-6156
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Bibin A Chundatt
> Assignee: Bibin A Chundatt
> Attachments: YARN-6156.0001.patch, YARN-6156.0002.patch
>
>
> As per the current implementation for AM blacklisting disabling is based on
> completed custer resource.
> Incase of partitioned cluster, nodes applicable for AM labels only should be
> considered as total nodes.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
