[
https://issues.apache.org/jira/browse/YARN-9942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16963236#comment-16963236
]
Íñigo Goiri edited comment on YARN-9942 at 10/30/19 5:12 PM:
-------------------------------------------------------------
[^YARN-9942.000.patch] adds a few constants to make reading easier.
I added:
* OVER_COMMIT_NO_TIMEOUT (-1) to mark when we don't timeout.
* OVER_COMMIT_IMMEDIATE (0) to mark as over committed (and kill) right away.
Open to suggestions for a better word than IMMEDIATE.
The main change is RMNodeImpl#1215 which actually changes the value of the
timeout to avoid triggering killing.
That part needs a unit test.
was (Author: elgoiri):
[^YARN-9942.000.patch] adds a few constants to make reading easier.
The main change is RMNodeImpl#1215 which actually changes the value of the
timeout to avoid triggering killing.
That part needs a unit test.
> Node resource update should use OVER_COMMIT_TIMEOUT_MILLIS_DEFAULT
> ------------------------------------------------------------------
>
> Key: YARN-9942
> URL: https://issues.apache.org/jira/browse/YARN-9942
> Project: Hadoop YARN
> Issue Type: Bug
> Affects Versions: 3.2.1
> Reporter: Íñigo Goiri
> Assignee: Íñigo Goiri
> Priority: Major
> Attachments: YARN-9942.000.patch
>
>
> Currently, there are a few instances where we change the amount of resources
> (e.g., decommissioning) which are using a value of 0. This triggers
> preemption events. We should set it to the default (not do anything).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
