[
https://issues.apache.org/jira/browse/YARN-5031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15270245#comment-15270245
]
sandflee commented on YARN-5031:
--------------------------------
We had a gpu cluster with just one queue, 10+ machines. most jobs run several
days or weeks.
1, if a job reserves resource on a node , it may wait days or weeks to really
get the resource.
2, to avoid starvation of large container request and other consideration , we
fill up one node then another, not spread containers to all nodes(YARN-4962),
and this behavior did't except container reservation either.
> Add a conf to disable container reservation
> -------------------------------------------
>
> Key: YARN-5031
> URL: https://issues.apache.org/jira/browse/YARN-5031
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: sandflee
>
> in a cluster with long running services, container reservation may starve
> other small request services
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
