[jira] [Commented] (YARN-4792) FairScheduler should do sanity checks on its configuration

["Yufei Gu (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/YARN-4792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15214912#comment-15214912 ] 

Yufei Gu commented on YARN-4792:
--------------------------------

After discussion with Karthik, we think it won't be necessary. It is absolutely not efficient to configure these settings manually, especially considering clusters keep growing and shrinking. We don't think any customer is willing to do that. Fair Scheduler with preemption features will take care of these automatically.

> FairScheduler should do sanity checks on its configuration
> ----------------------------------------------------------
>
>                 Key: YARN-4792
>                 URL: https://issues.apache.org/jira/browse/YARN-4792
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: fairscheduler
>    Affects Versions: 2.9.0
>            Reporter: Yufei Gu
>            Assignee: Yufei Gu
>
> The FairScheduler does not perform any sanity checks on the configuration that it uses.
> This can lead to a configuration that is legal but does not make sense in a cluster to use, and causes support cases to be filed:
> limit the root queue to a certain size for memory or vcores: the cluster resources limit the root queue already and there should be no need for an extra artificial limit
> setting max running applications on a leaf queue larger than the parent queue(s)
> setting max running applications larger than the queue size (example: minimum vcore allocation * number of apps > total vcores in the queue)
> There are possibly more checks



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)