[
https://issues.apache.org/jira/browse/DRILL-143?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16458866#comment-16458866
]
ASF GitHub Bot commented on DRILL-143:
--------------------------------------
Github user paul-rogers commented on the issue:
https://github.com/apache/drill/pull/1239
Just to be clear, I have no objection Drill enforcing its own cgroup limits.
My point is rather that CPU limits must be integrated with YARN, via the
DoY config file, so that the user specifies CPU limits in one place and that
limit is the same one passed to YARN for container allocation and to whichever
code is doing cgroup enforcement. By default, it will be YARN if cgroups are
enabled in YARN, as they can be for Apache YARN.
It is then fine to have an additional option to enable self-enforcement in
Drill for those odd cases where users cannot or choose not to use versions of
YARN that do the work.
The problem would be if the user has to configure CPU in two places: in a
shell script to enable self-enforcement, and in the DoY config to tell YARN the
container size. This goes against the ease-of-use experience DoY was designed
to provide. (That is, getting multiple to work consistently correct is very
hard if you're just trying to get Drill to run and are not a Drill internals
expert.)
> Support CGROUPs resource management
> -----------------------------------
>
> Key: DRILL-143
> URL: https://issues.apache.org/jira/browse/DRILL-143
> Project: Apache Drill
> Issue Type: New Feature
> Reporter: Jacques Nadeau
> Assignee: Kunal Khatua
> Priority: Major
> Labels: doc-impacting
> Fix For: 1.14.0
>
> Attachments: 253ce178-ddeb-e482-cd64-44ab7284ad1c.sys.drill
>
>
> For the purpose of playing nice on clusters that don't have YARN, we should
> write up configuration and scripts to allows users to run Drill next to
> existing workloads without sharing resources.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
