[ 
https://issues.apache.org/jira/browse/YARN-6384?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16217644#comment-16217644
 ] 

Daniel Templeton commented on YARN-6384:
----------------------------------------

A couple comments:

# Should this feature be on by default (when strict mode is on)?  Seems to me 
the default should be not to limit the consumption of an individual container 
in strict mode.  Maybe the default should be max int? Or maybe make it -1, and 
have -1 mean disabled?
# In {{CGroupsCpuResourceHandlerImpl}}, there's no need to initialize 
{{maxResourceUsagePercentInStrictMode}} to 2.0.
# I give everyone a hard time about user-facing docs.  The message for 
{{yarn-default.xml}} could be a bit clearer.  What about something like, "This 
property configuration sets the maximum amount of CPU resources that can be 
consumed by a single container when 
yarn.nodemanager.linux-container-executor.cgroups.strict-resource-usage is set 
to true, as a multiple of the number of vcores assigned to the container. The 
default value is 2.0, which means that each container will be prevented from 
consuming more than 2.0 times the number of vcores assigned to it when strict 
usage mode is enabled."

> Add configuratin to set max cpu usage when strict-resource-usage is false 
> with cgroups
> --------------------------------------------------------------------------------------
>
>                 Key: YARN-6384
>                 URL: https://issues.apache.org/jira/browse/YARN-6384
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: dengkai
>         Attachments: YARN-6384-0.patch, YARN-6384-1.patch, YARN-6384-2.patch, 
> YARN-6384-3.patch
>
>
> When using cgroups on yarn, if 
> yarn.nodemanager.linux-container-executor.cgroups.strict-resource-usage is 
> false, user may get very more cpu time than expected based on the vcores. 
> There should be a upper limit even resource-usage is not strict, just like a 
> percentage which user can get more than promised by vcores. I think it's 
> important in a shared cluster.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to