hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sunil G (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-6164) Expose maximum-am-resource-percent in YarnClient
Date Mon, 27 Feb 2017 11:51:45 GMT

    [ https://issues.apache.org/jira/browse/YARN-6164?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15885641#comment-15885641
] 

Sunil G commented on YARN-6164:
-------------------------------

Thanks [~benson.qiu] for the patch

There are some more high level thoughts after i was checking various CLI and other PBImpl
code.

# We are not sharing per-label Queue resource usage
# We are not sharing per-label Queue configurations (capacity etc).

More or less I am thinking in line of thoughts shared by [~bibinchundatt] earlier. 
bq.Since many informations related to configuration is already exposed by QueueInfo, Properties
may not a become a better logical grouping unless we move all existing config items to new
subclass. 
I still kind of feel above statement makes sense. However this is similar to a problem addressed
in {{QueueCapacities}} etc. And in long run, this may be helpful.

My proposal is something like below.

In {{QueueInfo}}, lets have an api called {{setQueueConfigurations}}.  A new api record could
be added named {{QueueConfigurations}} and it can have below sub items
# capacity
# abs-capacity
# max-capacity
# abs-max-capacity
# max-am-perc

If we have a map in {{QueueInfo}} per-label, then {{QueueConfigurations}} could be mapped
against its accessable-labels.

I would like to have discussion on this approach and we can make it more generic. [~benson.qiu],
[~bibinchundatt] and [~rohithsharma] , thoughts?

> Expose maximum-am-resource-percent in YarnClient
> ------------------------------------------------
>
>                 Key: YARN-6164
>                 URL: https://issues.apache.org/jira/browse/YARN-6164
>             Project: Hadoop YARN
>          Issue Type: Improvement
>    Affects Versions: 2.7.2
>            Reporter: Benson Qiu
>            Assignee: Benson Qiu
>         Attachments: YARN-6164.001.patch, YARN-6164.002.patch, YARN-6164.003.patch, YARN-6164.004.patch,
YARN-6164.005.patch
>
>
> `yarn.scheduler.capacity.maximum-am-resource-percent` is exposed through the [Cluster
Scheduler API|http://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/ResourceManagerRest.html#Cluster_Scheduler_API],
but not through [YarnClient|https://hadoop.apache.org/docs/current/api/org/apache/hadoop/yarn/client/api/YarnClient.html].
> Since YarnClient and RM REST APIs depend on different ports (8032 vs 8088 by default),
it would be nice to expose `maximum-am-resource-percent` in YarnClient as well. 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

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


Mime
View raw message