You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Raymond Liu (JIRA)" <ji...@apache.org> on 2015/09/08 05:06:46 UTC

[jira] [Commented] (YARN-4125) In addition to aggregate availability for an app, headroom should provide information on largest container that can be allcoated

    [ https://issues.apache.org/jira/browse/YARN-4125?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14734168#comment-14734168 ] 

Raymond Liu commented on YARN-4125:
-----------------------------------

[~kasha@cloudera.com] regarding provide the information about largest container allocatable.  how would you decide the value? 
say if the <memory, cpu> capability is:
<4096, 0>
<3172, 1>
<2048, 2>
<1024, 3>
<0, 4>
which one will be the largest container allocatable?

> In addition to aggregate availability for an app, headroom should provide information on largest container that can be allcoated
> --------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-4125
>                 URL: https://issues.apache.org/jira/browse/YARN-4125
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: scheduler
>    Affects Versions: 2.7.1
>            Reporter: Karthik Kambatla
>
> Today, headroom is an indicator to an application on the resources available to it. It is an aggregate across all nodes. If all the nodes have either only memory or only cpu available, no container can be allocated. The headroom should hint at this too, so apps can make decisions sooner. MAPREDUCE-6470 is a case of this, and has an example for the curious. 
> May be, we should add an API that returns both aggregate headroom and large container allocatable. 



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