[ 
https://issues.apache.org/jira/browse/YARN-2990?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14278300#comment-14278300
 ] 

Karthik Kambatla commented on YARN-2990:
----------------------------------------

Looked more closely, unfortunately this appears to be by design. Each 
application has an "allowed locality level" - initially node-local, and 
transitions to rack-local and off-switch after the corresponding delays. 
Instead, it might be better to track this allowed locality level per 
{{ResourceRequest}}. I propose:
# In the short-term, to address the case where the AM has to go through the 
node-local and rack-local delays, we could start with the default locality 
level of off-switch and reset it node-local after the AM is allocated. 
# In the long-term, let us augment ResourceRequest to include allowed locality 
level. 

Thoughts? 

> FairScheduler's delay-scheduling always waits for node-local and rack-local 
> delays, even for off-rack-only requests
> -------------------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-2990
>                 URL: https://issues.apache.org/jira/browse/YARN-2990
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: fairscheduler
>    Affects Versions: 2.6.0
>            Reporter: Karthik Kambatla
>            Assignee: Karthik Kambatla
>         Attachments: yarn-2990-test.patch
>
>
> Looking at the FairScheduler, it appears the node/rack locality delays are 
> used for all requests, even those that are only off-rack. 
> More details in comments. 



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

Reply via email to