[ 
https://issues.apache.org/jira/browse/YARN-10503?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17307760#comment-17307760
 ] 

Andras Gyori commented on YARN-10503:
-------------------------------------

Thank you [~zhuqi] for the patch. I agree with [~pbacsko] and [~ebadger] in 
that the only primary resources are vcore and memory, everything else should be 
considered as a secondary resource. I think the new revision mirrors this 
approach.

The compiling the Resource string is extremely error prone in my opinion, this 
should be handled in ResourceUtils (it even has a createResourceFromString 
method). With that in mind, I think problem is more complex than it seems. I 
believe there are a lot of places  where the secondary resources are hardcoded 
(eg. ResourceUtils#parseResourcesString, where FPGA and GPU is hardcoded), and 
also there could be numerous scenarios, where these secondary resources are not 
taken into consideration (since capacity/resource calculation is not 
centralised).

> Support queue capacity in terms of absolute resources with custom 
> resourceType.
> -------------------------------------------------------------------------------
>
>                 Key: YARN-10503
>                 URL: https://issues.apache.org/jira/browse/YARN-10503
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Qi Zhu
>            Assignee: Qi Zhu
>            Priority: Critical
>         Attachments: YARN-10503.001.patch, YARN-10503.002.patch, 
> YARN-10503.003.patch, YARN-10503.004.patch
>
>
> Now the absolute resources are memory and cores.
> {code:java}
> /**
>  * Different resource types supported.
>  */
> public enum AbsoluteResourceType {
>   MEMORY, VCORES;
> }{code}
> But in our GPU production clusters, we need to support more resourceTypes.
> It's very import for cluster scaling when with different resourceType 
> absolute demands.
>  
> This Jira will handle GPU first.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to