[ https://issues.apache.org/jira/browse/YARN-5255?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Arun Suresh updated YARN-5255:
------------------------------
    Description: 
YARN-4879 proposes to enhance the Allocate Request by introducing an id to explicitly identify a ResourceRequest.

Currently, if relaxLocality == true, a Node specific request is expanded to rack and *ANY* request by the {{AMRMClient}} before being sent to the Scheduler. The requires the 3 copies to perform locality specific delay scheduling.

It would be better to perform the expansion in the Scheduler itself, rather than the client, since:
# The expansion is not really specified in the ApplicationMasterProtocol, a non-java client will have to duplicate the expansion logic.
# Refactor out a lot of unnecessary code in the {{AMRMClientImpl}}

  was:
YARN-4879 proposes to enhance the Allocate Request by introducing a way to explicitly identify a ResourceRequest explicitly

Currently, if relaxLocality == true, a Node specific request is expanded to rack and *ANY* request by the {{AMRMClient}} before being sent to the Scheduler. The requires the 3 copies to perform locality specific delay scheduling.

It would be better to perform the expansion in the Scheduler itself, rather than the client, since:
# The expansion is not really specified in the ApplicationMasterProtocol, a non-java client will have to duplicate the expansion logic.
# Refactor out a lot of unnecessary code in the {{AMRMClientImpl}}


> Move ResourceRequest expansion for Delay Scheduling to the Scheduler
> --------------------------------------------------------------------
>
>                 Key: YARN-5255
>                 URL: https://issues.apache.org/jira/browse/YARN-5255
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Arun Suresh
>            Assignee: Arun Suresh
>
> YARN-4879 proposes to enhance the Allocate Request by introducing an id to explicitly identify a ResourceRequest.
> Currently, if relaxLocality == true, a Node specific request is expanded to rack and *ANY* request by the {{AMRMClient}} before being sent to the Scheduler. The requires the 3 copies to perform locality specific delay scheduling.
> It would be better to perform the expansion in the Scheduler itself, rather than the client, since:
> # The expansion is not really specified in the ApplicationMasterProtocol, a non-java client will have to duplicate the expansion logic.
> # Refactor out a lot of unnecessary code in the {{AMRMClientImpl}}



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

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