[
https://issues.apache.org/jira/browse/YARN-7402?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17711214#comment-17711214
]
Shilun Fan commented on YARN-7402:
----------------------------------
[~elgoiri] Thanks a lot for your quick reply, This feature creates a new module
hadoop-yarn-server-globalpolicygenerator, which is used to generate a global
resource view, which allows for allocation and preemption decisions to be based
on a global view of the cluster allocation and demand. The code of this feature
is in the YARN-7402 branch. I want to cherry-pick the developed part codes to
the trunk branch, and then continue to improve this feature on the trunk
branch. It should It is necessary to cherry-pick branch-7402 to the trunk.
> Federation V2: Global Optimizations
> -----------------------------------
>
> Key: YARN-7402
> URL: https://issues.apache.org/jira/browse/YARN-7402
> Project: Hadoop YARN
> Issue Type: New Feature
> Components: federation
> Reporter: Carlo Curino
> Assignee: Carlo Curino
> Priority: Major
>
> YARN Federation today requires manual configuration of queues within each
> sub-cluster, and each RM operates "in isolation". This has few issues:
> # Preemption is computed locally (and might far exceed the global need)
> # Jobs within a queue are forced to consume their resources "evenly" based on
> queue mapping
> This umbrella JIRA tracks a new feature that leverages the
> FederationStateStore as a synchronization mechanism among RMs, and allows for
> allocation and preemption decisions to be based on a (close to up-to-date)
> global view of the cluster allocation and demand. The JIRA also tracks
> algorithms to automatically generate policies for Router and AMRMProxy to
> shape the traffic to each sub-cluster, and general "maintenance" of the
> FederationStateStore.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
