[
https://issues.apache.org/jira/browse/YARN-7402?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16221498#comment-16221498
]
Carlo Curino commented on YARN-7402:
------------------------------------
This started with conversations with Bill Ramsey, [~roniburd], [~subru],
[~asuresh], [~kkaranasos] and [~chris.douglas].
The goal is to extend YARN ability to enforce global invariant across a
federated cluster, while retaining the scalability of
federation. For this purpose the sharing of information among sub-cluster is on
heartbeats and limited to very summarized
view of the world (queue-level aggregates only).
> Federation: Global Queues
> -------------------------
>
> 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
>
> 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.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
