[
https://issues.apache.org/jira/browse/YARN-6486?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16005704#comment-16005704
]
Karthik Kambatla commented on YARN-6486:
----------------------------------------
The non-HB driven approach allows certain optimizations. However, in our
experience, the current implementation doesn't scale well. Current
implementation continuously simulates node heartbeats and exposes lock
contention issues in the scheduler. We have been recommending turning this off
for large clusters.
Global scheduling is likely the better approach that is not driven by
heartbeats.
I am open to less-lock-contention approaches.
> FairScheduler: Deprecate continuous scheduling in 2.9
> -----------------------------------------------------
>
> Key: YARN-6486
> URL: https://issues.apache.org/jira/browse/YARN-6486
> Project: Hadoop YARN
> Issue Type: Task
> Components: fairscheduler
> Affects Versions: 2.9.0
> Reporter: Wilfred Spiegelenburg
> Assignee: Wilfred Spiegelenburg
>
> Mark continuous scheduling as deprecated in 2.9 and remove the code in 3.0.
> Removing continuous scheduling from the code will be logged as a separate jira
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
