[
https://issues.apache.org/jira/browse/YARN-5035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15269318#comment-15269318
]
Karthik Kambatla commented on YARN-5035:
----------------------------------------
bq. I was thinking of integrating YARN-4056 with the fair scheduler. My initial
thought process was to replace the assignMultiple with the Bundler.
Sounds like a good long-term approach. Until we gain confidence, I would like
for the bundler to be opt in. Would like to get this in sooner to help with
more immediate concerns.
bq. But if continuous-scheduling is turned on, we can always set maxAssign=1 to
get a balanced workload distribution, right?
Yes. We have seen some issues with continuous-scheduling on large clusters -
the continuous-scheduling thread hogs the CPU rendering other functionality
slow/unresponsive.
> FairScheduler: Adjust maxAssign dynamically when assignMultiple is turned on
> ----------------------------------------------------------------------------
>
> Key: YARN-5035
> URL: https://issues.apache.org/jira/browse/YARN-5035
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: fairscheduler
> Affects Versions: 2.8.0
> Reporter: Karthik Kambatla
> Assignee: Karthik Kambatla
>
> When assignMultiple is turned on, today the number of containers allocated to
> node within a single heartbeat is determined by maxAssign. Picking a higher
> value for this prevents the spreading of the workload, and a smaller number
> leads to slow ramp up.
> The proposal is to make it dynamic - half of the unallocated resources on the
> node.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]