[
https://issues.apache.org/jira/browse/YARN-2022?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14044270#comment-14044270
]
Wangda Tan commented on YARN-2022:
----------------------------------
Hi [~mayank_bansal],
For existing CapacityScheduler implementation, each user can use at most
getMaxAMResourcePerQueuePercent() * getAbsoluteCapacity(). In your case, each
user can use 10% * 10% = 1% cluster resource. So in extreme case, if the queue
has 10 users, each user have 1 application, after preemption, all resource will
be used by AM.
I'll create a separated JIRA to discuss this. What [~sunilg] has done should be
consistent with existing CapacityScheduler behavior. Is this make sense to you?
Thanks,
Wangda
> Preempting an Application Master container can be kept as least priority when
> multiple applications are marked for preemption by
> ProportionalCapacityPreemptionPolicy
> ---------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-2022
> URL: https://issues.apache.org/jira/browse/YARN-2022
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Affects Versions: 2.4.0
> Reporter: Sunil G
> Assignee: Sunil G
> Attachments: YARN-2022-DesignDraft.docx, YARN-2022.2.patch,
> YARN-2022.3.patch, YARN-2022.4.patch, YARN-2022.5.patch, YARN-2022.6.patch,
> YARN-2022.7.patch, Yarn-2022.1.patch
>
>
> Cluster Size = 16GB [2NM's]
> Queue A Capacity = 50%
> Queue B Capacity = 50%
> Consider there are 3 applications running in Queue A which has taken the full
> cluster capacity.
> J1 = 2GB AM + 1GB * 4 Maps
> J2 = 2GB AM + 1GB * 4 Maps
> J3 = 2GB AM + 1GB * 2 Maps
> Another Job J4 is submitted in Queue B [J4 needs a 2GB AM + 1GB * 2 Maps ].
> Currently in this scenario, Jobs J3 will get killed including its AM.
> It is better if AM can be given least priority among multiple applications.
> In this same scenario, map tasks from J3 and J2 can be preempted.
> Later when cluster is free, maps can be allocated to these Jobs.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
