[
https://issues.apache.org/jira/browse/YUNIKORN-1990?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17765739#comment-17765739
]
Rainie Li commented on YUNIKORN-1990:
-------------------------------------
Hi [~ccondit], I have some questions about this part (Line 201~ Line 210)
[https://github.com/apache/yunikorn-core/blob/5f509ff8fb903c6e883211782f2f30c04a31ea50/pkg/scheduler/objects/preemption.go#L201]
Is currentQueue the same queue as snapshot?
* If currentQueue = snapshot: why need Line 201 an additional loop to find
snapshot? since Line 198 already has currentQueue?
* If currentQueue != snapshot: with current logic, currentQueue will never
get updated since Line 203 just updated snapshot, why check currentQueue in
Line 204 without actually updating it?
Can you provide some info here?
Thanks!
> Intra-queue preemption happens
> ------------------------------
>
> Key: YUNIKORN-1990
> URL: https://issues.apache.org/jira/browse/YUNIKORN-1990
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - scheduler
> Reporter: Rainie Li
> Assignee: Rainie Li
> Priority: Major
> Attachments: Screenshot 2023-09-15 at 10.07.46 AM.png, Screenshot
> 2023-09-15 at 10.08.02 AM.png
>
>
> *Background*
> We noticed preemption happens for jobs within same queue.
>
> *Test we have done*
> * Submitted job1 to root.test-queue.tier1 queue, queue allocated usage went
> higher than guaranteed resource.
> * Then submitted another Job2 to same queue, queue allocated usage decreased
> to lower than guaranteed resource.
> * Most executors of Job1 got preempted by Job2
> !Screenshot 2023-09-15 at 10.07.46 AM.png!
> !Screenshot 2023-09-15 at 10.08.02 AM.png!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
