[ 
https://issues.apache.org/jira/browse/YUNIKORN-2736?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17867028#comment-17867028
 ] 

Manikandan R commented on YUNIKORN-2736:
----------------------------------------

Two test cases covered in 
[https://github.com/apache/yunikorn-core/pull/911/files#diff-7b65cc904d1c0a0395b409e51db43bfe65238432eb96b66831c950060feac911.]
 
 # TestTryPreemption_VictimsUnavailable_InsufficientResource

Can be passed by setting the correct start and index in preemption mock plugin. 
It is really not a bug needs to be addressed.

2. TestTryPreemption_OnNode_GuaranteedSetOnPreemptorSide

Yes, it is an issue. We can provide a fix to address this issue. But fix can 
increase the complexity and make it further difficult for reasoning. More 
importantly, fix can lead to preemption storm or loop by which pods running in 
leaf queues preempt each other continuously without going anywhere causing 
instability to overall functioning of the queues. Filed a new 
[Jira](https://issues.apache.org/jira/browse/YUNIKORN-2761) to explain this 
problem and possible alternative solutions in usage doc. Reviews are almost 
done. Will merge the changes soon. Please read the doc for details.

Hence, closing this as won't fix.

 

 

> Preemption fails between two siblings even guaranteed set on one child and 
> parent
> ---------------------------------------------------------------------------------
>
>                 Key: YUNIKORN-2736
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-2736
>             Project: Apache YuniKorn
>          Issue Type: Sub-task
>          Components: core - scheduler
>            Reporter: Elad Dolev
>            Assignee: Manikandan R
>            Priority: Major
>
> Test set up:
> root.parent.parent1 guaranteed - vcores:10, usage - vcores:2
> root.parent.parent1.c1 usage - vcores:2
> root.parent.parent1.c2 guaranteed - vcores:1
> root.parent.parent2.c3
> Expected o/p: Ask is pending on root.parent.parent1.c2 waiting for resources, 
> expecting preemption to kick in to kill 1 pod running in c1. 
> Please refer 
> [https://github.com/apache/yunikorn-core/pull/911/files#diff-7b65cc904d1c0a0395b409e51db43bfe65238432eb96b66831c950060feac911R860]
>  for more details.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to