[
https://issues.apache.org/jira/browse/YUNIKORN-2977?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17898683#comment-17898683
]
Peter Bacsko edited comment on YUNIKORN-2977 at 11/15/24 4:41 PM:
------------------------------------------------------------------
Test case to repro #3:
https://github.com/pbacsko/incubator-yunikorn-k8shim/commit/928c504e94b04af395e68a00994bbe96996d79c1
Note that behavior depends on line 546 as indicated by a comment (I found this
out by accident). If the {{NodeSelectorRequirement.Operator}} is removed, then
the DS pod will never be allocated despite making room for it with preemption.
was (Author: pbacsko):
Test case to repro #3:
https://github.com/pbacsko/incubator-yunikorn-k8shim/commit/928c504e94b04af395e68a00994bbe96996d79c1
Note that behavior depends on line 546 as indicated by a comment (I found this
out by accident). If the operator is removed, then the DS pod will never be
allocated despite making room for it with preemption.
> [Umbrell] DaemonSet preemption hardening
> ----------------------------------------
>
> Key: YUNIKORN-2977
> URL: https://issues.apache.org/jira/browse/YUNIKORN-2977
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler
> Reporter: Peter Bacsko
> Priority: Major
>
> We identified a couple of issues with DaemonSet preemption. The current
> implementation it's not stable.
> Notable issues:
> 1. Flooding the logs from {{tryAllocate()}} because unreservation fails
> 2. Flooding the logs if there are no victims found
> 3. Allocation is stuck if the DS pod cannot run on the target node due to
> predicate errors.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
