You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@yunikorn.apache.org by "Weiwei Yang (Jira)" <ji...@apache.org> on 2021/01/28 04:46:00 UTC
[jira] [Comment Edited] (YUNIKORN-520) Gang scheduling waits
indefinitely for placeholder pod allocation even where is no quota left in
the queue
[ https://issues.apache.org/jira/browse/YUNIKORN-520?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17273305#comment-17273305 ]
Weiwei Yang edited comment on YUNIKORN-520 at 1/28/21, 4:45 AM:
----------------------------------------------------------------
{quote}
I checked the shim code and I do not see it setting the field placeholderAsk in the handleSubmitApplicationEvent code as per the design document which means that the app is not rejected as it should since the core has no idea. The calculation of the placeholder size should be in that function or in the setTaskGroups function.
{quote}
That's correct, we can build that into the shim to pass placeholderAsk to the core. I believe the core side already has done the check.
was (Author: wwei):
{noformat}
I checked the shim code and I do not see it setting the field placeholderAsk in the handleSubmitApplicationEvent code as per the design document which means that the app is not rejected as it should since the core has no idea. The calculation of the placeholder size should be in that function or in the setTaskGroups function.
{noformat}
That's correct, we can build that into the shim to pass placeholderAsk to the core. I believe the core side already has done the check.
> Gang scheduling waits indefinitely for placeholder pod allocation even where is no quota left in the queue
> ----------------------------------------------------------------------------------------------------------
>
> Key: YUNIKORN-520
> URL: https://issues.apache.org/jira/browse/YUNIKORN-520
> Project: Apache YuniKorn
> Issue Type: Sub-task
> Components: shim - kubernetes
> Affects Versions: 0.10
> Reporter: Ayub Pathan
> Priority: Major
> Attachments: job.yaml, ns.yaml
>
>
> * Create a namespace([^ns.yaml] with quota.
> * Submit a job([^job.yaml] where the placeholder pods resource requests are more than queue quota.
> The YK waits indefinitely for the placeholder pods minMember limit to be satisfied. Gang scheduling can action to avoid this indefinite lock situation, incase of resource limits.
>
> {noformat}
> NAME↑ READY STATUS RS CPU MEM %CPU/R %MEM/R %CPU/L %MEM/L IP NODE QOS AGE │
> │ batch-sleep-job-4-hzc49 0/0 Pending 0 n/a n/a n/a n/a n/a n/a n/a n/a BU 4h28m │
> │ batch-sleep-job-4-jmqbf 0/0 Pending 0 n/a n/a n/a n/a n/a n/a n/a n/a BU 4h28m │
> │ batch-sleep-job-4-kg7hl 0/0 Pending 0 n/a n/a n/a n/a n/a n/a n/a n/a BU 4h28m │
> │ tg-tg1-batch-sleep-job-4-0 1/1 Running 0 n/a n/a n/a n/a n/a n/a 100.100.134.233 ip-10-192-136-201.ca-central-1.compute.internal BU 4h28m │
> │ tg-tg1-batch-sleep-job-4-1 1/1 Running 0 n/a n/a n/a n/a n/a n/a 100.100.87.75 ip-10-192-159-151.ca-central-1.compute.internal BU 4h28m │
> │ tg-tg1-batch-sleep-job-4-2 0/0 Pending 0 n/a n/a n/a n/a n/a n/a n/a n/a BU 4h28m │ {noformat}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@yunikorn.apache.org
For additional commands, e-mail: issues-help@yunikorn.apache.org