[
https://issues.apache.org/jira/browse/YUNIKORN-408?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17197724#comment-17197724
]
Adam Antal commented on YUNIKORN-408:
-------------------------------------
Hi [~vbm],
What you've experienced is not a bug, it is by design. Let me explain what you
see in detail.
- I myself have seen this message often in the logs, it is probably due to the
fact as [~wwei] mentioned. As Wangda is making a big change (merging the two
caches in the scheduler) this message will be obsolete after that commit.
- Container history a strictly increasing number. It increases by one when a
container is allocated, so 1 is the correct number. Also running containers
counts the actively running containers which is zero after the pod is deleted,
so both numbers looks good to me.
- In YUNIKORN-201, a team is working on a different approach to track
Application with K8s CRDs. Until that, there isn't really a way to tell
YuniKorn when an application is "finished". Therefore 1 running application
(even after the pod is finished) is also correct: YuniKorn is waiting for more
pods to spawn for that application. It considers the application running, after
the first pod has been spawned, and there isn't a way currently to tell if the
application has created and deleted all its pods.
We're doing our best to put the concepts into its place, but currently the
behaviour what you see is not a bug.
> Result mismatch between Container_Status and Container_History
> --------------------------------------------------------------
>
> Key: YUNIKORN-408
> URL: https://issues.apache.org/jira/browse/YUNIKORN-408
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: webapp
> Reporter: Vishwas
> Assignee: Adam Antal
> Priority: Major
> Attachments: appStatus_vs_contStatus.PNG, scheduler.log
>
>
> I have created a single pod to use the yunikorn scheduler and it gets
> allocated properly to the queue.
> When I delete the pod using the kubectl delete comand, I still see the
> container count as 1 in container_history but number of running container as
> 0 and also in application status I see it is running.
> I see below log in scheduler, not sure if it is related:
> {code:java}
> 2020-09-10T08:43:41.895Z DEBUG cache/context.go:241 failed to
> update pod in cache {"podName": "app-sleep-0", "error": "pod
> e9a531a8-e001-4cc9-bd3e-2d63852eacd7 is not added to scheduler cache, so
> cannot be updated"}{code}
> I have attached the snapshot of the UI and the scheduler.log from pod.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
