You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@yunikorn.apache.org by "Peter Bacsko (Jira)" <ji...@apache.org> on 2024/03/21 14:00:00 UTC
[jira] [Updated] (YUNIKORN-2510) Placeholder processing starts immediately despite maxApplications limit

     [ https://issues.apache.org/jira/browse/YUNIKORN-2510?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Peter Bacsko updated YUNIKORN-2510:
-----------------------------------
    Description: 
If the maxApplication limit is hit inside {{Application.tryAllocate()}}, we still create the placeholder pods in the shim. This has two consequences:

-1. Although they stay in Pending state, these pods still consume resources in the cluster, because they are tracked by the API server.-
2. Placeholder timer is started immediately. If the timeout expires, it affects the application, even if it's not in a runnable state.

#2 is more problematic and needs to be addressed. #1 is up to discussion.

EDIT: #1 doesn't seem to be a problem. After further reviewing the code, we only start the timer after the first allocation. So looks like we need to decide whether immediate creation of PH pods is OK or not.



  was:
If the maxApplication limit is hit inside {{Application.tryAllocate()}}, we still create the placeholder pods in the shim. This has two consequences:

1. Although they stay in Pending state, these pods still consume resources in the cluster, because they are tracked by the API server. 
2. Placeholder timer is started immediately. If the timeout expires, it affects the application, even if it's not in a runnable state.

#2 is more problematic and needs to be addressed. #1 is up to discussion.




> Placeholder processing starts immediately despite maxApplications limit
> -----------------------------------------------------------------------
>
>                 Key: YUNIKORN-2510
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-2510
>             Project: Apache YuniKorn
>          Issue Type: Bug
>          Components: core - scheduler
>            Reporter: Peter Bacsko
>            Assignee: Peter Bacsko
>            Priority: Major
>
> If the maxApplication limit is hit inside {{Application.tryAllocate()}}, we still create the placeholder pods in the shim. This has two consequences:
> -1. Although they stay in Pending state, these pods still consume resources in the cluster, because they are tracked by the API server.-
> 2. Placeholder timer is started immediately. If the timeout expires, it affects the application, even if it's not in a runnable state.
> #2 is more problematic and needs to be addressed. #1 is up to discussion.
> EDIT: #1 doesn't seem to be a problem. After further reviewing the code, we only start the timer after the first allocation. So looks like we need to decide whether immediate creation of PH pods is OK or not.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@yunikorn.apache.org
For additional commands, e-mail: issues-help@yunikorn.apache.org