You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@yunikorn.apache.org by "Wilfred Spiegelenburg (Jira)" <ji...@apache.org> on 2022/02/22 23:47:00 UTC
[jira] [Commented] (YUNIKORN-1076) Robust handling of invalid Task Group annotation
[ https://issues.apache.org/jira/browse/YUNIKORN-1076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17496389#comment-17496389 ]
Wilfred Spiegelenburg commented on YUNIKORN-1076:
-------------------------------------------------
Based on the comments in the PR we limit this Jira to implementing just the event creation.
Rejecting an application is a big code change and could cause a bad user experience.
> Robust handling of invalid Task Group annotation
> ------------------------------------------------
>
> Key: YUNIKORN-1076
> URL: https://issues.apache.org/jira/browse/YUNIKORN-1076
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: shim - kubernetes
> Reporter: Si Latt
> Assignee: ted
> Priority: Major
> Labels: newbie, pull-request-available
>
> For gang scheduling, task group information has to be defined in the annotation section. If the provided YAML for task group info is invalid, such as missing double quote for keys, it results in parse exception and gets logged in YK log. However, when looking at Kubernetes event log, there is no indication that exception happened during gang scheduling. Other gang scheduling events are logged and hence give users the wrong impression that pods are gang scheduled without any issue.
> Current behavior is dangerous as it cause gang scheduling to not work in production without users realizing any issue. We should probably take the following actions:
> # Reject / fail any app with invalid task group annotation
> # Emit exceptions in kubernetes events for surfacing critical events happening in the cluster.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@yunikorn.apache.org
For additional commands, e-mail: issues-help@yunikorn.apache.org