[
https://issues.apache.org/jira/browse/YUNIKORN-2372?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17815510#comment-17815510
]
Wilfred Spiegelenburg commented on YUNIKORN-2372:
-------------------------------------------------
To clarify: "maxResources" and "resourceUsage" are the maximum quota and usage
respectively when you compare it to a queue. We have the data to show there is
nothing we need to add.
I don't think we want to expose user and group quotas i.e.tracked usage as part
of the queue object. One user could be running jobs in multiple queues. They
might not be running anything. User and group quotas are centred on the user
and group with the queue structure per user or group.
User and group quotas should be accessible directly using a user or group name.
When a user or group is selected we display the queue structure from the root
with all details.
To display all users and groups details in the queue you would need to range
over all the user and group objects. Traverse to the queue level you need and
aggregate them together. That aggregation can then be displayed. That might
work nice for small numbers of queues, users and groups. When you run into the
100s of each it will not scale.
>From troubleshooting a pending pod perspective you would not start with the
>user or group but with the application or queue to find the outstanding
>requests. We do not track pending resources per user or group. Without the
>pending size you have nothing to base any analysis on..
An application belongs to one user or group. When you know the user you also
know if that application is tracked against a specific group. The order would
thus be:
# queue has pending resources
# display pending requests for the queue
# pending request links to an application
# application links to the user and group quotas
If you start at the application point, step 3, you have a link back to the
queue and a link forward to the user.
> Show user/group quota information on the UI
> -------------------------------------------
>
> Key: YUNIKORN-2372
> URL: https://issues.apache.org/jira/browse/YUNIKORN-2372
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler
> Reporter: Peter Bacsko
> Assignee: Dong-Lin Hsieh
> Priority: Major
>
> From a supportability point of view, it's very helpful to show user/group
> quotas on the UI. If a pod is unschedulable due to an insufficient quota,
> then it's immediately obvious why once the user looks at the quotas and sees
> a red bar for a given user/group.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
