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 2022/11/10 02:14:00 UTC

[jira] [Commented] (YUNIKORN-1389) Pass node labels from shim to core.

    [ https://issues.apache.org/jira/browse/YUNIKORN-1389?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17631388#comment-17631388 ] 

Wilfred Spiegelenburg commented on YUNIKORN-1389:
-------------------------------------------------

We are trying to stay away from pulling all that detail into the scheduler. It is not used during the scheduling cycle. During scheduling we use the K8s predicates to select nodes. None of these details are used as part of the scheduling cycle as they are not stable. Most of them depend on the cloud provider used or the administrator adding labels.

How do you propose to use the labels and make them usable in the UI? Without a design on the UI side just passing the labels from the k8shim to the core leaves things hanging. Should we not look at the UI side first and then decide on what we do for propagation of the labels?

There is also special casing for rack and name which is a left over from YARN like days. That special casing does not fit in with this change it feels like.

> Pass node labels from shim to core.
> -----------------------------------
>
>                 Key: YUNIKORN-1389
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-1389
>             Project: Apache YuniKorn
>          Issue Type: Task
>          Components: shim - kubernetes
>            Reporter: Qi Zhu
>            Assignee: Qi Zhu
>            Priority: Major
>              Labels: pull-request-available
>
> Suggested by YUNIKORN-1362:
> The YuniKorn UI shows the list of all the nodes in the cluster. There is no way of filtering the list.
> We have only certain nodes that we want our jobs to be scheduled in the cluster and that is dictated by the nodeSelector. I would good to have a way to show only those nodes on the UI to give a better idea of the resources that we have for the jobs to be scheduled 
> The functionality will depend on the details we pull for the nodes into the core. Propagating some info for a node from the shim into the core might be needed.
> Here we first to pass node labels from shim to core.



--
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