[ 
https://issues.apache.org/jira/browse/YUNIKORN-941?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17452463#comment-17452463
 ] 

Kinga Marton commented on YUNIKORN-941:
---------------------------------------

Some note on the newly created charts:
 * as [~yuchaoran2011] suggested, we will use a subchart for the admission 
controller. 
 * since during a YK upgrade we want to make sure that no pods will be handled 
by the default scheduler during the YK downtime, it is essentially to have the 
admission controlller running when the scheduler will be upgraded. By using 
seubcharts this is possible with the following steps:
 ** update the admission controller (helm upgrade will do the upgrade only if 
there are some chnges, so for this steps we need to make sure that there are no 
changes in the scheduler)
 ** after the admission controller is updated, we can update the helm 
deployment again and include the scheduler changes as well. Since helm will 
detect the admission controller it is already up to date, it won't touch it.
 * We need to do the upgrade in this two steps, because during a normal upgrade 
helm aggregates all the manifests into one and then it will sort them according 
to their type and alphabetically, but will not wait for the dependeies being 
installed first. See more details in the following Helm documentation: 
[https://helm.sh/docs/topics/charts/#operational-aspects-of-using-dependencies]

> split scheduler and admission controller deployment
> ---------------------------------------------------
>
>                 Key: YUNIKORN-941
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-941
>             Project: Apache YuniKorn
>          Issue Type: Improvement
>          Components: shim - kubernetes
>            Reporter: Kinga Marton
>            Assignee: Kinga Marton
>            Priority: Major
>              Labels: pull-request-available
>
> To support proper YuniKorn upgrades and restarts we should move the admission 
> controller out of the scheduler deployment and make it a separate deployment.
> This could also allow the admission controller to be made high available and 
> allow simpler no down time upgrades possible. 



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

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

Reply via email to