[
https://issues.apache.org/jira/browse/YUNIKORN-668?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17343686#comment-17343686
]
Chaoran Yu edited comment on YUNIKORN-668 at 5/13/21, 3:56 AM:
---------------------------------------------------------------
Are you saying that you want to achieve high availability of the scheduler?
There's not really a need to run two replicas. You can just run one replica and
reply on the K8s Deployment to restart the pod in case it fails. The scheduler
is a stateless process, so there's nothing to persist and recover (except that
the newly started pod will scan and recover existing allocations in the cluster)
was (Author: yuchaoran2011):
Are you saying that you want to achieve high availability of the scheduler?
There's not really a need to run two replicas. You can just run one replica and
reply on the K8s Deployment to restart the pod in case it fails. The scheduler
is a stateless process, so there's no thing to persist and recover (except that
the newly started pod will scan and recover existing allocations in the cluster)
> There is no fail over mechanism in shim project.
> ------------------------------------------------
>
> Key: YUNIKORN-668
> URL: https://issues.apache.org/jira/browse/YUNIKORN-668
> Project: Apache YuniKorn
> Issue Type: Bug
> Reporter: MichealShin
> Priority: Critical
>
> We want to have the fail over feature in the shim project, but can not work
> when just set the replication to 2 in k8s. The two pods just cannot work
> together . When pod 1 get the config, the pod 2 will not. To make it worse,
> the working pod may not work after some time, and the other will get the
> right config.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
