[
https://issues.apache.org/jira/browse/FLINK-30920?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17685052#comment-17685052
]
Gaurav Miglani commented on FLINK-30920:
----------------------------------------
yes you are right, but in our cases, load came suddenly, for which we need sink
to be of define parallelism, although I have found out property
`kubernetes.operator.job.autoscaler.vertex.min-parallelism` which can be set,
then autoscaler will not trigger downscale below this threshold, but it seems
to me as we are scaling vertices and we already have list of vertices, why not
user can control which vertices he didn't want autoscaler to perform any
action, your thoughts ?
> K8 cluster autoscaler | exclude operator ids from scaler
> --------------------------------------------------------
>
> Key: FLINK-30920
> URL: https://issues.apache.org/jira/browse/FLINK-30920
> Project: Flink
> Issue Type: Improvement
> Components: Kubernetes Operator
> Reporter: Gaurav Miglani
> Priority: Major
> Attachments: sample-logs
>
>
> Sometime in cases of sink operator ids, where logic is heavy group
> aggregation and scan mode is earliest, flink k8 operator tries to
> scale/downscale sink operator ids as well, there should be a way where user
> can give list of operator ids/vertices where cluster autoscaler doesn't
> perform any scaling action on a configurable list of operator ids/vertices
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
