[
https://issues.apache.org/jira/browse/FLINK-35896?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17868639#comment-17868639
]
yuanfenghu commented on FLINK-35896:
------------------------------------
> RescaleApiScalingRealizer#realizeParallelismOverrides [1] will catch the
>Exception, and only log it. So does your approach1 remove this catch, right?
There will be log printing, but I feel it would be better to record it in the
event. Obviously we record the event when we succeed, so should we also record
it when an exception occurs? Obvious, the Scaling action it is relatively low
frequency. It looks good to log it into an event
> Also, may I know what exception happens in
>ScalingRealizer.realizeParallelismOverrides?
Yes, our environment encapsulates the flink rest api, and there are some
internal container call errors. This is different from the community. However,
I think if the jobmanager of flink is directly called in the rest method, there
may be exceptions, such as timeout.
> eventHandler in RescaleApiScalingRealizer handle the event of scaling failure
> ------------------------------------------------------------------------------
>
> Key: FLINK-35896
> URL: https://issues.apache.org/jira/browse/FLINK-35896
> Project: Flink
> Issue Type: Improvement
> Components: Autoscaler
> Affects Versions: 2.0.0
> Reporter: yuanfenghu
> Priority: Minor
>
> When using flink-autoscaler-standalone, if fail during the Scaling process,
> need to handle the failure event through evenhandler
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
