[ 
https://issues.apache.org/jira/browse/FLINK-22134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17320042#comment-17320042
 ] 

Till Rohrmann commented on FLINK-22134:
---------------------------------------

Thanks for testing this feature [~xintongsong]. Your feedback is very valuable.

B2: I like the idea of calling {{Executing.notifyNewResourcesAvailable}} or 
{{Executing.Context.canScaleUp}} only if the resources have stabilized. That 
way, it should be possible to immediately restart the job after cancellation if 
the resources have not changed in the meantime. What this means for the 
implementation is that the {{AdaptiveScheduler}} needs to be responsible for 
the stabilization timeout.

UI2: I see the point that this can be confusing. Technically, a rescale 
operation is currently a restart with a potentially modified parallelism. I am 
also not sure whether the distinction is actually possible to do because for 
scaling down, we rely on a proper failover. Hence, all explicit scale down 
events won't be distinguishable from a true failure. 

> Test the reactive mode
> ----------------------
>
>                 Key: FLINK-22134
>                 URL: https://issues.apache.org/jira/browse/FLINK-22134
>             Project: Flink
>          Issue Type: Test
>          Components: Runtime / Coordination
>    Affects Versions: 1.13.0
>            Reporter: Till Rohrmann
>            Assignee: Xintong Song
>            Priority: Blocker
>              Labels: release-testing
>             Fix For: 1.13.0
>
>         Attachments: flink-xtsong-standalonejob-1-xtsong-mac.log, 
> 截屏2021-04-12 上午10.30.25.png, 截屏2021-04-12 上午10.32.50.png
>
>
> The newly introduced reactive mode (FLINK-10407) allows Flink to make use of 
> newly arriving resources while the job is running. The feature documentation 
> with the current set of limitations can be found here [1].
> In order to test this new feature I recommend to follow the documentation and 
> to try it out wrt the stated limitations. Everything which is not explicitly 
> contained in the set of limitations should work.
> [1] 
> https://ci.apache.org/projects/flink/flink-docs-master/docs/deployment/elastic_scaling/



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to