[
https://issues.apache.org/jira/browse/FLINK-3970?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15300746#comment-15300746
]
Stephan Ewen commented on FLINK-3970:
-------------------------------------
I think the best way to completely isolate jobs against each other (CPU,
network, etc) is to run them in different processes.
The best way to do that is to start a flink cluster per job.
That is quite simple with YARN: Submitting the job in a per-job-cluster mode
brings up the Flink cluster on YARN, runs the job, and shuts the cluster down.
You can do something similar in other containerized environments.
Hope that solution works for you.
> How to deal with "resouce isolation" problem
> --------------------------------------------
>
> Key: FLINK-3970
> URL: https://issues.apache.org/jira/browse/FLINK-3970
> Project: Flink
> Issue Type: Wish
> Reporter: ZhengBowen
>
> For example, 'big query' and 'small query' are executed at the same time, you
> need isolate 'big query' and 'small query' to prevent 'big query' exhaust
> resouce(including i/o,mem,network) to make the 'small query' can complete
> quickly.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
