[
https://issues.apache.org/jira/browse/FLINK-17714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17169695#comment-17169695
]
Zhu Zhu edited comment on FLINK-17714 at 8/3/20, 3:46 AM:
----------------------------------------------------------
One way in my mind is allow users to configure the factory class name in flink
conf and load it with user classloader.
Regarding the plugin mechanism, I have 2 concerns:
1. The plugin mechanism might be a bit too complex for users to operator for a
lightweight custom `RestartBackoffTimeStrategy`.
IIRC, the plugin mechanism requires the plugin jar to be put in {{/plugin}}
folder. (Correct me if I'm wrong here.)
2. A custom `RestartBackoffTimeStrategy` may rely on users' job code (e.g. to
recognize a specific error). However, job users may not be able to add job
specific plugins to flink client folder, in the case that the flink client is
shared for multiple users and controlled by a separate infra team.
was (Author: zhuzh):
One way in my mind is allow users to configure the factory class name in flink
conf and load it with user classloader.
The plugin mechanism might be a bit too complex for users to operator for a
lightweight custom `RestartBackoffTimeStrategy`.
IIRC, the plugin mechanism requires the plugin jar to be put in {{/plugin}}
folder. (Correct me if I'm wrong here.)
Besides that, a custom `RestartBackoffTimeStrategy` may rely on users' job code
(e.g. to recognize a specific error). However, job users may not be able to add
job specific plugins to flink client folder, in the case that the flink client
is shared for multiple users and controlled by a separate infra team.
> Support custom RestartBackoffTimeStrategy
> -----------------------------------------
>
> Key: FLINK-17714
> URL: https://issues.apache.org/jira/browse/FLINK-17714
> Project: Flink
> Issue Type: Wish
> Components: Runtime / Coordination
> Reporter: Zhu Zhu
> Priority: Major
>
> There are cases that users need to customize RestartBackoffTimeStrategy to
> better control job recovery.
> One example is that users want a job to restart only on certain errors and
> fail on others. See this ML
> [disscusion|https://lists.apache.org/thread.html/rde685552a83d0d146cf83560df1bc6f33d3dd569f69ae7bbcc4ae508%40%3Cuser.flink.apache.org%3E].
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
