[
https://issues.apache.org/jira/browse/FLINK-10790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16674971#comment-16674971
]
Piotr Nowojski commented on FLINK-10790:
----------------------------------------
Is there some difference between batch and streaming in this regard? I would
prefer to keep streaming and batch consistent between one another as much as
possible, so if we can simplify `ChannelSelector` for both of them, we should
do it.
> Refactor all the StreamPartitioner implementations into runtime module
> ----------------------------------------------------------------------
>
> Key: FLINK-10790
> URL: https://issues.apache.org/jira/browse/FLINK-10790
> Project: Flink
> Issue Type: Sub-task
> Components: Network
> Affects Versions: 1.8.0
> Reporter: zhijiang
> Assignee: zhijiang
> Priority: Minor
>
> The motivation is for refactoring {{ChannelSelector}} interface mentioned in
> [FLINK-10662|https://issues.apache.org/jira/browse/FLINK-10662].
> The {{ChannelSelector}} implementations are separate for stream and batch
> currently, resulting in many duplicated logics for selecting channels. The
> root cause is that they are locating in different modules to not reuse the
> existing ones.
> In general, the core runtime codes should be unified for both stream and
> batch jobs as much as possible, then we can reduce the work overhead to
> maintain the duplicated codes.
> This jira would migrate the existing partitioner package from
> {{flink-streaming-jave}} to {{flink-runtime}}. Next we want to de-duplicate
> implementations for {{RoundRobinChannelSelector}} and {{OutputEmitter}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
