[
https://issues.apache.org/jira/browse/FLINK-37351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18054450#comment-18054450
]
Rion Williams commented on FLINK-37351:
---------------------------------------
[~arvid]
I wouldn't mind taking a stab at implementing this. I'm assuming that we'd
likely detect the type of topology (i.e. pre-commit or not) and based on that
handle it accordingly when we assign the co-location group key?
I'm sure there are other approaches (e.g., configuration, etc.), but this one
seems like it would leverage some of the existing grouping infrastructure
that's already in place. Any initial thoughts or preferences as you're likely
far more familiar with this than I would be.
> Ensure writer and committer are colocated
> -----------------------------------------
>
> Key: FLINK-37351
> URL: https://issues.apache.org/jira/browse/FLINK-37351
> Project: Flink
> Issue Type: Bug
> Components: API / Core
> Affects Versions: 2.1.0
> Reporter: Arvid Heise
> Priority: Major
>
> Writer and committer should always be co-located in sinks that do not have a
> pre-commit topology to improve performance and potentially allow backchannels
> (in the style of statefun) to work.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
