You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "ASF GitHub Bot (JIRA)" <ji...@apache.org> on 2016/01/18 20:05:39 UTC
[jira] [Commented] (FLINK-3255) Chaining behavior should not depend
on parallelism
[ https://issues.apache.org/jira/browse/FLINK-3255?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15105675#comment-15105675 ]
ASF GitHub Bot commented on FLINK-3255:
---------------------------------------
GitHub user StephanEwen opened a pull request:
https://github.com/apache/flink/pull/1518
[FLINK-3255] [streaming] Disable parallelism-dependent chaining optimization
This pull request disables the condition to chain shuffles/broadcasts/etc if the consumer parallelism is one.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/StephanEwen/incubator-flink dop1_chaining
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/1518.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #1518
----
commit 4bdb03173e398c8dbcf9fe68bd05f0d325cb0b69
Author: Stephan Ewen <se...@apache.org>
Date: 2016-01-18T17:27:53Z
[FLINK-3255] [streaming] Disable parallelism-dependent chaining optimization
----
> Chaining behavior should not depend on parallelism
> --------------------------------------------------
>
> Key: FLINK-3255
> URL: https://issues.apache.org/jira/browse/FLINK-3255
> Project: Flink
> Issue Type: Bug
> Components: Streaming
> Affects Versions: 1.0.0
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
> Fix For: 0.10.1
>
>
> Currently, operators are chained more aggressively when the parallelism is one. That makes debugging tougher as it changes threading behavior.
> The benefits are also limited: Real installations where that type of efficiency would be needed would not run in parallelism 1, or would not use a partitioning/broadcast step there (if explicitly required to run parallelism 1).
> In the future, when we want to allow parallelism to be adjusted dynamically, this will be even more tricky.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)