[
https://issues.apache.org/jira/browse/FLINK-5473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15831939#comment-15831939
]
ASF GitHub Bot commented on FLINK-5473:
---------------------------------------
GitHub user StefanRRichter opened a pull request:
https://github.com/apache/flink/pull/3182
[FLINK-5473] Limit MaxParallelism to 1 for non-parallel operators and
improve choice of max parallelism without explicit configuration
This PR limits the maximum parallelism for non-parallel operator to 1.
Furthermore, this improves the default behaviour if the user did not
explicitly specify a maximum parallelism. In particular, maximum parallelism
can now be derived from savepoints, allowing users that migrate from Flink 1.1
to Flink 1.2 to keep their job unchanged.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/StefanRRichter/flink recover-max-para
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/3182.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 #3182
----
commit 20d0a3fc88c85a71b692c5408fc7b2fd33da8ff2
Author: Stefan Richter <s.rich...@data-artisans.com>
Date: 2017-01-16T13:31:22Z
[FLINK-5473] Limit max parallelism to 1 for non-parallel operators
commit f6081f319b7f2ef8615557743d906bc4585445f7
Author: Stefan Richter <s.rich...@data-artisans.com>
Date: 2017-01-16T17:41:37Z
[FLINK-5473] Better default behaviours for unspecified maximum parallelism
----
> setMaxParallelism() higher than 1 is possible on non-parallel operators
> -----------------------------------------------------------------------
>
> Key: FLINK-5473
> URL: https://issues.apache.org/jira/browse/FLINK-5473
> Project: Flink
> Issue Type: Bug
> Components: DataStream API
> Affects Versions: 1.2.0
> Reporter: Robert Metzger
> Assignee: Stefan Richter
>
> While trying out Flink 1.2, I found out that you can set a maxParallelism
> higher than 1 on a non-parallel operator.
> I think we should have the same semantics as the setParallelism() method.
> Also, when setting a global maxParallelism in the execution environment, it
> will be set as a default value for the non-parallel operator.
> When restoring a savepoint from 1.1, you have to set the maxParallelism to
> the parallelism of the 1.1 job. Non-parallel operators will then also get the
> maxPar set to this value, leading to an error on restore.
> So currently, users restoring from 1.1 to 1.2 have to manually set the
> maxParallelism to 1 for all non-parallel operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
