[
https://issues.apache.org/jira/browse/FLINK-1505?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14329028#comment-14329028
]
ASF GitHub Bot commented on FLINK-1505:
---------------------------------------
GitHub user uce opened a pull request:
https://github.com/apache/flink/pull/428
[FLINK-1505] Separate reader API from result consumption
@gyfora, can you please rebase on this branch and verify that everything is
still working as expected for you?
This PR separates the reader API (record and buffer readers) from result
consumption (input gate). The buffer reader was a huge component with mixed
responsibilities both as the runtime component to set up input channels for
intermediate result consumption and as a lower-level user API to consume
buffers/events.
The separation makes it easier for users of the API (e.g. flink-streaming)
to extend the handling of low-level buffers and events. Gyula's initial
feedback confirmed this.
In view of FLINK-1568, this PR makes it also easier to test the result
consumption logic for failure scenarios.
I will rebase #356 on this changes.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/uce/incubator-flink flink-1505-input_gate
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/428.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 #428
----
commit db1dc5be12427664a418ce6e4fb41de39838fac0
Author: Ufuk Celebi <[email protected]>
Date: 2015-02-10T14:05:44Z
[FLINK-1505] [distributed runtime] Separate reader API from result
consumption
----
> Separate buffer reader and channel consumption logic
> ----------------------------------------------------
>
> Key: FLINK-1505
> URL: https://issues.apache.org/jira/browse/FLINK-1505
> Project: Flink
> Issue Type: Improvement
> Components: Distributed Runtime
> Affects Versions: master
> Reporter: Ufuk Celebi
> Priority: Minor
>
> Currently, the hierarchy of readers (f.a.o.runtime.io.network.api) is
> bloated. There is no separation between consumption of the input channels and
> the buffer readers.
> This was not the case up until release-0.8 and has been introduced by me with
> intermediate results. I think this was a mistake and we should seperate this
> again. flink-streaming is currently the heaviest user of these lower level
> APIs and I have received feedback from [~gyfora] to undo this as well.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
