[ 
https://issues.apache.org/jira/browse/FLINK-2802?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14941348#comment-14941348
 ] 

ASF GitHub Bot commented on FLINK-2802:
---------------------------------------

GitHub user gyfora opened a pull request:

    https://github.com/apache/flink/pull/1216

    [FLINK-2802] [streaming] Remove cyclic watermark dependencies for iterations

    This PR contains a simple change so that if timestamps are enabled, 
iteration sources automatically emit a `Long.MAX_VALUE` watermark.
    
    While this will not help making event/processing time windows consistent 
across cycles (which is inherently problematic due to the cyclic time 
dependency), at least these programs will not deadlock.
    
    I still need to add a test to validate the behaviour.

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/gyfora/flink watermark

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/1216.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 #1216
    
----
commit 3671742d99fe9afdb71a00ed16f5ec47f099b0bb
Author: Gyula Fora <[email protected]>
Date:   2015-10-02T16:26:32Z

    [FLINK-2802] [streaming] Remove cyclic watermark dependencies for iterations

----


> Watermark triggered operators cannot progress with cyclic flows
> ---------------------------------------------------------------
>
>                 Key: FLINK-2802
>                 URL: https://issues.apache.org/jira/browse/FLINK-2802
>             Project: Flink
>          Issue Type: Bug
>          Components: Streaming
>            Reporter: Gyula Fora
>            Priority: Blocker
>
> The problem is that we can easily create a cyclic watermark (time) dependency 
> in the stream graph which will result in a deadlock for watermark triggered 
> operators such as  the `WindowOperator`.
> A solution to this could be to emit a Long.MAX_VALUE watermark from the 
> iteration sources.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to