You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@beam.apache.org by "Jan Lukavský (Jira)" <ji...@apache.org> on 2020/05/05 14:59:00 UTC
[jira] [Commented] (BEAM-9888) @RequiresTimeSortedInput might feed
data out of order
[ https://issues.apache.org/jira/browse/BEAM-9888?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17099982#comment-17099982 ]
Jan Lukavský commented on BEAM-9888:
------------------------------------
At the end, dropping data based on input watermark is safer and way easier to implement. Moreover, the difference will be pretty much negligible.
> @RequiresTimeSortedInput might feed data out of order
> -----------------------------------------------------
>
> Key: BEAM-9888
> URL: https://issues.apache.org/jira/browse/BEAM-9888
> Project: Beam
> Issue Type: Bug
> Components: runner-core
> Affects Versions: 2.20.0
> Reporter: Jan Lukavský
> Assignee: Jan Lukavský
> Priority: Major
> Fix For: 2.21.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> StatefulDoFnRunner drops late data based on output watermark, but output watermark might not advance fast enough to discard data that should have been dropped. We have to either:
> a) drop data based on input watermark, or
> b) drop data based on timestamp of last emitted element
> I'm a little in favor of b), because it can drop less data.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)