[ https://issues.apache.org/jira/browse/BEAM-9573?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17065015#comment-17065015 ]
Maximilian Michels commented on BEAM-9573: ------------------------------------------ I understand that this may mean additional work for you as a release manager. I'm sorry but we can't release with a bug like this. Is this a regression from a previous release? Yes. Is this a new feature or related to a new feature? It's a new feature. Timer output timestamps. What percentage of users would be impacted by this issue if it is not fixed? 100% of the users. Would it be possible for the impacted users to skip this version? I can't answer this question. > Watermark hold for timer output timestamp is not computed correctly > ------------------------------------------------------------------- > > Key: BEAM-9573 > URL: https://issues.apache.org/jira/browse/BEAM-9573 > Project: Beam > Issue Type: Bug > Components: runner-flink > Affects Versions: 2.20.0 > Reporter: Maximilian Michels > Assignee: Maximilian Michels > Priority: Blocker > Fix For: 2.20.0 > > > With the introduction of timer output timestamp, a new watermark hold had > been added to the Flink Runner. The watermark computation works on the keyed > state backend which computes a key-scoped watermark hold and not the desired > operator-wide watermark hold. > Computation: > https://github.com/apache/beam/blob/b564239081e9351c56fb0e7d263495b95dd3f8f3/runners/flink/src/main/java/org/apache/beam/runners/flink/translation/wrappers/streaming/DoFnOperator.java#L1140 > Key-scoped state: > https://github.com/apache/beam/blob/b564239081e9351c56fb0e7d263495b95dd3f8f3/runners/flink/src/main/java/org/apache/beam/runners/flink/translation/wrappers/streaming/DoFnOperator.java#L1130 > We need to change this to operate on all keys. This has to be done before > fixing BEAM-9566. -- This message was sent by Atlassian Jira (v8.3.4#803005)