[
https://issues.apache.org/jira/browse/FLINK-32496?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17738922#comment-17738922
]
Rui Fan commented on FLINK-32496:
---------------------------------
Thanks [~haishui] creating this JIRA.
Hi [~martijnvisser] , after analysis, I think these 2 tickets are not duplicate.
This ticket is a new bug, and it is the root cause of this mail list :
[https://lists.apache.org/thread/znbhls5cdw2o1jl7gl1t1nvsqkgjps5r]
After my test and analysis, I found the bug is when one source is always active
and other sources are idle(Either there is no data, or alignment causes idle.),
these source won't resume.
> Sources with idleness and alignment always wait for alignment when part of
> multiple sources is idle
> ---------------------------------------------------------------------------------------------------
>
> Key: FLINK-32496
> URL: https://issues.apache.org/jira/browse/FLINK-32496
> Project: Flink
> Issue Type: Bug
> Components: API / DataStream
> Affects Versions: 1.16.2, 1.17.1
> Reporter: haishui
> Priority: Major
>
> Sources with idleness and alignment always wait for alignment when part of
> multiple sources is idle.
> *Root cause:*
> In
> [SourceOperator|https://github.com/apache/flink/blob/master/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/operators/SourceOperator.java],
> `lastEmittedWatermark` is Long.MAX_VALUE if a source is idle.
> When other source is active, the `currentMaxDesiredWatermark` is less then
> Long.MAX_VALUE.
> So the `shouldWaitForAlignment` method is always true for idle sources.
>
> What's more, the source will become idle if a source wait for alignment for a
> long time, which also should be considered.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
