[
https://issues.apache.org/jira/browse/FLINK-26668?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17508004#comment-17508004
]
Lijie Wang commented on FLINK-26668:
------------------------------------
[~kingWang]
In fact, I don't know much about the Flink SQL plan generation, and don't know
why this change affects the parallelism of multiinput. But the truth is, this
change does affect the parallelism of multiinput, as the error in screenshot
above. As discussed in [PR19040|https://github.com/apache/flink/pull/19040] ,
given that we had multiple issues trying to fix this, and that a temporary fix
would complicate and add more risk, we think a redesign approach with a new
source interface later would be a better solution.
> Unsupportedoperation exception caused by createActualEdge error in adaptive
> batch scheduler
> -------------------------------------------------------------------------------------------
>
> Key: FLINK-26668
> URL: https://issues.apache.org/jira/browse/FLINK-26668
> Project: Flink
> Issue Type: Bug
> Components: API / DataStream
> Affects Versions: 1.15.0
> Environment:
>
> Reporter: kangTwang
> Priority: Major
> Attachments: aaa.jpg, bbb.jpg
>
>
> In the AdaptiveBatch TPC DS test, the following error messages appear when
> executing query_23a，If the upstreamnode is Calc [50], the downstreamnode is
> multipleinput [75], and they are forwardpartitioners, the parallelism is
> inconsistent, and an unsupported operationexception is thrown：
> !aaa.jpg!
> Because the error is thrown in the streamgraph stage, the specific DAG
> diagram cannot be seen. Therefore, non adaptive batch is adopted. The DAG
> diagram executed is as follows:
> !bbb.jpg!
> Check Calc [50] and find that the downstream operator is HashAggregate[51],
> not MultipleInput[75]
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
