You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by "Bo Cui (Jira)" <ji...@apache.org> on 2023/03/26 23:12:00 UTC

[jira] [Commented] (FLINK-31595) MiniBatchLocalGroupAggFunction produces wrong aggregate results with state clean up

    [ https://issues.apache.org/jira/browse/FLINK-31595?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17705135#comment-17705135 ] 

Bo Cui commented on FLINK-31595:
--------------------------------

like the sql:

select a, max(b), max(c) from (

  select * from (

    select *, row_number(...) as rn from T

  ) where rn=1

) group by a

 

the ttl of job is 1h and minibatch is enabled, so row_number ttl is 1.5h, group ttl is 1h.

The time of retract data may be 1.2h,  and minbatch state has expired,  so the aggregation result is incorrect.

> MiniBatchLocalGroupAggFunction produces wrong aggregate results with state clean up
> -----------------------------------------------------------------------------------
>
>                 Key: FLINK-31595
>                 URL: https://issues.apache.org/jira/browse/FLINK-31595
>             Project: Flink
>          Issue Type: Bug
>            Reporter: Bo Cui
>            Priority: Critical
>              Labels: pull-request-available
>
> If the upstream operator supports retract data, and the first data in a batch may be retract data, and the retract data should be ignored.
> https://github.com/apache/flink/blob/a64781b1ef8f129021bdcddd3b07548e6caa4a72/flink-table/flink-table-runtime/src/main/java/org/apache/flink/table/runtime/operators/aggregate/MiniBatchLocalGroupAggFunction.java#L68



--
This message was sent by Atlassian Jira
(v8.20.10#820010)