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

Posted to issues@flink.apache.org by "JingWei Li (Jira)" <ji...@apache.org> on 2026/04/26 15:24:00 UTC

[jira] [Created] (FLINK-39548) CommonExecTableSourceScan uses physical row type instead of output type for upsert source key partitioning

JingWei Li created FLINK-39548:
----------------------------------

             Summary:  CommonExecTableSourceScan uses physical row type instead of output type for upsert source key partitioning 
                 Key: FLINK-39548
                 URL: https://issues.apache.org/jira/browse/FLINK-39548
             Project: Flink
          Issue Type: Bug
          Components: Table SQL / Planner
            Reporter: JingWei Li


    • When  CommonExecTableSourceScan  wraps an upsert/changelog source with configured parallelism, it derives primary-key indices from the    
  physical row type.                                                                                                                            
    • If the source output is projected/reordered before the partitioning wrapper is created, those indices no longer match the actual output   
  schema.                                                                                                                                       
    • This causes the generated source partitioning key selector to use wrong fields.  



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