SEZ9 commented on issue #13280:
URL: https://github.com/apache/seatunnel/issues/13280#issuecomment-6887812137

   Thanks for the detailed, deterministic reproduction and the code 
investigation.
   
   If I follow correctly: `SqlServerStreamingChangeEventSource` calls 
`afterHandleLsn(partition, offsetContext.getOffset())` after querying changes, 
and 
`SqlServerTransactionLogFetchTask.TransactionLogSplitReadTask.afterHandleLsn` 
compares that row-event offset with `lsnSplit.getStopOffset()` before emitting 
END. Since HIGH is database-wide, when only `Other` advances it and the 
selected table `A` has no changes in the LOW-HIGH interval, the offset of A's 
last event can never reach HIGH, so END is not emitted and the buffered 
snapshot records are never released. That matches the missing `backfillReadTask 
execute end` line in your log.
   
   Your local fix (capping each bounded query at the snapshot HIGH watermark 
and emitting END at the original HIGH once the complete query results, 
including an empty result, have been processed) sounds like a reasonable 
direction. Would you be willing to open a PR with that change, along with an 
integration test covering the "another table advances HIGH while the selected 
table has no changes" scenario? Your reproduction steps are a good basis for 
it. If you would rather not, please let us know and we can pick it up.
   
   <!-- streview-comment:2355 -->


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to