zephyr45 commented on PR #13060:
URL: https://github.com/apache/seatunnel/pull/13060#issuecomment-6916834611

   @SEZ9 Thanks for the recheck. Confirming the two remaining points against 
unchanged head `7c272c5a`:
   
   **F3:** The complete sentence is: “Flink's adapter counters can also differ 
between writers.” Aligning transaction IDs with those counters cannot identify 
a transaction that was never captured in retained checkpoint state. A fresh 
writer fences sequence one; a later abandoned transaction absent from retained 
state may hold back `read_committed` consumers until `transaction.timeout.ms` 
expires. That is the documented timeout boundary for this case, not a claim of 
immediate cleanup. It is already stated beside the recovery semantics in 
[EN](https://github.com/apache/seatunnel/blob/7c272c5a246e617bff1f737fa4e78ed3a1ab7bf5/docs/en/connectors/sink/Kafka.md#L104)
 and 
[ZH](https://github.com/apache/seatunnel/blob/7c272c5a246e617bff1f737fa4e78ed3a1ab7bf5/docs/zh/connectors/sink/Kafka.md#L100).
 Users should account for this possible delay when choosing the timeout, 
alongside normal transaction/checkpoint duration.
   
   **F5/F6:** This head does **not** introduce a single cross-writer 
legacy-fencing owner or a new hard bound on the probe chain. [Deduplication is 
within each 
writer](https://github.com/apache/seatunnel/blob/7c272c5a246e617bff1f737fa4e78ed3a1ab7bf5/seatunnel-connectors-v2/connector-kafka/src/main/java/org/apache/seatunnel/connectors/seatunnel/kafka/sink/KafkaSinkWriter.java#L333-L349);
 [the sender scans until a fresh producer 
epoch](https://github.com/apache/seatunnel/blob/7c272c5a246e617bff1f737fa4e78ed3a1ab7bf5/seatunnel-connectors-v2/connector-kafka/src/main/java/org/apache/seatunnel/connectors/seatunnel/kafka/sink/KafkaTransactionSender.java#L200-L216).
 Concurrent sibling probes can still bump epochs and extend each other's scans. 
I consider this acceptable as the one-time legacy migration cost for this PR 
and am retaining coordinated cleanup as a separate optimization follow-up. 
Retries can incur cleanup work again. A safe optimization needs per-prefix 
ownership, owner failover
  and mixed-prefix handling; arbitrarily choosing writer zero is insufficient.
   
   I have updated the PR description to make these boundaries explicit and to 
reflect the successful [Apache 
Build](https://github.com/apache/seatunnel/runs/127712599445) and [fork 
Build](https://github.com/zephyr45/seatunnel/actions/runs/41960786230/attempts/2)
 at this head. The previously reported 583-test validation remains for this 
same unchanged SHA; this clarification does not claim a new test run.
   
   Please complete your diff recheck of F1/F2/F4/F7 and the disposition of 
F3/F5/F6 when ready. Maintainer approval remains pending.
   
   


-- 
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