SEZ9 commented on PR #13060:
URL: https://github.com/apache/seatunnel/pull/13060#issuecomment-6961534688

   @zephyr45 Not done yet: I haven't updated my review of `10e618cb`. The 
thread tells me what changed in `cf52689e`, but not how each of my findings was 
resolved, so I can't yet confirm the fixes address them.
   
   **Settled**
   - **F5/F6** (legacy shared-namespace fencing repeated by every restoring 
writer): deferring these to #13298 is fine with me. They are low-severity 
performance items and don't need to block the correctness fixes here.
   
   **What I need to close the rest**
   - **F1**: Which change stops `validateRestoredStateOwner` from assuming 
modulo redistribution, so a Flink scale-down restore of an EXACTLY_ONCE sink no 
longer throws `IllegalStateException`? Is there a test for it?
   - **F2**: How does a scale-down restore behave now when `transaction_prefix` 
is not configured and the restored states carry different generated prefixes?
   - **F3**: For a writer that restores with no state (scale-up), where does 
its transaction sequence start now, and is its namespace fenced before the 
first attempt?
   - **F4**: You wrote that "Kafka's Spark retry identity is unchanged" after 
`cf52689e`. Does that mean a Spark task retry still gets a new transactional 
namespace and leaves the previous attempt's open transaction unfenced? If so, 
please say whether you are fixing it here or deferring it, and where it is 
tracked.
   - **F7**: Are the `transaction_prefix` option row and the engine scope of 
the rescaling claim now aligned with the Semantics paragraph in the Kafka sink 
docs?
   
   A short per-finding note (commit plus the test that covers it) is enough. 
Once I have that, I'll re-review the latest head and update the review.
   
   The multi-sink point is with @nzw921rx to confirm after the override removal.
   
   <!-- streview-comment:2405 -->


-- 
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