MartijnVisser commented on PR #313:
URL: 
https://github.com/apache/flink-connector-kafka/pull/313#issuecomment-5623700914

   This isn't fixing FLINK-40585. Runs 34473915461 and 30181683657 both show a 
precommitted id still ONGOING, so they're FLINK-40626 and #314 is what closes 
them. Run 31916828021 is the only one without that signature, and it's the v4.0 
cell: v4.0 still has `ThrottleUntilFirstCheckpointStrategy`, which 
FLINK-38696's `a90e8f48` replaced on main with `perCheckpoint(10)`, so the 
third job writes the rest in one transaction there and loses 966 records where 
main would lose ten. I can't tell from the log whether that last transaction 
was cut off by a late marker or hadn't finished committing.
   
   The direction is right, the condition isn't. Wait on transaction state, not 
on the LSO catching up with the high watermark. I'd take #314 first and then 
move this to v4.0 under its own ticket, and I wouldn't carry it on main: the 
marker lands inside 2 ms and the drain spends over 3 ms building the consumer 
before it reads the offsets, so I couldn't hit the race here at all.


-- 
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