kezhuw commented on pull request #14831:
URL: https://github.com/apache/flink/pull/14831#issuecomment-783149666
@gaoyunhaii Sorry, I propably targets wrong method, I means
`StreamTask.triggerCheckpoint` and its overrides. May be not that serious, but
I think it would be good to code trigger-checkpoint in one place.
I have taken an quick look at new implementation:
* For handling of `EndOfPartitionEvent`, I think it is much straightforward
than old `FinalizeBarrierComplementPrcoessor`.
* For "overtaking buffers for unaligned checkpoint", I think both
methods("flushed to network stack" and "processed by downstream") are
achievable.
* "No, the waiting happens after all the operators have been closed"
* Please aware of FLINK-16383 which ignores `notifyCheckpointCompleted`
after operator closed. Personally, I think it is problem of
`StreamOperator.close` which is documented as "flushing buffered data" but
misused(?) by many operators as "cleanup-resources".
* What if `StreamOperator.close` generates some new records, assumes
that `StreamOperator.close` is the end-of-stream-flush? Will these records be
duplicated in recovery ?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
