1996fanrui commented on code in PR #22392:
URL: https://github.com/apache/flink/pull/22392#discussion_r1172355862
##########
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/SubtaskCheckpointCoordinatorImpl.java:
##########
@@ -381,7 +383,7 @@ private void registerAlignmentTimer(
registerTimer.registerTask(
() -> {
try {
-
operatorChain.alignedBarrierTimeout(checkpointId);
+
operatorChain.alignedBarrierTimeout(checkpointId, metrics);
Review Comment:
> All in all, I'm started to think that maybe your original idea, to
approximate the true/false flag based on the bytesPersistedDuringAlignment > 0
might be the lesser evil. The case when bytesPersistedDuringAlignment == 0 but
the checkpoint barrier actually timed out in the output buffers is quite
extreme/rare, and shouldn't be that significant to the end user and probably
not worth of making the code so much more complicated.
I agree with you, these solutions are complicated, and it probably not worth
of making the code so much more complicated, so I prefer generate unaligned
checkpoint type based on persisted data.
Do you think it's ok? If yes, I can go ahead.
--
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]
