AHeise commented on a change in pull request #13562:
URL: https://github.com/apache/flink/pull/13562#discussion_r502233073
##########
File path:
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/io/CheckpointBarrierTracker.java
##########
@@ -82,7 +82,9 @@ public void processBarrier(CheckpointBarrier receivedBarrier,
InputChannelInfo c
// fast path for single channel trackers
if (totalNumberOfInputChannels == 1) {
- notifyCheckpoint(receivedBarrier, 0);
+ markCheckpointStart(receivedBarrier.getTimestamp());
+ markAlignmentEnd(0);
Review comment:
I'd just decouple the two metrics of checkpoint start duration and
alignment times. For tracker, there is no semantics of alignment. We have an
extra call to `System.nanoTime()` that is completely unnecessary (here I don't
worry about performance). It feels like a mismatch of abstraction. Similarly,
when you optimize 1 channel cases, there is just no alignment happening, which
you simulate by ending it early and explicitly overwrite the measurement.
----------------------------------------------------------------
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:
us...@infra.apache.org
