AHeise commented on a change in pull request #14348:
URL: https://github.com/apache/flink/pull/14348#discussion_r543196339
##########
File path:
flink-tests/src/test/java/org/apache/flink/test/checkpointing/UnalignedCheckpointITCase.java
##########
@@ -321,7 +321,7 @@ public void invoke(Long value, Context context) throws
Exception {
if (backpressure) {
// induce backpressure until enough checkpoints
have been written
if (random.nextInt(100) == 42) {
- Thread.sleep(0, 100_000);
+ Thread.sleep(100);
Review comment:
Err, I meant to say that two instances of backpressure may not occur
that I like to see covered:
- Backpressure results in full buffers to be snapshotted and restored. If we
just accumulate 100 record, then with the current record size and the min
buffer size (4kb), it's likely that only a few bytes are snapshotted and
restored and also only individual buffers and not multiples. So, we may never
see in the ITCase if multiple buffers are snapshotted/recovered at all and in
the correct order.
- Consequently, we may also not see backpressure in all exchanges, but just
the last exchange. But I think it's important to cover all exchanges to test
the different kinds of recovery, especially during rescaling (to come). It's
also good to cover backpressure being propagated to the source ultimately.
----------------------------------------------------------------
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]
