aromanenko-dev commented on a change in pull request #11090: [BEAM-9470]
:sdks:java:io:kinesis:test is flaky
URL: https://github.com/apache/beam/pull/11090#discussion_r391779556
##########
File path:
sdks/java/io/kinesis/src/test/java/org/apache/beam/sdk/io/kinesis/ShardReadersPoolTest.java
##########
@@ -324,10 +325,15 @@ public void shouldCallRateLimitPolicy()
}
}
+ ArgumentCaptor<List<KinesisRecord>> recordsCaptor =
ArgumentCaptor.forClass(List.class);
+ verify(customRateLimitPolicy,
atLeastOnce()).onSuccess(recordsCaptor.capture());
+ List<List<KinesisRecord>> capturedRecords = recordsCaptor.getAllValues();
+ assertThat(capturedRecords).contains(
+ ImmutableList.of(a, b),
+ singletonList(c),
+ singletonList(d),
+ Collections.emptyList()
+ );
verify(customRateLimitPolicy).onThrottle(same(e));
- verify(customRateLimitPolicy).onSuccess(eq(ImmutableList.of(a, b)));
- verify(customRateLimitPolicy).onSuccess(eq(singletonList(c)));
- verify(customRateLimitPolicy).onSuccess(eq(singletonList(d)));
Review comment:
I did some testing on my side and I think that the problem is with using the
same mock of `RateLimitPolicy` for both `ShardRecordsIterator`s calling in
different threads. For example, when failed assertion says that "d" record is
absent in `capturedRecords`, actually I see (using debug output) that it was
read with `shardRecordsIterator.readNextBatch()` in `readLoop()`.
Since, in case of using the real `RateLimitPolicy` in normal code, the
instances of them will be created per every `readLoop()` thread (right?). So,
we need to test it in the same way with mock policies.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
