KAFKA-19481: Fix flaky test testConsumerGroupHeartbeatWithRegex

[jmmonte2]

Description

This change fixes the flaky test testConsumerGroupHeartbeatWithRegex, which fails with the following log. Develocity link
org.opentest4j.AssertionFailedError: Unexpected assignment ConsumerGroupHeartbeatResponseData(throttleTimeMs=0, errorCode=0, errorMessage=null, memberId='OGfeiEjOQbqUTsJgtGMCdQ', memberEpoch=1, heartbeatIntervalMs=5000, assignment=null) ==> expected: not <null>

I addressed the issue by using TestUtils.tryUntilNoAssertionError() to allow for retries.

Root Cause:
The failure occurs because the test depends on an async operation, refreshRegularExpressions, within GroupMetadataManager , which may not complete running before assertion runs.

• Also 1 indicator async operation did not finish is the log memberEpoch=1 because when this test successful runs memberEpoch=2 at end. Initially updated from epoch 0 -> 1 here. Once refreshRegularExpressions is done, handleRegularExpressionsResult updates epoch from 1 -> 2 here

refreshRegularExpressions is responsible for resolving regular expression based subscriptions to the current set of matching topic names in the cluster.

Testing

• Used this command
  for i in {1..100}; do echo "Run #$i"; ./gradlew :core:integrationTest --rerun-tasks --tests kafka.api.AuthorizerIntegrationTest.testConsumerGroupHeartbeatWithRegex; if [ $? -ne 0 ]; then echo "Test failed on run #$i"; exit 1; fi; done; echo "All 100 runs passed successfully."
• Also added intentional wait to async call and confirmed it passed

Note:
I ran this 100 times locally, but was unable to reproduce the same error. Only way I was able to reproduce was by adding a sleep of a second to async call then got the same exact error.

[squah-confluent]

lgtm, since testConsumerGroupHeartbeatWithRegexWithTopicDescribeAclAddedAndRemoved follows the same approach.

Can we update the PR description to explain what the async operation and test failure are? It's not immediately clear unless you're familiar with regex subscriptions.

[jmmonte2]

lgtm, since testConsumerGroupHeartbeatWithRegexWithTopicDescribeAclAddedAndRemoved follows the same approach.

Can we update the PR description to explain what the async operation and test failure are? It's not immediately clear unless you're familiar with regex subscriptions.

@squah-confluent thanks for the review. I went ahead and updated.

[squah-confluent]

@jmmonte2 That's more than enough detail, thank you for updating the description!

[jmmonte2]

@squah-confluent Thank you! I noticed the approval did not go through due to lack of write access. Do you have another reviewer I can add on?

[jmmonte2]

@FrankYang0529 Could you please also review?