MINOR: Remove redundant SuppressIntegrationTests#5896
MINOR: Remove redundant SuppressIntegrationTests#5896guozhangwang merged 2 commits intoapache:trunkfrom vvcephei:test-resilience
Conversation
|
We will run the tests 10 times and verify no SuppressIntegrationTest failures before merging. |
|
Java 8: Java 11: |
bbejeck
left a comment
There was a problem hiding this comment.
LGTM, but I think we still need some tests with a live broker especially when suppression with spill to disk is implemented. Given what you said in the PR comments, maybe we can include a system test?
mjsax
left a comment
There was a problem hiding this comment.
Would be good to get @guozhangwang thoughts on this, too.
|
hah. Looks like I only needed one run to fail my own target: https://builds.apache.org/job/kafka-pr-jdk11-scala2.12/552/testReport/junit/org.apache.kafka.streams.integration/SuppressionIntegrationTest/shouldShutdownWhenRecordConstraintIsViolated/ |
|
@bbejeck I share your concern, and I guess it's an important thing to hash out up front... Do you think this tradeoff is worth it? In particular, are we giving up meaningful test coverage and losing the value of testing if we make changes like this? |
|
The single remaining SuppressionIntegrationTest still timed out after 30s, so I'm also experimenting with bumping the timeout to 60s. |
|
no java 8 fails. retest this, please. |
|
no java 8 fails. java 11 fails: kafka.admin.ResetConsumerGroupOffsetTest.testResetOffsetsNotExistingGroup |
|
retest this, please |
|
I personally don't believe that the risk for this particular test is large. It's testing with a single partition and everything is based on event-time. The only difference I see is, that using Btw: I am wondering, why we don't remove the whole test? The two remaining error scenarios can be tested as unit-test, too, IMHO. |
|
I wanted to actually test that Streams shuts down, since that's what the feature is advertised to do. My reasoning was that I could have a unit test that verifies it throws an exception, but the only way to make sure that Streams shuts down is actually to run the whole thing. Does this seem like good thinking, or am I overengineering? |
|
Java 11 fails: kafka.api.PlaintextConsumerTest.testLowMaxFetchSizeForRequestAndPartition retest this, please. |
Depends if we have a test Note: if we don't have this case, it might be worth to add this test to get rid of the other two. I personally believe in decomposition, and both tests tests from this class have a huge overlap (testing the shutdown part) that the new test could extract and cover at once. |
|
Since opening this ticket, the Java11 version has been upgraded (https://issues.apache.org/jira/browse/INFRA-17235), and an issue with our build configuration has also been fixed, so this PR may not be needed to improve stability. I'll put the question to the reviewers whether we should still proceed. Retest this, please. |
|
That is an unrelated failure: and it's the same for java 8 and 11. retest this, please. |
The removed tests have counterparts covered by SuppressScenarioTest using the TopologyTestDriver. This will speed up the build and improve stability in the CPU-constrained Jenkins environment. Reviewers: Bill Bejeck <bill@confluent.io>, Matthias J. Sax <matthias@confluent.io>, Guozhang Wang <wangguoz@gmail.com>
The removed tests have counterparts covered by SuppressScenarioTest using the TopologyTestDriver.
This will speed up the build and improve stability in the CPU-constrained Jenkins environment.
Committer Checklist (excluded from commit message)