You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Tony Reix (JIRA)" <ji...@apache.org> on 2015/02/20 09:54:11 UTC
[jira] [Created] (KAFKA-1970) Several tests are not stable
Tony Reix created KAFKA-1970:
--------------------------------
Summary: Several tests are not stable
Key: KAFKA-1970
URL: https://issues.apache.org/jira/browse/KAFKA-1970
Project: Kafka
Issue Type: Bug
Affects Versions: 0.8.1
Environment: Several:
- RHEL7.1/x86_64 <=== My reference
- RHEL7.0/x86_64
- Ubuntu /x86_64
- RHEL7.1/PPC64LE
- RHEL7.0/PPC64BE
- OpenJDK 1.7
- IBM JVM 1.7
Reporter: Tony Reix
I'm porting Kafka 0.8.1 on RHEL 7.1/PPC64LE.
Since it looked that tests were unstable, I've launched the tests on several environments, in order to have a wide view.
I'm using:
- ./gradlew build -x signArchives
or:
- ./gradlew test -x signArchives
Results seem to show:
- Tests are unstable everywhere (very few on your Ubuntu/x86_64 test env)
- IBM JVM shows some more issues than OpenJDK
- Sometimes, tests are not lauched, with no reason.
But not on my reference environment (RHEL7.1/x86_64/OPenJDK)
- Open JDK : Tests runs results:
- dorado-vm2 - RHEL7.0/x86_64 :
- 238 tests completed, 82 failed
- 238 tests completed, 94 failed
- BUILD SUCCESSFUL
- dorado-vm3 - Ubuntu /x86_64 :
- BUILD SUCCESSFUL
- soe01x - RHEL7.1/x86_64 :
- 238 tests completed, 4 failed x 2 times
- soe07-vm1 - RHEL7.1/PPC64LE:
- BUILD SUCCESSFUL
- 238 tests completed, 2 failed
- 238 tests completed, 3 failed
- IBM JVM : Tests runs results:
- dorado-vm2 - RHEL7.0/x86_64 :
- 1 failed + Tests Blocked
- BUILD SUCCESSFUL
- soe01x - RHEL7.1/x86_64 :
- 238 tests completed, 6 failed
- 238 tests completed, 4 failed x 3 times
- 238 tests completed, 5 failed
- soe07-vm1 - RHEL7.1/PPC64LE:
- 238 tests completed, 1 failed
- BUILD SUCCESSFUL
- laurel6 - RHEL7.0/PPC64BE:
- 238 tests completed, 1 failed
- BUILD SUCCESSFUL
=========================================================
I think that these tests are unstable:
kafka.server.LogRecoveryTest > testHWCheckpointNoFailuresMultipleLogSegments
kafka.server.LogRecoveryTest > testHWCheckpointWithFailuresMultipleLogSegments
kafka.admin.DeleteTopicTest > testAutoCreateAfterDeleteTopic
kafka.admin.DeleteTopicTest > testPreferredReplicaElectionDuringDeleteTopic
kafka.server.RequestPurgatoryTest > testRequestExpiry
=========================================================
These tests are failing often (always on my reference environment (RHEL7.1/x86_64/OpenJDK), but not on Ubuntu) :
kafka.server.LogOffsetTest > testEmptyLogsGetOffsets
kafka.server.LogOffsetTest > testGetOffsetsBeforeLatestTime
kafka.server.LogOffsetTest > testGetOffsetsBeforeEarliestTime
kafka.server.LogOffsetTest > testGetOffsetsBeforeNow
=========================================================
As an example or random failures, on my reference environment (RHEL7.1/x86_64/OpenJDK) , the test:
kafka.server.LogRecoveryTest > testHWCheckpointNoFailuresMultipleLogSegments
failed 2 times out of 12.
=========================================================
On Ubuntu/x86_64/OpenJDK , out of 3 runs of :
gradlew test -x signArchive
I've got:
- 3 Full success
- 1 launch that did NOT run the tests
=========================================================
Still on x86_64/OpenJDK , I'm surprised to always have 4 failures with RHEL 7.1 and none on Ubuntu.
Some issue within RHEL 7.1 and/or Java ?
=========================================================
On RHEL 7.1 / PPC64LE / IBM JVM, I see a wide unstability.
I've run 12 tests.
Parsing them about "FAILED" tests with:
for i in 1 2 10 11 12 13 14 15 16 17 18 19; do $i; N=`grep FAILED gradlew.build.IBMJVM.res$i | wc -l`; echo $i":"$N; done
gave:
1:3
2:0
10:0
11:3
12:4
13:49
14:3
15:4
16:0
17:0
18:0
19:0
Doing the same about "PASSED" tests, I've got:
1:372
2:238
10:0
11:372
12:236
13:191
14:237
15:236
16:238
17:0
18:0
19:0
Showing:
- 4 launches did NOT run the tests
- 2 launches were SUCCESSFUL
- for the others, there were 3, 4 or 49 FAILED tests
=========================================================
Conclusions: I think that it would be useful for Kafka project:
- to run tests with IBM JVM in addition to OpenJDK.
- to run tests on a different Linux distrib than Ubuntu: RHEL .
- to check (by running it many times) that the following test is stable in your standard test environment:
kafka.server.LogRecoveryTest > testHWCheckpointNoFailuresMultipleLogSegments
On my side, there are other causes of unstability in my specific environments (PPC64) that I have to study.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)