[
https://issues.apache.org/jira/browse/KAFKA-10084?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17124224#comment-17124224
]
John Roesler commented on KAFKA-10084:
--------------------------------------
Thanks to some advice from [~mjsax] , it seems like this is just a race
condition, as we may check the uncommitted end offset only once and might miss
the end/abort transaction marker, while we repeatedly check the committed end
offset and would eventually see the end transaction marker, which accounts for
the off-by-one error apparent in the stacktrace.
> System test failure: StreamsEosTest.test_failure_and_recovery_complex
> ---------------------------------------------------------------------
>
> Key: KAFKA-10084
> URL: https://issues.apache.org/jira/browse/KAFKA-10084
> Project: Kafka
> Issue Type: Task
> Affects Versions: 2.6.0, 2.7.0
> Reporter: John Roesler
> Assignee: John Roesler
> Priority: Major
> Fix For: 2.6.0, 2.7.0
>
>
> The test failed with message:
> {code:java}
> RemoteCommandError: ubuntu@worker14: Command 'grep ALL-RECORDS-DELIVERED
> /mnt/streams/streams.stdout' returned non-zero exit status 1.{code}
> And I found in the verifier's stderr:
> {code:java}
> java.lang.IllegalStateException: Offset for partition echo-1 is larger than
> topic endOffset: 2422 > 2421
> at
> org.apache.kafka.streams.tests.EosTestDriver.verifyAllTransactionFinished(EosTestDriver.java:604)
> at
> org.apache.kafka.streams.tests.EosTestDriver.verify(EosTestDriver.java:184)
> at
> org.apache.kafka.streams.tests.StreamsEosTest.main(StreamsEosTest.java:82){code}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
