[
https://issues.apache.org/jira/browse/BEAM-10471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17168544#comment-17168544
]
Etienne Chauchot commented on BEAM-10471:
-----------------------------------------
Hey Ahmet ! Usually for flakes I leave 2 weeks to have enough historical data
before considering that the flakiness is solved. And it isn't in that case but
the PR allowed me to see that the estimated size returned by the IO can be
multiplied by 2 in some cases (maybe apache cassandra internal size estimates
are not updated in some cases) way above the 20% margin I let in the test
condition. I'll take a look and change to a proper test condition.
> Failure / flake in CassandraIOTest > testEstimatedSizeBytes
> -----------------------------------------------------------
>
> Key: BEAM-10471
> URL: https://issues.apache.org/jira/browse/BEAM-10471
> Project: Beam
> Issue Type: Bug
> Components: io-java-cassandra
> Reporter: Kenneth Knowles
> Priority: P1
> Labels: flake
> Time Spent: 40m
> Remaining Estimate: 0h
>
> https://ci-beam.apache.org/job/beam_PreCommit_Java_Commit/12328/testReport/junit/org.apache.beam.sdk.io.cassandra/CassandraIOTest/testEstimatedSizeBytes/
> {{AssertionError}} with no error message. Please give a meaningful error
> message. Preferably do not use {{assertTrue}} and other less meaningful
> asserters.
> Digging into the code, it is a bounds check that the estimated bytes are +/-
> 10% of expected, since the result is nondeterministic. It fell outside of the
> range but since the assertion is written as {{assertTrue}} we do not see the
> actual value.
> https://github.com/apache/beam/blob/66bcc9a967746cad12dfdeed473fb60c77033514/sdks/java/io/cassandra/src/test/java/org/apache/beam/sdk/io/cassandra/CassandraIOTest.java#L291
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
