[
https://issues.apache.org/jira/browse/IMPALA-15205?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18098350#comment-18098350
]
Joe McDonnell commented on IMPALA-15205:
----------------------------------------
It looks like there is a timing issue. If the retry() call notices that the
memory usage is above 270MB shortly after it reaches it, then it still has some
distance to grow and the check will fail. If the retry() call notices a bit
later, the query could be well past 270MB and won't have as much to grow. Code
coverage on a particular test configuration is on the wrong side of this, but
this rarely reproduces on my workstation. If I change the retry() call's
parameters to sleep 0.1 seconds (still up to 10 seconds) rather than 1 second,
I can reproduce it easily.
> TestConjuncts.test_create_cache_many_tables fails on coverage build
> -------------------------------------------------------------------
>
> Key: IMPALA-15205
> URL: https://issues.apache.org/jira/browse/IMPALA-15205
> Project: IMPALA
> Issue Type: Bug
> Components: Test
> Affects Versions: Impala 5.0.0
> Reporter: Joe McDonnell
> Priority: Critical
> Labels: broken-build, flaky
>
> On our exhaustive build with code coverage,
> TestConjuncts.test_create_cache_many_tables fails with:
> {noformat}
> custom_cluster/test_conjuncts.py:121: in test_create_cache_many_tables
> assert mem_after - mem_before < 25 * 1024 * 1024, "Memory usage increased
> during " \
> E AssertionError: Memory usage increased during sleep. Before: 292945920
> After: 320700416
> E assert (320700416 - 292945920) < ((25 * 1024) * 1024)
> _TestConjuncts__wait_for_270mb = <function
> TestConjuncts.test_create_cache_many_tables.<locals>.__wait_for_270mb at
> 0x7f2aa0e1db20>
> handle = <tests.common.impala_connection.OperationHandle object
> at 0x7f2a3a920ad0>
> mem_after = 320700416
> mem_before = 292945920
> self = <tests.custom_cluster.test_conjuncts.TestConjuncts
> object at 0x7f2a973b5310>
> unique_database = 'test_create_cache_many_tables_49a1d922'{noformat}
> This seems to be unique to the coverage build. We could use a different limit
> for the coverage build or disable this specific test.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
