[
https://issues.apache.org/jira/browse/IMPALA-12194?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17731170#comment-17731170
]
ASF subversion and git services commented on IMPALA-12194:
----------------------------------------------------------
Commit 05197c6a4c59437fd2f6ac0402d4f3753f4516eb in impala's branch
refs/heads/master from Joe McDonnell
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=05197c6a4 ]
IMPALA-12194: Fix flakiness in DataCacheTest.SetReadOnly
DataCacheTest.SetReadOnly spawns a bunch of threads and
then sets the cache to read only. The threads do a bunch
of stores to the cache and then a bunch of reads. The
test DCHECKs that at least one of the reads will be a cache
miss, because at least one store can't happen because the
cache is read only. This is racy, because it is possible for
the threads to complete all the stores before the cache
is set to read only.
This modifies the test to spawn half the threads, then set
the cache read only, then spawn the other half of the threads.
This guarantees that some threads can't complete their stores,
so the DCHECK won't fire.
Testing:
- Ran DataCacheTest.SetReadOnly* in a loop for 1000 iterations
Change-Id: Id086c8be27200965c60f716b7303a0627b769281
Reviewed-on: http://gerrit.cloudera.org:8080/20026
Reviewed-by: Joe McDonnell <joemcdonn...@cloudera.com>
Tested-by: Joe McDonnell <joemcdonn...@cloudera.com>
> DataCacheTest.SetReadOnly intermittently fails on debug builds
> --------------------------------------------------------------
>
> Key: IMPALA-12194
> URL: https://issues.apache.org/jira/browse/IMPALA-12194
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 4.3.0
> Reporter: Joe McDonnell
> Assignee: Joe McDonnell
> Priority: Blocker
> Labels: broken-build, flaky
>
> Some precommit jobs have been hitting intermittent test failures on
> DataCacheTest.SetReadOnly:
> {noformat}
> /home/ubuntu/Impala/be/src/runtime/io/data-cache-test.cc:165
> Expected: (cache_misses) > (0), actual: 0 vs 0{noformat}
> The test has a race condition that means it can't guarantee that the
> cache_misses is strictly greater than zero.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
