[
https://issues.apache.org/jira/browse/CASSANDRA-10492?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14993852#comment-14993852
]
Ariel Weisberg commented on CASSANDRA-10492:
--------------------------------------------
Turns out this is still a test bug, but different from what I thought. The
server isn't timing our requests as fast as expected because the configuration
is wrong.
What I found is that the timeouts for each read request is 10 seconds instead
of the requested one second. This means that with six attempts it hits the
driver's 60 second timeout some of the time.
I set request_timeout_in_ms and range_request_timeout_in_ms and now it seems to
run through the timeouts in 2.2.1 and 2.2.2 quickly.
I updated the pull request.
> Fix failing upgrade tests on 2.1->2.2 path
> ------------------------------------------
>
> Key: CASSANDRA-10492
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10492
> Project: Cassandra
> Issue Type: Bug
> Reporter: Jim Witschey
> Assignee: Ariel Weisberg
> Labels: triage
> Fix For: 2.1.x, 2.2.x
>
> Attachments: test_stdout.txt
>
>
> 2 dtests for the storage engine upgrade on the 2.1->2.2 path fail:
> http://cassci.datastax.com/view/Upgrades/job/storage_engine_upgrade_dtest-21_HEAD-22_tarball/lastCompletedBuild/testReport/
> (If there are 3 failures at that link, don't worry; one was a dtest bug
> that's been fixed.)
> {{upgrade_tests/cql_tests.py:TestCQL.large_count_test}} flaps when a TRUNCATE
> operation times out:
> http://cassci.datastax.com/view/Upgrades/job/storage_engine_upgrade_dtest-21_HEAD-22_tarball/20/testReport/upgrade_tests.cql_tests/TestCQL/large_count_test/history/
> {{upgrade_tests/paging_test.py:TestPagingWithDeletions.test_failure_threshold_deletions}}
> also fails on this upgrade path:
> http://cassci.datastax.com/view/Upgrades/job/storage_engine_upgrade_dtest-21_HEAD-22_tarball/15/testReport/upgrade_tests.paging_test/TestPagingWithDeletions/test_failure_threshold_deletions/history/
> These are in the same ticket for the moment, but if one becomes a longer
> project, it should be broken out into a separate ticket.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
