[
https://issues.apache.org/jira/browse/HBASE-20387?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16584591#comment-16584591
]
Sean Busbey commented on HBASE-20387:
-------------------------------------
[we check the "rerun all the flaky tests" for up to 40
runs|https://github.com/apache/hbase/blob/master/dev-support/flaky-tests/flaky-reporting.Jenkinsfile#L44]
{code}
flaky_args=("${flaky_args[@]}" --urls
"${JENKINS_URL}/job/HBase-Flaky-Tests/job/${BRANCH_NAME}" --is-yetus False
--max-builds 40)
{code}
Running once an hour, means this is about 2 days lag if a job is no longer
flaky. I believe the "1/5" in the current report was just an artifact of when
it ran since the flaky run job was new.
[we check nightly tests for up to 5
runs|https://github.com/apache/hbase/blob/master/dev-support/flaky-tests/flaky-reporting.Jenkinsfile#L43]
{code}
flaky_args=("${flaky_args[@]}" --urls
"${JENKINS_URL}/job/HBase%20Nightly/job/${BRANCH_NAME}" --is-yetus True
--max-builds 5)
{code}
Running 1/day means that it's about a work week of lag if a job is no longer
flaky.
[the old
job|https://builds.apache.org/job/HBase-Find-Flaky-Tests-old-just-master/configure]
checked the flaky runs for 30 builds and checked the nightly tests for 6
builds.
A test has to avoid being in either list to be run in the normal builds, which
means the failing rate needs to be below 2.5% to get out.
> flaky infrastructure should work for all branches
> -------------------------------------------------
>
> Key: HBASE-20387
> URL: https://issues.apache.org/jira/browse/HBASE-20387
> Project: HBase
> Issue Type: Improvement
> Components: test
> Reporter: Sean Busbey
> Assignee: Sean Busbey
> Priority: Critical
> Fix For: 3.0.0, 1.5.0, 1.2.7, 1.3.3, 2.0.2, 2.2.0, 2.1.1, 1.4.7
>
> Attachments: HBASE-20387.0.patch, HBASE-20387.1.patch
>
>
> We need a flaky list per-branch, since what does/does not work reliably on
> master isn't really relevant to our older maintenance release lines.
> We should just make the invocation a step in the current per-branch nightly
> jobs, prior to when we need the list in the stages that run unit tests. We
> can publish it in the nightly job as well so that precommit can still get it.
> (and can fetch it per-branch if needed)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
