[
https://issues.apache.org/jira/browse/GEODE-8230?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17127179#comment-17127179
]
ASF GitHub Bot commented on GEODE-8230:
---------------------------------------
onichols-pivotal opened a new pull request #5217:
URL: https://github.com/apache/geode/pull/5217
before, we ran Build, then all <x>Test jobs in parallel (which takes ~5
hours), then all Benchmark<x> jobs (which takes ~5 hours).
Presumably the benchmarks were run after all tests, on the premise that
benchmarks are meaningless if the code is not correct. However, required PR
checks now prevent incorrect code from getting into develop in the first place,
and the remaining occasional failures are flaky, so it's rather arbitrary to
gate benchmarks on a lottery.
However, faster feedback has clear benefits, hence this PR to cut our CI
pipeline from 10 hours to 5 hours.
Also eliminates redundant jobs (ApiCheck and UnitTest are deterministic and
already have required JDK11 flavors in the PR pipeline, so JDK8 flavor only is
more than sufficient in the develop pipeline)
Also decreases benchmarks max_in_flight from 3 to 2, since now that we have
support branches, we don't want to max our quota just from develop. Some sort
of global "pool" resource might be a more elegant way to manage this resource
quota across pipelines.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> run benchmarks in parallel with other CI
> ----------------------------------------
>
> Key: GEODE-8230
> URL: https://issues.apache.org/jira/browse/GEODE-8230
> Project: Geode
> Issue Type: Improvement
> Components: ci
> Reporter: Owen Nichols
> Assignee: Owen Nichols
> Priority: Major
>
> currently, we run Build, then all <x>Test jobs in parallel (which takes ~5
> hours), then all Benchmark<x> jobs (which takes ~5 hours).
> Presumably the benchmarks were run after all tests, on the premise that
> benchmarks are meaningless if the code is not correct. However, required PR
> checks now prevent incorrect code from getting into develop in the first
> place, and the remaining occasional failures are flaky, so it's rather
> arbitrary to gate benchmarks on a lottery.
> However, faster feedback has clear benefits, and cutting our CI pipeline from
> 10 hours to 5 hours seems like low-hanging fruit.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
