> Thanks Geoffrey for standing in my defense
Hopefully this is just a passing comment, not meant to be indicative of
hurt feelings but, as VP, I feel like I should address it. Sorry for
something the heavy-handedness to follow:
The results of votes shouldn't be taken personally. Because they are
discrete values, they have the tendency to "feel bad". It's like seeing
a bad grade on a test, but the reality is that there's a very broad
spectrum of a reason that you might get a -1 on a vote.
To this point, Andrew's suggestion that this shouldn't have been a vote
in the first place is a good one and would've avoided any hurt feelings.
I just want to make sure it's clear that my -1 was a terse way to say
the following:
* Your proposal would reduce the efficacy of pre-commit testing (ITs not
being run _successfully_)
* Reliance on hardware we don't control is a major impediment to getting
consistently passing tests (lots of prior-art here, Andrew alludes to it
via the work done in HBase)
* We have a solution to integration testing on pull requests with the
same setup we have today
* I've seen lots of issues running non-trivial tests on Travis
(underpowered, again, as Andrew said)
All of this is a long way to say: thanks for trying to make testing
better/faster in Phoenix, Priyank. I hope you're not discouraged by
going back to the drawing board. Feel free to ask for review/feedback
early and often. We're all here to make things better. We will sometimes
have differing opinions on how to get there, but this is natural and
human :)
On 6/21/19 3:41 PM, Priyank Porwal wrote:
Thanks Geoffrey for standing in my defense and Andrew for consensus
suggestion!
We can ignore the vote for now and continue the discussion though, which is
great. Hopefully it will lead us to some next steps; and maybe consensus.
I'll explore HBase's usage of Yetus. Does anyone have suggestions to get to
auto-triggered build&test runs and coverage reports other than Yetus?
Thanks,
Priyank
On Fri, Jun 21, 2019 at 11:18 AM Priyank Porwal <[email protected]>
wrote:
The IT do successfully start in Travis just fine, but Travis has a time
limit of 50 minutes for open-source projects, and Phoenix ITs run for
almost 3 hours. This limit is applied per 'job' and not to the complete
cycle. I did attempt to break down Phoenix ITs into multiple jobs, but
couldn't complete that due to lack of experience with Maven and other
plugins. Hence, I proposed to initiate the on-boarding process which would
be no-op for existing Phoneix contributors. It is not yet ready to be a
replace ASK infra, at least not yet.
Ignoring TravisCI and CodeCov specifically (I have no affinity or
anti-affinity to either), the aim was to get a similar experience - devs
publishing new PRs and/or pushing new commits to existing PRs should
trigger automated build and test-runs; and show code-coverage reports in
addition to test reports. Currently, devs are required to squash all their
commits into a single one and upload patches to trigger test runs. Also,
there is no visibility into coverage info from our test runs. If Yetus or
anything else in ASF infra can support that, I can attempt onboarding to
them.
On Fri, Jun 21, 2019 at 10:49 AM Andrew Purtell <[email protected]>
wrote:
Travis is underpowered compared to ASF test infrastructure. Do the
integration tests run there?
Recommend using ASF infra and Yetus (yetus.apache.org). HBase provides a
completely worked example of how to integrate and use it for precommit.
On Fri, Jun 21, 2019 at 12:15 AM Priyank Porwal <[email protected]>
wrote:
[Converting this thread to a community vote]
I'd like to start Travis-CI and CodeCov integration after getting some
success with both on a fork in my personal account. Checkout -
https://github.com/priyankporwal/phoenix/pull/3
Things to note:
1. TravisCI kicked-off as soon as the PR is created and/or new commits
are
pushed. No additional developer action is necessary.
2. Once completed, code-coverage report is uploaded to CodeCov which
produced a nice color-coded graph of different folders/files. Detailed
reports linked from the PR as well.
3. Confirmed that compilation and test failures resulted in CI flagging
the
PR.
4. Currently, TravisCI only runs unit-tests. "mvn verify" takes too long
for it to be included in Travis' scipt stage (max allowed time per job
is
50 mins) - I made several attempts to break up the tests into several
jobs,
but lack of maven skills prevented me from achieving that goal.
5. Repo-admin permissions only needed to start this integration
(one-time)
and thereafter, incremental improvements can be made via any regular PR.
Perhaps folks with maven expertise can get to it sooner.
Please vote on proceeding with the integration with TravisCI and
CodeCov.
Thanks,
Priyank
On Tue, May 28, 2019 at 2:54 PM Thomas D'Silva
<[email protected]> wrote:
I assume we want to run all the ITs. Whevenver a PR is created Travis
CI
will automatically runs all the tests
and post the results to the PR.
On Tue, May 28, 2019 at 2:08 PM Geoffrey Jacoby <[email protected]>
wrote:
I don't know much about this particular tool, but something like
this
would
be good.
Our current toolchain, with HadoopQA needing a JIRA patch and our
code
reviews mostly migrating to Github is really awkward to deal with,
so
TravisCI's Github integration's a definite plus.
An example of Tephra's integration is here[1]: and on TravisCI's
home
page[2] they mention that open source projects are free.
Assuming there are no licensing, scalability or implementation
gotchas
I'd
be a +1.
Geoffrey
[1] https://travis-ci.org/apache/incubator-tephra
[2] https://travis-ci.org/
On Tue, May 28, 2019 at 1:31 PM William Shen <
[email protected]
wrote:
+1 It would be awesome to be able to do this.
Any concerns if we choose to run long IT as part of this setup?
On Tue, May 28, 2019 at 1:00 PM Pedro Boado <[email protected]>
wrote:
What IT would you suggest to run? Testsuite (including long IT)
takes
~2h.
On Tue, 28 May 2019, 20:40 Thomas D'Silva, <
[email protected]
.invalid>
wrote:
+1 I think its a great idea. This would make it easier for new
contributors
to run tests
and also make it easier for committers to verify a patch
doesn't
break
functionality.
On Tue, May 28, 2019 at 12:34 PM Priyank Porwal <
[email protected]
wrote:
Hi,
What do you guys think about this work to setup Travis-CI
and
CodeCoverage
for Phoenix? The objective would be to run unit and
integration
tests
on
each PR, show code-coverage reports and perhaps also do
checkstyle
checks
(after initial scrubbing effort). This would help rid of
manual
patch
uploads that we need currently, plus bring visibility into
code
health.
Thoughts?
https://issues.apache.org/jira/browse/PHOENIX-4863
Thanks,
Priyank
--
Best regards,
Andrew
Words like orphans lost among the crosstalk, meaning torn from truth's
decrepit hands
- A23, Crosstalk
