zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Enrico Olivelli <eolive...@gmail.com>
Subject Re: What about using Travis-CI ?
Date Thu, 07 Feb 2019 22:04:27 GMT
Il giorno gio 7 feb 2019, 22:56 Norbert Kalmar <nkalmar@cloudera.com.invalid>
ha scritto:

> Keep in mind we have to run the tests on one thread in 3.4, so keep
> -Dsurefire-forkcount=1.
>

With forkcount=1 tests take too long for Travis.
I will perform experiments on my private fork, we can disable travis for
3.4 or
we can limit Travis to a simple 'full build without tests' in the mean time.

If we will invest more resources in splitting Jenkins jobs we can drop
Travis as it seems too limited for Zookeeper (in Bookkeeper we are using it
only for validation of the build without tests for instance).
The advantage of Travis is the the contributor is free to restart the build
and security risks are over Travis infra.

Also I would like to try to run the suite without requiring fsync to disk,
maybe the build may complete faster.

Enrico



> Thanks,
> Norbert
>
> On Thu, Feb 7, 2019 at 10:32 PM Enrico Olivelli <eolivelli@gmail.com>
> wrote:
>
> > Thank you Andor for mergin Travis configuration file.
> >
> > Unfortunately the build sometimes hangs and Travis fails the build
> >
> > Timeout (30 minutes) reached. Terminating "mvn verify
> > -Dmaven.test.redirectTestOutputToFile=true -Dsurefire-forkcount=4"
> > The command mvn verify -Dmaven.test.redirectTestOutputToFile=true
> > -Dsurefire-forkcount=4 exited with 137.
> >
> >
> > Enrico
> >
> > Il giorno mar 15 gen 2019 alle ore 09:50 Norbert Kalmar
> > <nkalmar@cloudera.com.invalid> ha scritto:
> > >
> > > That's strange, it should work with multiple forks with maven. I had
> > > similar problems, but the problem was PortAssigment didn't get the
> > required
> > > parameters. It has the parameters in zookeeper-server pom.xml:
> > > -Xmx512m -Dtest.junit.threads=${surefire-forkcount}
> > > -Dzookeeper.junit.threadid=${surefire.forkNumber}
> > > And if this would be the problem, much more tests would fail... (Plus
> > > Travis is calling the same surefire plugin so...)
> > >
> > > Tests are very slow with 1 fork, around the 50 minutes mark if I
> remember
> > > correctly.
> > >
> > > Anyway, I would go with option 2).
> > > My problem with option 3) is that which tests would we choose? If you
> > have
> > > a good subset of tests already, than this option also works for me :)
> > >
> > > Regards,
> > > Norbert
> > >
> > > On Mon, Jan 14, 2019 at 11:06 PM Enrico Olivelli <eolivelli@gmail.com>
> > > wrote:
> > >
> > > > Last update:with fork count = 4 we have the build finishing wihin 20
> > > > minutes but we have a lot of test failures !
> > > >
> > > > https://travis-ci.org/apache/zookeeper/builds/479607886
> > > >
> > > > Enrico
> > > >
> > > > Il giorno lun 14 gen 2019 alle ore 22:24 Enrico Olivelli
> > > > <eolivelli@gmail.com> ha scritto:
> > > > >
> > > > > I am sorry, we have two problems:
> > > > > 1) the build has some tests which is failing (flakes??)
> > > > > 2) the build takes too much time, we have an hard limit on Travis
> of
> > > > > max 50 minutes
> > > > >
> > > > > see this example:
> > > > >
> > > >
> >
> https://travis-ci.org/apache/zookeeper/builds/479429591?utm_source=github_status&utm_medium=notification
> > > > >
> > > > > I see these ways:
> > > > > 1) Say good bye to Travis
> > > > > 2) Limit Travis build to "only" compile/spotbugs/rat...
> > > > > 3) Limit Travis to only a subset of tests (we can tag tests with
> > JUnit
> > > > > Categories for instance)
> > > > >
> > > > > I don't know if there is some way to speed up the build, in the
> > > > > configuration file I have disabled 'fsyncs' and set forkcount = 1.
> > > > > I don't know how many tests are missing in the job logs, maybe we
> are
> > > > > close to the end of the suite.
> > > > >
> > > > >
> > > > > Enrico
> > > > >
> > > > >
> > > > > Il giorno lun 14 gen 2019 alle ore 14:16 Enrico Olivelli
> > > > > <eolivelli@gmail.com> ha scritto:
> > > > > >
> > > > > > First Travis build:
> > > > > > https://github.com/apache/zookeeper/pull/751
> > > > > >
> > > > > > Il giorno mar 8 gen 2019 alle ore 13:47 Enrico Olivelli
> > > > > > <eolivelli@gmail.com> ha scritto:
> > > > > > >
> > > > > > > Il giorno mar 8 gen 2019 alle ore 11:05 Enrico Olivelli
> > > > > > > <eolivelli@gmail.com> ha scritto:
> > > > > > > >
> > > > > > > > Il giorno mar 8 gen 2019 alle ore 11:02 Andor Molnar
> > > > > > > > <andor@apache.org> ha scritto:
> > > > > > > > >
> > > > > > > > > Sure, let’s try it. Please go ahead and enable
it.
> > > > > > >
> > > > > > > INFRA ticket:
> > > > > > > https://issues.apache.org/jira/browse/INFRA-17584
> > > > > > >
> > > > > > > Cheers
> > > > > > > Enrico
> > > > > > >
> > > > > > > > >
> > > > > > > > > What are the benefits over Jenkins?
> > > > > > > >
> > > > > > > > 1) non committers can rebuild easily just by using
the
> > Travis-UI
> > > > > > > > 2) easy to change the configuration, it is committed
on the
> > branch
> > > > to
> > > > > > > > test (merged with target branch)
> > > > > > > > 3) less security issues on ASF side, pull requests
built from
> > > > unknown
> > > > > > > > constributors are built on Travis hardware, not ASF
one
> > > > > > > > 4) easy to add new Java versions
> > > > > > > >
> > > > > > > > Enrico
> > > > > > > >
> > > > > > > >
> > > > > > > > >
> > > > > > > > > Andor
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > > On 2019. Jan 8., at 7:28, Enrico Olivelli
<
> > eolivelli@gmail.com>
> > > > wrote:
> > > > > > > > > >
> > > > > > > > > > Shall we ask infra to enable Travis?
> > > > > > > > > > This way we will see it working on my PR
> > > > > > > > > >
> > > > > > > > > > I can create the INFRA ticket,  with the
permission of
> the
> > > > PMC/committers
> > > > > > > > > >
> > > > > > > > > > Enrico
> > > > > > > > > >
> > > > > > > > > > Il mer 2 gen 2019, 16:16 Enrico Olivelli
<
> > eolivelli@gmail.com>
> > > > ha scritto:
> > > > > > > > > >
> > > > > > > > > >> very soon:
> > > > > > > > > >> https://github.com/apache/zookeeper/pull/751
> > > > > > > > > >>
> > > > > > > > > >> I have pushed the script I am using
in other projects,
> > with
> > > > tests on
> > > > > > > > > >> jdk11 + linux, default profile.
> > > > > > > > > >>
> > > > > > > > > >> We should ask INFRA in order to enable
Travis-CI, this
> is
> > > > kind of
> > > > > > > > > >> important infrastructural change
> > > > > > > > > >> so I think some PMC should sponsor this
and approve
> > > > > > > > > >>
> > > > > > > > > >> Once we have Travis, we can enable https://coveralls.io
> > Code
> > > > Coverage
> > > > > > > > > >> for free, all of them is well integrated
with GitHub.
> > > > > > > > > >> I am doing the same during this days
with on of the open
> > > > source
> > > > > > > > > >> projects from my company (
> > > > https://github.com/diennea/blazingcache) and
> > > > > > > > > >> it works like a charm.
> > > > > > > > > >>
> > > > > > > > > >> I had set up Coveralls+Jenkins for Apache
BookKeeper, it
> > > > needs an
> > > > > > > > > >> INFRA step because of the management
of a shared secret,
> > but
> > > > with
> > > > > > > > > >> Travis+ Coveralls everything is automatic.
> > > > > > > > > >>
> > > > > > > > > >>
> > > > > > > > > >>
> > > > > > > > > >> Enrico
> > > > > > > > > >>
> > > > > > > > > >> Il giorno mer 2 gen 2019 alle ore 16:09
Enrico Olivelli
> > > > > > > > > >> <eolivelli@gmail.com> ha scritto:
> > > > > > > > > >>>
> > > > > > > > > >>> I will sent a PR soon
> > > > > > > > > >>>
> > > > > > > > > >>> this is the Ticket
> > > > > > > > > >>> https://issues.apache.org/jira/browse/ZOOKEEPER-3234
> > > > > > > > > >>>
> > > > > > > > > >>> Enrico
> > > > > > > > > >>>
> > > > > > > > > >>> Il giorno mer 2 gen 2019 alle ore
15:48 Andor Molnar
> > > > > > > > > >>> <andor@cloudera.com.invalid>
ha scritto:
> > > > > > > > > >>>>
> > > > > > > > > >>>> +1
> > > > > > > > > >>>>
> > > > > > > > > >>>>
> > > > > > > > > >>>> On Sat, Dec 22, 2018 at 2:48
PM Enrico Olivelli <
> > > > eolivelli@gmail.com>
> > > > > > > > > >> wrote:
> > > > > > > > > >>>>
> > > > > > > > > >>>>> Hi,
> > > > > > > > > >>>>> Now that we are Mavenizing
ZookKeeper it will be
> super
> > > > easy to add
> > > > > > > > > >>>>> Travis-CI to the story.
> > > > > > > > > >>>>>
> > > > > > > > > >>>>> It will enable non Apache
Committers to have more
> > access
> > > > to logs of
> > > > > > > > > >>>>> their patches.
> > > > > > > > > >>>>>
> > > > > > > > > >>>>> I am thinking about Travis
only for Pull Requests.
> > > > > > > > > >>>>> It is super easy and ZooKeeper
build is not super
> > heavy.
> > > > > > > > > >>>>>
> > > > > > > > > >>>>> I would suggest to run tests
with a single degree of
> > > > parallelism
> > > > > > > > > >>>>>
> > > > > > > > > >>>>> Thoughts ?
> > > > > > > > > >>>>> Enrico
> > > > > > > > > >>>>>
> > > > > > > > > >>
> > > > > > > > > > --
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > -- Enrico Olivelli
> > > > > > > > >
> > > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message