[Vm-dev] Errored: OpenSmalltalk/opensmalltalk-vm#1258 (Cog - 0ce1378)

[Nicolas Cellier]

2018-04-04 6:58 GMT+02:00 Ben Coman <btc at openinworld.com>:

>
> On 4 April 2018 at 06:52, Travis CI <builds at travis-ci.org> wrote:
> >
> >
> > OpenSmalltalk / opensmalltalk-vm
> >
> > Cog
> >
> > Build #1258 has errored
> >
> > 2 hours, 29 minutes, and 57 seconds
> >
> > Eliot Miranda
> >
> > 0ce1378 CHANGESET →
> >
> > Nuke obsolete files that uploaded VMs to mirandabanda.org
>
> I notice https://travis-ci.org/OpenSmalltalk/opensmalltalk-
> vm/builds/361820486
> shows the failures due to Raspberry Pi are
>
>  1258.38   16 min   Failed in travis_install.sh
>  1258.39   49 min   Failed in travis-build.sh = The job exceeded the
> maximum time limit for jobs, and has been terminated.
>  1258.40   16 min   Failed in travis_install.sh
>  1258.41   16 min   Failed in travis_install.sh
>  1258.42   49 min   Failed in travis_build.sh = The job exceeded the
> maximum time limit for jobs, and has been terminated.
>  1258.43   45 min   Success,   travis_install.sh time = 22 minutes,
> travis_build.sh = 21 minutes
>
> So over half the builds failed directly due to installing from remote site.
> The successful job shows over half the time taken with the install, so
> this indirectly causes the two timeout failures.
> The install says... "After this operation, 427 MB of additional disk
> space will be used".
>
> So I wonder if it wold be useful to have an install-only stage that
> tar's itself up to carry forward as an artifact
> for a subsequent build-only job to untar & chroot into that artifact ??
> This might save 20 minutes per build-only job and give the build more
> headroom to succeed.
>
> Also the install-only stage might have some magic to detect a stalled
> install before Travis does, and restart it.
> And at least it will be very obvious that install failed rather than a
> build.
>
> cheers -ben
>

Yes, those timeouts sound like the source of our problems.
I understand that build worker machines are linux amd64 and that we use
sort of cross-compilation -march=armv6 to produce our artifacts, so amd64
messages seems normal, though we go probably thru tons of useless stuff.

If we can accelerate production of arm builds that's all good, they are
dead slow...
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20180404/a3dd7906/attachment.html>