[Qgis-developer] CI testing: the straight truth

[Matthias Kuhn]

Hi Andreas, Nyall

I would also prefer this, I just sent them a mail and will get back to
you as soon as I get a response.

Mid-term we will still need to launch a project to update the system and
get away from the pre-built dependencies in osgeo4travis to a system
that is easier to maintain collectively but I'm happy if we can wait
with this a bit.

@Tom Chadwin:
agreed, that's a risk indeed

Best regards
Matthias

On 01/04/2017 08:21 AM, Neumann, Andreas wrote:
> Hi all,
> 
> I agree with Nyall. We should try contracting the travis ci team and ask
> them if we could switch to unlimited build time. Even if this means
> paying a certain amount for the services (if it is a reasonable price).
> 
> Matthias - could you please take a lead in contacting them?
> 
> Thanks,
> Andreas
> 
> On 2017-01-03 23:40, Nyall Dawson wrote:
> 
>> On 4 January 2017 at 07:59, Matthias Kuhn <matthias at opengis.ch
>> <mailto:matthias at opengis.ch>> wrote:
>>> Hi René-Luc, hi all,
>>>
>>> This is something I have been afraid of for a couple of weeks.
>>>
>>> Short story: our builds just grew too big and a full build is taking
>>> more than 50 minutes to complete on the travis infrastructure.
>>
>> Argh... what a pain! Still, I think in some ways this is a good
>> reflection of the exponential growth in QGIS unit tests since your
>> initial work introducing the CI infrastructure. I certainly would
>> never want to go back to the pre-CI days!
>>
>> Just wondering- is there any chance sending an email to the Travis
>> crew could get an extension to this build time? I can't see anything
>> on their publicised plans, but perhaps they have a cheaper unlimited
>> build time option available for open source projects?
>>
>> I'd like to see us exhaust these "easier" options first before
>> requiring someone to donate time into tweaking/changing the CI
>> infrastructure.
>>
>> Nyall
>>
>>
>>> What
>>> keeps us alive at the moment is the persistent ccache so a full build
>>> actually never happens. If the ccache gets lost, we will need some black
>>> magic to get the builds started again (it's possible to slowly get the
>>> cache warm but not straightforward).
>>>
>>> What we could possibly do:
>>>
>>>  - Modularize the build: e.g. astyle could very well be moved to a
>>>    separate job along with other static checks like spelling. That would
>>>    even have the advantage of a faster feedback for these analyses.
>>>    But I'm not sure how much time we can actually get out of that.
>>>
>>>  - Move dependencies into separate packages. We have some deps like
>>>    qspatialite that are built as part of QGIS which I think could be
>>>    installed as dependencies from a .deb. This will require quite a bit
>>>    of work: moving travis out of the container based infrastructure to
>>>    the sudo-enabled infrastructure, setting up a repo (e.g. ppa) with
>>>    all sort of dependencies including qt 5 etc. This would be an option
>>>    again since travis enabled caching also on sudo systems recently.
>>>    This will be a quite large task to do.
>>>
>>>  - Move to another system (managed infrastructure or self-hosted, there
>>>    are a couple of services like circle ci, drone.io, jenkins, gitlab
>>>    ci ...).
>>>    This will be a quite large task to do with the risk that we run into
>>>    yet another timeout, other technical issue or run a self-hosted
>>>    infrastructure for which nobody really has the resources to maintain.
>>>
>>> Apart from this, there is also the dependency on some pre-compiled
>>> libraries in the osgeo4a repository. Keeping these up to date is not
>>> something we will want in the long term (I was hoping that travis would
>>> ship a more recent distro than trusty but haven't seen many hints in
>>> this direction yet).
>>>
>>> Bottomline: I don't think there's an easy fix but I think it's time to
>>> start thinking about the future.
>>>
>>> Matthias
>>>
>>> On 01/02/2017 03:03 PM, René-Luc Dhont wrote:
>>>> Hi Devs,
>>>>
>>>> I would like to merge a PR https://github.com/qgis/QGIS/pull/3897 but
>>>> Travis cannot ccomplete the tests.
>>>>
>>>> Would it be simple to fix travis ?
>>>>
>>>> Regards,
>>>> René-Luc
>>>> _______________________________________________
>>>> Qgis-developer mailing list
>>>> Qgis-developer at lists.osgeo.org <mailto:Qgis-developer at lists.osgeo.org>
>>>> List info: http://lists.osgeo.org/mailman/listinfo/qgis-developer
>>>> Unsubscribe: http://lists.osgeo.org/mailman/listinfo/qgis-developer
>>> _______________________________________________
>>> Qgis-developer mailing list
>>> Qgis-developer at lists.osgeo.org <mailto:Qgis-developer at lists.osgeo.org>
>>> List info: http://lists.osgeo.org/mailman/listinfo/qgis-developer
>>> Unsubscribe: http://lists.osgeo.org/mailman/listinfo/qgis-developer
>> _______________________________________________
>> Qgis-developer mailing list
>> Qgis-developer at lists.osgeo.org <mailto:Qgis-developer at lists.osgeo.org>
>> List info: http://lists.osgeo.org/mailman/listinfo/qgis-developer
>> Unsubscribe: http://lists.osgeo.org/mailman/listinfo/qgis-developer
> 
>  
> 
>