Re: Renamed .gitlab-ci.yml to disable CI builds (at least for now)

[Wouter Verhelst <w@uter.be>]

Hi,

On Tue, Nov 27, 2018 at 11:00:34AM +0000, Damyan Ivanov wrote:
> -=| Wouter Verhelst, 27.11.2018 08:20:08 +0100 |=-
> > Hi Laura,
> > 
> > On Sun, Nov 25, 2018 at 09:27:19PM +0100, Laura Arjona Reina wrote:
> > > Hello all
> > > We've been told that the webwml CI builds in Salsa use too much
> > > resources, affecting Salsa's ability to serve other users.
> > 
> > That's not good, obviously.
> > 
> > How are we overdoing it? E.g., are we using too many runners, or are we
> > using too much space for artifacts?
> 
> Too many runners, I think. That's from #alioth:
> 
>     webwml starts about 20 builds in parallel, 20 times docker pull of 
>     a large image, 20 times git pull of the not too small repo, this 
>     produces a lot of load
> 
>     i asked them to disable the builds, as the result of those 
>     builds are not used, not even to check for errors. and 
>     breaking the concurrent build restriction to do nothing is 
>     a bit bold
> 
> HTH to pinpoint the issue.

It does, thanks.

The reason we currently do it that way is that, while I know you can
tell buildbot to skip certain builds if they are not necessary based on
certain criteria, I couldn't find an obvious way to make "were files
under a certain directory touched in the most recent push" be part of
those criteria. I did add a small shell test to the start of every
translation that skips the build if the current translation did not see
any changes in the push for which we're building, but didn't consider
the fact that even the setup for running that really small script is
significant, too.

I'll go through the documentation and see if I can find a way to make
that work better, so that translation jobs aren't even started if they
don't need to do anything.

Regards,

-- 
To the thief who stole my anti-depressants: I hope you're happy

  -- seen somewhere on the Internet on a photo of a billboard