On Sun, Nov 15, 2009 at 9:06 AM, Lieven Govaerts <svnlgo_at_mobsol.be> wrote:
> while we're migrating mailing lists and the svn repo to the ASF
> infrastructure, let's discuss what we want to do with the buildbot.
> The ASF provides 4 different build services (see
> http://ci.apache.org/) and Buildbot (the one we use) is one of them.
> Based on a long chat with Gavin McDonald from ASF infra on IRC last
> week, I propose to integrate our Buildbot master into the ASF's. This
> has some (obvious) benefits:
> - although our buildbot master has been stable since we migrated it to
> Justin's Mac, I assume the ASF infrastructure team can do a better job
> of hosting and maintaining such setup in a secure environment.
+1 in general
> - ASF's buildbot already has Ubuntu and Windows server 2008 slaves
> (and according to Gavin also Windows XP 64 bit and Solaris) and there
> are plans to add more. There's no Mac yet, but we have one in our
In the case of Windows, do they also have Visual Studio and all the
extras needed to actually build SVN?
> - it has more features (http://ci.apache.org/buildbot.html), and -
> more importantly - the maintainers are open (and have time) to improve
> the setup based on our requests.
> Sticking with buildbot allows us to keep the existing buildbot slaves,
> if the owners choose to do so. Also, out of the 4 available options it
> seems best fitted to our use case (C code, custom test scripts and
> builds on multiple platforms).
Given our familiarity with it, I think it makes sense to stick with
buildbot. That said, I like Hudson and we have done some work
recently so that our build and test can integrate nicely. For
example, Hudson can display the test results a lot more nicely and it
graphs things like the execution time that can help spot trends. I
personally prefer buildbots red/green model over Hudson's "weather
> There are some things that aren't really clear yet:
> - the ASF buildbot is shared for 20+ projects. I know slaves can be
> reserved for certain projects, but what's the impact of building
> projectA first, and then build subversion on the same slave.
> - do we use slave's system libraries, or do we need to build a
> specific set of versions of libraries used only for the subversion
> - the waterfall view is shared between all projects and buiders
> (http://ci.apache.org/waterfall). Is that okay for us? Or should we
> work on a Subversion specific view.
> - if there are project + slave specific build problems, what are our
> options to get the needed logs/info and fix the issues?
I really wonder about all these things as well. Not to mention just
that the long build and test time for Subversion is different than a
lot of the typical Java projects that can do all this in a few
minutes. Our buildbots cannot even keep up with our activity, how
well will this work in a massively shared model where there are a lot
of projects all on the same master?
> So, thoughts? Remarks?
With everything going on, my only comment would be to slow down. I do
not see any reason to rush into this (too many changes happening at
Received on 2009-11-15 17:46:59 CET