[Buildbot-devel] getting buildslaves to exit after one build?

Dan Kegel dank at kegel.com
Tue Aug 28 17:14:43 UTC 2012


On Tue, Aug 28, 2012 at 10:03 AM, Tom Prince <tom.prince at ualberta.net> wrote:
>> As I explained at the start of the thread, I think need to use plain old slaves.
>
>> Latent slaves run on the master,
>
> as does the code snippet you posted

What I mean is, with latent slaves, buildbot master is responsible
for starting the slaves.  This requires remote access.  Remote access
is harder to set up and debug IMHO than just scripting the slave
outside of buildbot to start a container in a loop.  Somebody
in my group can easily see what the slave is doing when I
hardcode that loop in shell, but would be harder pressed to
trace through the buildbot machinery if I used latent slaves.

I suppose if I really understood buildbot and libvirt's code, I might
feel differently, and would feel strongly that doing things the Right
Way with buildbot and libvirt was worth the effort.
Sadly, I don't really understand buildbot's internals, let along libvirt's,
but am comfortable using buildbot with traditional buildslaves,
and as a result I'm going for the quick hack that lets me treat
buildbot as much like a black box as possible.

Does this make me lame?  Yes.  But I have my hands full,
and fully understanding buildbot is not in the cards for me at the moment.
(If you haven't guessed by now, I'm the kind of guy who has
hand-crank windows in his car because he doesn't trust power windows.)

> I think it is probably a reasonable use case (where you want to stop the
> slave automatically, but it will come up on its own). I suspect that an
> empty start_instance, and the stop_instance I posted will likely do the
> right thing (more or less). ... Actually, it looks like the code will
> reject latent slaves that try connecting when they haven't been asked
> to, so this becomes slightly more complicated.

Yeah, and it's complication I'd like to avoid for the moment.
I will probably limp along with my patch until it becomes evident that it's
too broken, at which point I might have time to try to understand libvirt
and latent slaves and do things 'right'.
- Dan




More information about the devel mailing list