> Well, there's certainly room for more than one master. Ideally, you'd > want a separate master for each different thing that's being built: > > 1) Stable tree. > 2) Development tree. > *) Each client.
Yeah, was figuring we were going to have to split it up. In our case we'd have to designate one person's branch as "next" and have the development tree buildbot point there. Clients... also useful. Think brian has/had a buildbot for libmemcached? > Slaves can attach to these to masters to ensure portability and stuff. > I like to use python's buildbot setup as an example: > > http://www.python.org/dev/buildbot/ Neat. Sorry, I hate buildbot's organization, but I'm accepting that this is what we need to do. > Anyone who would want to run a slave would just need to install the > prerequisites and ask for slave credentials on the master. Of course, > in order to be useful, one would have to be offering something different. Sounds good. I definitely started this for the stable tree but need to follow up with more people to ensure their slaves are running. Have to get an e-mail notification when a slave's been offline for more than X hours and deal with it. I'll still offer to host buildmasters, by the by. My box is basically idle for development and it's trivial to host buildmasters. Can give shells, whatever. Buildbot runs on there already and I'll back up the data on the host. -Dormando
