To make a long story short: I prefer * one source-checkout per slave * ability to kick in more slaves per builder category, i.e., making windows installers * failing slaves shouldn't make the build fail That means: We have to lock down the source for all builds on one slave for all repacks. Checking out the source needs to be optional, and not all slaves need to succeed. That basically rules out Dependent schedulers and slave locks, given that with l10n repacks, we need a builder per language and platform to get some scheduling freedom. I am currently hacking up a solution to this: A scheduler that blocks new builds from starting until all builds have completed. A OneSlaveLock, which actually triggers a callback on all but the first owner to take that lock, enabling us to switch off check-out on following steps. Code coming up as soon as it passes self-review.
First of trac issues blocking this would be http://buildbot.net/trac/ticket/36.
(In reply to comment #1) > First of trac issues blocking this would be http://buildbot.net/trac/ticket/36. > That's closed.
Let's resolve this in one way or another, picking INCOMPLETE as the initial comment in this bug doesn't map reality anymore.