Comments on: Builds, shuttle busses and cabs http://blog.mozilla.com/axel/2008/02/07/builds-shuttle-busses-and-cabs/ Free your mind and your ass will follow. Thu, 12 Nov 2009 18:58:01 -0800 http://wordpress.org/?v=2.8.6 hourly 1 By: Axel Hecht http://blog.mozilla.com/axel/2008/02/07/builds-shuttle-busses-and-cabs/comment-page-1/#comment-887 Axel Hecht Fri, 08 Feb 2008 15:51:55 +0000 http://blog.mozilla.com/axel/2008/02/07/builds-shuttle-busses-and-cabs/#comment-887 I somehow don't believe that we should have enough slaves to catch up with Reed's check-in parties :-) I'm really not sure if it's cost-effective to beef up the infrastructure for the maximum traffic, and even if we do, going to the limit of that should be part of the plan. I think the treat is to not blend patches, yes. The bigger treat would be to not compile the non-merged source stamps in chronological order, but latest-first, and to put older source stamps onto some kind of idle scheduler. Wow, that'd be horrible for a waterfall display, but there's probably a good chance to find something to make up for that. I somehow don’t believe that we should have enough slaves to catch up with Reed’s check-in parties :-)

I’m really not sure if it’s cost-effective to beef up the infrastructure for the maximum traffic, and even if we do, going to the limit of that should be part of the plan.

I think the treat is to not blend patches, yes. The bigger treat would be to not compile the non-merged source stamps in chronological order, but latest-first, and to put older source stamps onto some kind of idle scheduler.

Wow, that’d be horrible for a waterfall display, but there’s probably a good chance to find something to make up for that.

]]> By: Rob Helmer http://blog.mozilla.com/axel/2008/02/07/builds-shuttle-busses-and-cabs/comment-page-1/#comment-886 Rob Helmer Fri, 08 Feb 2008 02:45:52 +0000 http://blog.mozilla.com/axel/2008/02/07/builds-shuttle-busses-and-cabs/#comment-886 So what doing one build/test cycle per checkin, and not "blending" patches? Pretty sure we have the capacity to allocate a set of slaves to every single checkin we get, without making anyone wait in line. This is how the tryserver works now in fact, but we haven't given it many slaves and we're not using buildbot's Source class (it can be told that checkins are unmergable for instance) so we could do a cleaner solution. It seems like this would enable the kind of things you are considering in the rest of the post, no? So what doing one build/test cycle per checkin, and not “blending” patches?

Pretty sure we have the capacity to allocate a set of slaves to every single checkin we get, without making anyone wait in line.

This is how the tryserver works now in fact, but we haven’t given it many slaves and we’re not using buildbot’s Source class (it can be told that checkins are unmergable for instance) so we could do a cleaner solution.

It seems like this would enable the kind of things you are considering in the rest of the post, no?

]]>