Cayetano Santos <[email protected]> writes: >>mar. 10 nov. 2026 at 10:46, Christopher Baines <[email protected]> wrote: > >> Untrusem <[email protected]> writes: >> >>> So I came across this article: "How Much Energy Does Building Distro >>> Packages Use" [0] >>> >>> And this made me think "How much energy Guix's infra would be using in >>> building packages?", "How can we calculate/monitor it". >>> >>> I don't think our infra is self-hosted or using solar (maybe in the >>> future). So how one would be able generate some data on this? > >> One thing that bothers me more than the energy usage is the waste, > > I think this is the key point here: non optimal usage of computing > resources, which also impacts time (to merge branches, to build pr, etc.) > > For example, the fact that ci builds (every force push of) pull > requests, which eventually get merged in master, ... and build again > from scratch.
There are some non-trivial things to consider when setting this up, but I think there was a viable setup back when the QA Frontpage was doing this with the bordeaux build farm. I'm working on bringing that back for Codeberg, but it's slow going https://codeberg.org/guix/qa-frontpage/issues/10 >> there's plenty of builds that run for a long time and always fail, so >> more quickly identifying and fixing/removing these packages would be >> good. There's also other efficiency issues like the data service ending >> up building stuff for other systems, using expensive emulation and often >> the builds fail, but get retried every time a revision is processed >> https://codeberg.org/guix/data-service/issues/25 > > Other than having a quick means to follow broken packages on master (I’m > thinking here on a dedicated mail list or rss feed), I see no other way > to improve current situation. Personally I've always seen rigerously testing changes (Pull Requests) as the important thing, once that's in place, you can also check for existing issues. >> We've also got in to a state recently where changes to packages are >> being made inefficiently, just see the recent example of the changes on >> the mesa-updates, gnome-team and python-team branches. Each set of >> changes affects more than half the packages in Guix, and there's a lot >> of overlap since they all affect python. By separating out the changes >> this way, the build farms are doing a lot more work, plus users end up >> also building/downloading more over and over again. It's a balance, but >> I think we need to do better to make changes efficiently as well as at >> pace. > > You mean that, for example, a monthly world rebuild branch, all teams > together, would be more efficient from a computing perspective ? I don't know where this coupling of branches and teams has come from, the package graph is far more important in my mind, and at the moment that isn't structured around "teams" (although we can move in that direction). But the cadence aspect (e.g. monthly) is important, as the optimal position is to be pacing changes such that users get important changes (e.g. security and other fixes) quickly, and other things efficiently. I've never used an "error budget" system, but I think something like that might be better than a fixed schedule. Measure what effect changes have and then plan around that.
signature.asc
Description: PGP signature
