So, I meant to reply to that, but I have been trying to only focus on mastodon and ignore twitter doom scrolling.

The reason why chromium builds started taking a long time again was that I had added some normal builders to the heavybuilder channel during the mass rebuild in order to get some things moving again.

So, they were still in there. I removed them yesterday, so now the aarch64 builders in the 'heavybuilder' channel that chromium uses are really our two heavy builders.

Our two aarch64 heavybuilders are cavium thunder x2's with 225CPUs and 256GB memory. Amazon does not offer any aarch64 instances with more than 128cpus, although you can get a bunch more memory.

So, things should be back to better for chromium, and aws doesn't help us here. :)

Might it still be useful to look at what we can take advantage of for mass rebuilds, Koschei, etc., beyond the specific chromium-is-a-beast problem?

Well, in mass rebuilds, we could use a bunch more s390x instances... if aws had those I would be all over it. ;)

koschei is designed to use the otherwise idle builders that exist.

I think there was a project/something to get koji more cloud aware a few years ago, but I never heard what happened to it. Perhaps we should ask upstream koji about it...