Rework/improve Koji capacity considerations

[mizdebsk]

From historical data Koschei should know which package builds on which architectures (noarch, all arches, or only some subset). This data could be used to better utilize available capacity. This became more important after addition of secondary arches to Fedora primary Koji.

For example, when we exceed ppc64 builder capacity, but still have x86_64 capacity, we can keep submitting noarch builds.

[mizdebsk]

@msimacek WDYT about solution like in PoC commit?

[msimacek]

Looks good to me (besides the mentioned missing call to all).

But the "keep submitting" is a bit misleading - if let's say x86_64 load is over limit, but ppc64 is not, it would submit a noarch build only if the build with highest priority is noarch. I assume it's intentional to ensure fairness.

Could you please add a simple testcase for it? (note I merged your VCR code some time ago, so you can use that if you want)

[mizdebsk]

But the "keep submitting" is a bit misleading - if let's say x86_64 load is over limit, but ppc64 is not, it would submit a noarch build only if the build with highest priority is noarch. I assume it's intentional to ensure fairness.

This is intentional for two reasons:

• to prevent priority inversion (or as you said, "ensure fairness") - otherwise high-priority packages could be preemptied by lower-priority ones
• because we don't have caching of Koji queries - I didn't want to implement this yet, and repeating Koji RPC calls for next builds seemed like bad idea

Could you please add a simple testcase for it? (note I merged your VCR code some time ago, so you can use that if you want)

Yes, sure. I'll add test cases. Thx for merging VCR code!

[msimacek]

Merged, thank you.