Now I'd like to check what was reason for rubygem-dalli failures 1 which happened previously to possibly confirm the build failure fix, but I can't because the logs are gone :/

The reason why builds are pruned so aggressively is the amount of storage used for each build. We don't have the ability to specify the retention by pass/fail. That would need a koji enhancement.

Several points

1. I am asking to keep (root and build) logs, not anything else
2. I care about successful as well as failed builds. The build might fail for different reasons and I want to compare them to successful builds.
3. Its strange that regular scratch build I typically don't care about more then a few minutes are kept around way longer (2 weeks?) then the logs of Koschei.
4. And finally, I understand that there are storage/budget constraints, on the other hand, this is the most useful service of Fedora Infrastructure, so it'd be nice to hear "yes, we try to organize budget for Koschei next year" or so. Not sure if anybody is kicking for this or it is just handwavy "its just Koschei, we can delete build output more aggressively". Yes, Koschei could keep the logs in its own storage (I can open RFE for this), but not sure if that would change anything.

1) it's all or nothing
2) successful builds take up even more space due to the binaries being kept too
3) it's different policies, the reason the koschei are kept less is there's a lot more of them and take an order of magnitude more space. You might not care about scratch, others do.
4) the usefulness is a matter of opinion. And while I have no opinion either way there are others where their opinions are completely the opposite of yours.

4) the usefulness is a matter of opinion. And while I have no opinion either way there are others where their opinions are completely the opposite of yours.

Certainly nobody from devels, although they can disable Koschei if they wish ... But I can imagine that the resources and maintenance might be PITA for somebody. But in total, the positives for community outweighs the negatives IMO.

I actually already filed this as: https://pagure.io/releng/issue/130

I think 1 day is pretty useless and we should move it to say 7 days. (1/2 of what normal scratch builds have at 14 days), but hopefully we can reach some consensus.