Try to use old mamba version

[mathbunnyru]

No description provided.

[mathbunnyru]

My approach should work, but we have a much bigger problem.

https://github.com/jupyter/docker-stacks/runs/4493047317?check_suite_focus=true#step:7:4475

#5 [1/4] FROM docker.io/jupyter/minimal-notebook@sha256:5d8ba694b92d9fe5802529b7ccf8bad23de6632d0bbaa92fc3fdc8a31cdc9c9c
4476
  #5 resolve docker.io/jupyter/minimal-notebook@sha256:5d8ba694b92d9fe5802529b7ccf8bad23de6632d0bbaa92fc3fdc8a31cdc9c9c done
4477
  #5 sha256:2f66eefcb707429dfa3d1b8d0cdcbf330bb3d2624db2821eb2eb2e7323e0358a 1.48kB / 1.48kB 0.0s done
4478
  #5 sha256:fd44b7e34afa5f14902e0239880773be42ca449f5d1dbab726b4a17cde4f48f4 41.11MB / 235.96MB 0.3s`

Because we build so many images, they take too much space and we end up pulling images from dockerhub.
Which means we're effectively building not the code which is at the current branch, but some mix of pas image and the new code.

[mathbunnyru]

@consideRatio I will for now completely disable arm images building at all :(
I think the approach "build everything with some hacks in one place" doesn't work well for us.

I will try to increase the volume using GitHub action for that, but I do think it's a fragile approach as well.

[mathbunnyru]

Nice, the only resolve docker.io I see are ubuntu:focal images.
Let's get mamba pinning back, but this time I will only pin it for arm images.

[consideRatio]

@consideRatio I will for now completely disable arm images building at all :(
I think the approach "build everything with some hacks in one place" doesn't work well for us.

That sounds reasonable. Given the limited maintenance capacity we have, and that this blocks everything if not maintained, cutting away something to save something else seems reasonable.

I think this strategy hasn't been robust enough given that each failure cause a lot of additional work and is hard to resolve due to the monolith CI jobs. I consider a resolution towards having arm again after disabling it would be to put in more work in making the CI system work in a more composable manner (#1407), I think that would make it handle multiple architectures in a more maintenance sustainable way.

---

Are you making some last attempts to get things to work atm?

[mathbunnyru]

Yes, let's wait for the latest build - it should run faster (because caching images works) and it gives expected result (because of caching).
It might be the issue with latest mamba, but it might be something else :(

[mathbunnyru]

Finally, this seems to work.
Let's merge this and hope it will work during push step.

[consideRatio]

@mathbunnyru it worked! You are amazing! Thank you for your thorough work!!! ❤️ 🎉