CI: Micromamba taking too long to resolve the environments in the CI

[datapythonista]

Our CI jobs are frequently failing now as they timeout after 90 minutes of execution. Of those 90 minutes, 25 are spent on micromamba resolving the environment.

In the past we have fixed this by limiting the number of packages to be considered. For example, if the environment just says numpy, maybe there are 200 versions that will be considered. While if we say numpy >= 2 the number can be limited to few.

I'm not sure which packages have lots of options, and we don't want to filter out the versions that make sense to install. But we should have a look and see if by adding few constraints we can get a reasonable time to solve the environment.

[microslaw]

take

[datapythonista]

The failing CI jobs is exactly what we are trying to fix. I'll close the PR for now, as it doesn't really help as it is. The dependencies that need to be fixed are not our development environment (not sure how long it takes, but not 25 minutes the last time I installed it). It's the dependencies for the CI in ci/deps.

Before opening a PR it would be good to share the analysis on what we're planning to do and why, so we can discuss what to pin exactly.

[microslaw]

On a closing note, I've tried to benchmark micromamba on every environment in ci/deps. Nothing came even close to the 25 minutes mentioned, on most environments it finished within two minutes, with few 7-ish minute outliers. Feel free to see the results here

[datapythonista]

Interesting. Looks like it could be the Python 3.13 jobs only, and in some particular cases. I saw the jobs timing out often, but I didn't see in some cases they run fast.

You can see a very slow case here: https://github.com/pandas-dev/pandas/actions/runs/15379937435/job/43269355646?pr=61533

Maybe a bug in micromamba in Python 3.13. If it happens only sometimes, it'll be difficult to find the problem, but maybe we can check in the micromamba repo if there is an issue for this, and open one if it's not.

[microslaw]

I looked through their issues, didn't see anything related, I'm not sure if this is micromamba's fault. I will try to reproduce and pinpoint it, so that we'll have something more concrete, unless the issue proves too rare/difficult

[datapythonista]

I opened mamba-org/mamba#3970