Why is Conda so much slower with yml file?

Question

I am trying to re-create an environment across different OS. Using conda env export --from-history, I created the following yml file:

name: mlflow_env
channels:
  - conda-forge
dependencies:
  - numpy
  - lightgbm
  - squarify
  - holidays
  - lunardate
  - lunarcalendar
  - sqlalchemy
  - pandas
  - geopy
  - pgeocode
  - psycopg2
  - cython
  - scikit-learn
  - mlflow
  - pyarrow

Once I run conda env create --file conda_lt.yml python=3.9 on any of my two systems, it takes hours to solve the environment, which it eventually solves.

However if I try to run conda create -c conda-forge -n mlflow_env numpy lightgbm squarify holidays lunardate lunarcalendar sqlalchemy pandas geopy pgeocode psycopg2 cython scikit-learn mlflow pyarrow, solving the environment takes less than 30 seconds.

In my understanding the two commands should really result in the same environment, so I am wondering what is the difference and why the yml installation is so much slower. Naturally, any suggestion to make the yml installation as quick as the other is welcomed.

Answer 1

You can try this:

conda update conda
conda update conda-build
conda install -n base conda-libmamba-solver
conda config --set solver libmamba

My answer is based on the issue raised in Conda's issue tracker.