Local Conda environments via Apptainer

The nice thing about this approach is that you don't need to install Conda and you don't need to modify your environment/bashrc/settings.

I use it to install dependencies that may be tough to install on a supercomputer or on my NixOS environment.

How to fetch the image:

$ apptainer pull https://github.com/bast/apptainer-conda/releases/download/0.6.0/conda.sif

Usage

• Reads the Conda environment file environment.yml.
• Creates the folder environment.

Run myscript.py inside the Conda environment defined by environment.yml:

$ ./conda.sif python myscript.py

Open Python shell inside the Conda environment defined by environment.yml:

$ ./conda.sif python

First time you run either of the above commands it will take a bit of time since it needs to install the dependencies into the environment folder. However, subsequent runs will start basically immediately since the environment is then there.

Micromamba and environment files

Under the hood, it uses Micromamba instead of Conda in order to speed up installations but it should not really matter for the functionality.

The one place where I found it to matter is that you have to specify channels in the environment file.

Instead of this (example taken from conda documentation):

name: stats
dependencies:
  - numpy
  - pandas

You need to do this:

name: stats
channels:
  - defaults
dependencies:
  - numpy
  - pandas

But I believe that specifying channels explicitly is anyway good practice.

Running on a supercomputer/cluster

On a cluster you might need to bind folders like here:

$ env APPTAINER_BIND="/cluster" ./conda.sif python

---

To build the image, I have used this wonderful guide as starting point and inspiration.