Container for offline conda installation

Installing conda packages on a computer without Internet is somewhat challenging. One common approach is to conda-pack your environment, transfer the pack, then “unpack” it on the other computer - link. Or cache all required packages in a local channel that you can bring offline - link.

Here we will demonstrate an easy way to cache the packages in the container, so you can easily install the same environment on a machine without Internet.

Here is a simple scenario.

$ conda create -y -n test
$ conda install -y -n test python=3.8 pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia

Here is a recipe to cache all necessary packages in the container.

Bootstrap: docker
From: continuumio/miniconda3

%post
  export LC_ALL=C
  conda info

/bin/bash <<EOF
  conda create -y -n test
  conda install -y -n test python=3.8 pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia
EOF

%runscript
  conda "$@"

$ sudo conda build conda.sif miniconda3.def

The final container is 6.1 GB ...

Then, copy the container on the remote computer without Internet and run exactly the same installation commands --offline. Note the additional --copy option as well.

./wharf/conda.sif create -n pytorch-test --offline
./wharf/conda.sif install -n pytorch-test python=3.8 pytorch torchvision
torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia --offline --copy

Here is another variant. You can export your environment

$ conda env export -n my_env > my_env.yaml

Bootstrap: docker
From: continuumio/miniconda3

%files
  my_env.yaml /opt

%post
  export LC_ALL=C
  conda info
  conda env create -n test -f /opt/my_env.yaml

  conda env remove -n test

%runscript
  conda "$@"

Then on the target machine:

$ ./wharf/conda.sif env create -n my_env -f my_env.yaml --offline --copy