How to use a docker container as environment for a conda project?

I am working on a python project that uses a conda environment. I got access to new resources for testing my project on a server but I can on ly run docker containers there. My supervisor told me to put in the docker container only the environment, but I'm new to docker so I have no idea on how to do it. Until now, I just put the environment and the code into the docker image and run everything from there, with this dockerfile

FROM continuumio/miniconda3

# needed to run docker container as non-root
ARG USER=<user>
ARG USER_ID=<userid>
ARG USER_GROUP=<userGroup>
ARG USER_GROUP_ID=<userGroupId>
ARG USER_HOME=/home/${USER}

RUN groupadd --gid $USER_GROUP_ID $USER && useradd --uid $USER_ID --gid $USER_GROUP_ID -m $USER

RUN apt-get update && apt-get install -y curl


COPY env.yml .
RUN conda env create -f env.yml

SHELL ["conda", "run", "-n", "env", "/bin/bash", "-c"]

WORKDIR /app

COPY main.py /app/
COPY src/ /app/src/

ENV PYTHONPATH=/app

USER $USER

ENTRYPOINT ["conda", "run", "-n", "env", "python", "/app/main.py", "-C", "src/configs/config.yaml", "--use-splits", "-y"]

In this way, I include the code in the file, but I do not understand how it is possible to use a docker container only as environment. How should I modify the dockerfile to make it work? And how should I run the script telling it to use the docker container as environment?

Solution

If I understand well, you've already built your docker image. If it's not the case, and assuming your are in the folder containing your Dockerfile, simply do:

docker build -t <image name> .

To run your script inside a docker container, the idea is NOT to tell your script that is should run inside it. Conversely, you need to create a docker container and tell it to run you script.

create a container from the image your built (docker run --rm <image name>)
run your script inside this container (conda run -n env python" /app/main.py ...)

However, you specified an entrypoint inside your Dockerfile. This command will automatically be executed whenever your do:

docker run --rm <image name>

However, I think you should use this entrypoint instead:

ENTRYPOINT ["conda", "run", "-n", "env", "python", "/app/main.py"]

It will allow you to change the options you pass to the script, without having to re-build the image again. You could then run your script in your container with:

docker run --rm <image name> -C src/configs/config.yaml --use-splits -y

In addition, I would suggest to run your container with a volume mapping. This way, you won't have to re-build your project everytime you make modifications in your source code. Assuming your project is like:

my_project
|__ main.py
|__ src/

Simply go to your project root folder and do:

docker run --rm --volume $(pwd):/app <image name> <your script options>

or whatever options you need to pass.

To sum it up

I would change the entrypoint as stated above, and run your container like:

docker run --rm --volume $(pwd):/app <image name> -C src/configs/config.yaml --use-splits -y