用于 Python 的 Nvidia-Docker API？

Question

I am currently running a lot of similar Docker containers which are created and run by a Python script via the official API. Since Docker natively doesn't support GPU mapping, I tested Nvidia-Docker, which fulfills my requirements, but I'm not sure how to integrate it seamlessly in my script.

I tried to find the proper API calls for Nvidia-Docker using Google and the docs, but I didn't manage to find anything useful.

My current code looks something like this:

# assemble a new container using the params obtained earlier
container_id = client.create_container(img_id, command=commands, stdin_open=True, tty=True, volumes=[folder], host_config=client.create_host_config(binds=[mountpoint,]),detach=False) 
# run it 
client.start(container_id)

The documentation for the API can be found here.

From Nvidia-Dockers Github page:

The default runtime used by the Docker® Engine is runc, our runtime can become the default one by configuring the docker daemon with --default-runtime=nvidia. Doing so will remove the need to add the --runtime=nvidia argument to docker run. It is also the only way to have GPU access during docker build.

Basically, I want to add the --runtime=nvidia-docker argument to my create_container call, but there is no support for that as it seems.

But since I need to switch between runtimes multiple times during the script execution (mixing Nvidia-Docker and native Docker containers) the quick and dirty way would be to run a bash command using subprocess but I feel like there has to be a better way.

TL;DR: I am looking for a way to run Nvidia-Docker containers from a Python script.

Answer 1

run() and create() methods have runtime parameter according to https://docker-py.readthedocs.io/en/stable/containers.html

Which has sense because docker cli tool is pretty simple and every command translate in a call to the docker engine service REST API