Support nvidia cuda ubuntu docker images

[raywonkari]

What would you like added?

Feature request to add more docker images with Nvidia cuda specific ubuntu version as the base image. For example: nvidia/cuda:11.4.2-devel-ubuntu20.04. Currently, the base image is set to ubuntu:20.04.

Why is this needed?

This would help teams making use of the runners to run workloads based out of Nvidia cuda. This will save lot of time and efforts to keep these custom images updated.

Additional context

Currently we have forked this repo, and have created our custom docker image, but this would mean we should continuously update these custom images as the original actions-runner-controller is updated. If we can keep these images in the original actions-runner-controller, the images are automatically update to date.

We can definitely help create a PR introducing these custom docker images, by adding new dockerfile's and plug that into the existing CI. Please let us know if you would like us to do so.

[github-actions]

Hello! Thank you for filing an issue.

The maintainers will triage your issue shortly.

In the meantime, please take a look at the troubleshooting guide for bug reports.

If this is a feature request, please review our contribution guidelines.

[ddelange]

- FROM ubuntu:20.04
+ FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04

would be something to do in a fork and serve from there...

[ddelange]

or would you say devel is a necessity? isn't it unnecessary dockersize of just a bunch of headers a user could install on the fly from preconfigured apt if needed? versioning of cudnn is iffy (see their dockerfiles) so I personally would opt for the first one below:

$ dockersize nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04
linux/arm64  1.58G
linux/amd64  1.62G
$ dockersize nvidia/cuda:11.7.1-cudnn8-devel-ubuntu20.04
linux/arm64  3.42G
linux/amd64  3.48G
$ dockersize nvidia/cuda:11.7.1-runtime-ubuntu20.04
linux/arm64  1.06G
linux/amd64  1.10G
$ dockersize nvidia/cuda:11.7.1-devel-ubuntu20.04
linux/arm64  2.35G
linux/amd64  2.41G

[ddelange]

Added a commit on my fork doing a search and replace to achieve the above diff just in time:
master...ddelange:actions-runner-controller:master ddelange/actions-runner-controller-releases@actions-runner-controller:releases:main...main

Now pushing images with CUDA base to docker.io (ref) and ghcr.io (ref) in addition to regular summerwind/actions-runner:

$ dockersize ddelange/actions-runner:ubuntu-20.04
linux/amd64  1.92G
linux/arm64  1.83G
$ dockersize ddelange/actions-runner:ubuntu-22.04
linux/amd64  1.85G
linux/arm64  1.75G
$ dockersize ghcr.io/ddelange/actions-runner-controller-releases/actions-runner-dind-rootless:ubuntu-20.04
linux/amd64  1.99G
linux/arm64  1.89G
$ dockersize ghcr.io/ddelange/actions-runner-controller-releases/actions-runner-dind-rootless:ubuntu-22.04
linux/amd64  1.92G
linux/arm64  1.82G

...

[ddelange]

cc @Link-

[ddelange]

update: since the release flow was refactored ref ddelange#1, I removed the commit I mentioned above, and instead added it here ddelange/actions-runner-controller-releases@actions-runner-controller:releases:main...main

• dockerhub: still the same
• ghcr.io: now publishes under ghcr.io/ddelange/actions-runner-controller-releases instead of ghcr.io/ddelange/actions-runner-controller

edit: reflected this change in the previous comment

[neumann-nico]

@ddelange Thanks for your awesome work!
Are you going to continue to update your Nvidia image?
Looks like it stopped a while ago

[ddelange]

Hi @neumann-nico 👋

I pulled master and manually triggered a build: https://github.com/ddelange/actions-runner-controller-releases/actions

Lets see if the daily cron kicks in again!