GitHub Custom Runners (with Virtual Machine Support)

[mmcky]

We have been able to use [cml]() project to setup GitHub custom runners on a QuantEcon server -- however it lacks virtual machine creation and teardown. It would be great to have virtual machine support on the machine we want to setup a custom runner on -- as then it is the same environment to GitHub Actions.

We can use their images

https://github.com/actions/runner-images

• [ ] find a project to orchestrate the vm setup and teardown + cache

[mmcky]

This has some promise but a new VM is not setup each time still

https://dev.to/peaceiris/start-github-actions-self-hosted-runner-with-virtualbox-and-vagrant-49ei

So it still would behave differently to the cloud hosted solutions.

One benefit here though is that it can comfortably run public repos on a server (with better security)

[mmcky]

@chappiewuzefan this is some information with regards to the GitHub custom runner project. It has some links you could start reading along with some ideas on what we might be able to configure and use.

• [ ] @mmcky to write up some configuration of cml that we currently use to run our containers on ec2

[chappiewuzefan]

1. Github-hosted or Self-hosted
2. If it is Self-hosted:
   • additional software the workflows need
   • extend docker image
   • use docker-compose

[chappiewuzefan]

Google collab for experiment on one a lec, below are useful links: https://github.com/jonico/awesome-runners

https://hub.docker.com/layers/mmcky/quantecon-lecture-python/cuda-12.1.0-anaconda-2023-03-py310-b/images/sha256-3f8345f3e7ca43bbe51b10bff8359c977a490059fa7efc116ce479f856717dba?context=repo

https://hub.docker.com/r/myoung34/github-runner

https://github.com/actions/runner-images

https://cml.dev/