Scaling Molecular Software Testing with Self-Hosted GitHub Actions Runners

Abstract
Molecular software is beginning to use more GPU-accelerated compute and traditional CI solutions can often be expensive when running longer GPU tests. In order to understand the existing solutions, we reviewed 6 potential self-hosted GitHub Actions Runner candidates[1-6] from a curated list[7]. This poster explains how the Open Molecular Software Foundation (OMSF) seeks to provide a new solution[8] for cloud-based GitHub Actions Runners to make scaling of GPU compute more effective for molecular software. We highlight the long-term community decisions discussed, such as language selection, ability to receive feedback, cloud-agnostic architecture, and ease of maintainability that lead to developing a new solution rather than updating existing tooling.  In addition, we discuss common pitfalls in working with cloud-based infrastructure and how we implemented a testing strategy to mitigate the potential costs of untested infrastructure code.

References

[1] Muransky, M (2023) ansible-github_actions_runner (1.18.7)  [Source code]. https://github.com/MonolithProjects/ansible-github_actions_runner 

[2] Young, M (2024) docker-github-actions-runner (Release 2.315.0) [Source code]. https://github.com/myoung34/docker-github-actions-runner

[3] Machula, V (2024) ec2-github-runner (v2.3.6) [Source code]. https://github.com/machulav/ec2-github-runner

[4] GitHub (2024) actions-runner-controller (v0.9.1) [Source code]. https://github.com/actions/actions-runner-controller

[5] Knape, S (2020) github-runner [Source code]. https://github.com/SanderKnape/github-runner

[6] Phillips Labs (2024) terraform-aws-github-runner (v5.9.0) [Source code]. https://github.com/philips-labs/terraform-aws-github-runner 

[7] Nicolai, J (2021) awesome-runner [Online] Available: https://jonico.github.io/awesome-runners/ 

[8] Open Molecular Software Foundation (2024) gha-runner [Source code]. https://github.com/omsf-eco-infra/gha-runner