Introducing GitHub Actions runner scale set client

[samus-aran]

We’re excited to introduce the GitHub Actions runner scale set client that is now available in public preview. The client is a standalone Go-based module that lets you build custom autoscaling solutions for GitHub Actions runners without requiring Kubernetes.

The client provides the core building blocks for implementing autoscaling on any infrastructure. Whether you're running containers, virtual machines, bare metal servers, or cloud services, by integrating directly with GitHub's scale set APIs, you gain full control over runner lifecycle management while GitHub handles the orchestration logic.

Key capabilities:

• Platform agnostic design: Works with containers, virtual machines, and bare metal servers across Windows, Linux, and macOS.
• Full provisioning control: You define how runners are created, scaled, and destroyed based on your specific requirements.
• Native multi-label support: Assign multiple labels to each scale set, optimizing resources for multiple build types.
• Agentic scenario support: Supports agentic workflows, including GitHub Copilot coding agent.
• Real-time telemetry: Monitor job execution and runner performance with built-in metrics.

How it works

The client orchestrates API interactions with GitHub's scale set infrastructure while leaving the actual runner provisioning entirely in your hands. This separation means you maintain complete control over your infrastructure decisions. Whether that's spinning up containers, launching cloud virtual machines, or provisioning physical hardware, all while benefiting from GitHub's scale set APIs.

Important: You'll manage all infrastructure setup, provisioning logic, and scaling strategies. GitHub provides the APIs and client libraries. You're responsible for implementing your infrastructure automation.

Relationship to Actions Runner Controller

The Runner Scale Set Client isn't a replacement for Actions Runner Controller (ARC), which remains a reference implementation of the scale set APIs and the recommended Kubernetes solution for autoscaling runners. Instead, the client is a new tool for interfacing with same scale set APIs for building custom autoscaling solutions outside of Kubernetes. Multi-label support will also be made available for ARC with the 0.14.0, which will be available in March 2025—see the issue for adding this support.

Get started

The GitHub Actions Runner Scale Set Client is fully open source and available for you to explore the source code and adapt the implementation to your specific infrastructure needs. You can also visit the GitHub Actions Runner Scale Set Client documentation for additional information.

Share your feedback

Got feedback, ideas, questions, or stories about how you’re using this? Post below in the comments - bug reports welcome there too. If you’ve got a clear bug to track, you can also open an Issue in the repo. Contributions are welcome! ❤️

[shubham055555]

This is a very interesting and valuable update.

The introduction of a platform-agnostic runner scale set client without requiring Kubernetes significantly improves flexibility for developers who prefer custom or lightweight infrastructure setups.

What stands out most:

• Support for multiple environments including virtual machines, containers, and bare metal
• Full control over provisioning and autoscaling logic
• Native multi-label support for handling diverse workloads
• Compatibility with agentic workflows such as GitHub Copilot integrations

I have a question regarding implementation outside Kubernetes:

Are there any recommended best practices or reference architectures for:

• Cost optimization during autoscaling
• Handling runner failures or crash recovery
• Efficient queue-based scaling strategies

It would be helpful to see practical examples for setups such as:

• Docker-based runners on a single server
• Cloud-based VM autoscaling (AWS, GCP, Azure)

Overall, this appears to be a powerful tool for teams that want full infrastructure control without the overhead of Kubernetes. Looking forward to exploring it further.