runnerd
runnerd runs ephemeral GitHub Actions runners on one Linux host: a fresh container per job, sized from what the repositories you serve have used.
Install
|
init writes a config sized for the host and a systemd user unit, runs runnerd doctor and prints the steps left. It never runs anything as root; the slice and Sysbox commands it prints are yours to run. Then:
Point a job at the host with runs-on: [self-hosted, linux, x64]. Serve private repositories only: on a public one, anyone who opens a pull request can run code on this host.
Requirements
- Linux with cgroup v2, and Docker with the systemd cgroup driver.
- Sysbox and a systemd slice whose
MemoryMaxis the runners' budget. - The GitHub CLI, logged in with the
reposcope.
Binaries ship for x86_64 and aarch64; the runner images are linux/amd64 only, so on aarch64 build the base stage of image/Dockerfile on the host.
Documentation
Full docs: https://ayagmar.github.io/runnerd/. Install covers pinned versions and attestation checks, Configuration every key, Why is my job waiting? every reason a job gives, and Security model what jobs, the token and the dashboard can reach. ARCHITECTURE.md says where the code lives, docs/adr why, and CONTRIBUTING.md how to send a change.
Building
Run "The gate" in AGENTS.md before sending a change. Tests that start real containers are ignored by default and need Docker with Sysbox and the images:
The mirror's repair test needs only Docker, and pulls the pinned registry image:
The Gradle build cache test also needs the network:
License
MIT