How to Scale Runners to Zero to Cut Idle Cost

A self-hosted runner sitting idle still bills for its instance; scaling to zero means you only pay while jobs actually run.

Use an autoscaling controller (for example actions-runner-controller on Kubernetes) that provisions runners when jobs queue and removes them when the queue drains.

Steps

• Deploy an autoscaler that watches the GitHub job queue.
• Set the minimum replica count to zero.
• Let the controller scale up on queued jobs and back to zero when idle.

Autoscaler config sketch

# actions-runner-controller: scale from 0 up to a ceiling on demand
apiVersion: actions.github.com/v1alpha1
kind: AutoscalingRunnerSet
spec:
  minRunners: 0        # no idle cost when the queue is empty
  maxRunners: 20
  githubConfigUrl: https://github.com/my-org

Gotchas

• Scaling from zero adds cold-start latency to the first queued job; keep a warm pool if that hurts.
• The autoscaler control plane itself has a small always-on cost.
• Combine scale-to-zero with ephemeral runners so no stale runner lingers and bills.

Related guides