How to Use Spot Instances for Self-Hosted Runners

Spot instances cost a fraction of on-demand, so idempotent CI jobs are a natural fit for interruptible capacity.

Run ephemeral runners on spot capacity. Because a reclaimed instance just drops the current job, keep jobs idempotent and let GitHub re-queue them.

Steps

• Provision spot instances in an autoscaling group as ephemeral runners.
• Register each runner with --ephemeral so it exits after one job.
• Handle reclamation by re-queuing the interrupted job on fresh capacity.

Register an ephemeral runner

# On the spot instance, register a one-shot ephemeral runner
./config.sh --url https://github.com/my-org \
  --token "$RUNNER_TOKEN" \
  --ephemeral \
  --labels spot,linux,x64
./run.sh   # exits after one job; ASG replaces the instance

Gotchas

• A reclaimed instance can kill a job mid-run; only use spot for retry-safe work.
• Spot prices fluctuate; set a max price and a fallback to on-demand for critical queues.
• Do not run deploys on spot capacity where an interruption could leave a half-done release.

Related guides