Question

Your enterprise runs thousands of CI jobs per day on self-hosted runners in a Kubernetes cluster using Actions Runner Controller (ARC). During peak hours, jobs sit queued for long periods because the fixed pool of runners is exhausted, while off-peak the same runners sit idle and waste cluster resources. You need runners to automatically scale up when workflow jobs are queued and scale down when demand drops. Which approach best meets this requirement?

Accepted Answer

Deploy runner scale sets with ARC configured to listen for workflow job events, so the number of runner pods automatically scales based on queued and in-progress jobs.

Answer

Increase the static replica count of the runner deployment to the maximum expected peak so no job ever waits in the queue.

Answer

Add more runner labels to the existing runners so multiple jobs can be assigned to a single runner concurrently during peak load.

Answer

Switch the affected repositories to GitHub-hosted runners only during peak hours using a scheduled workflow that edits the runs-on value.