GitHub Actions Runner Controller 0.14.0 adds multilabel scale sets

GitHub made Actions Runner Controller 0.14.0 generally available on March 19. The release adds multiple labels to runner scale sets, replaces ARC’s earlier internal service client with the public actions/scaleset Go library, and expands Kubernetes metadata controls. GitHub’s release announcement also describes experimental Helm chart rewrites and a Linux node selector for listener pods.

Fewer scale sets for combined requirements

Before this release, ARC supported one label per scale set. Platform teams often had to create separate sets for combinations such as operating system, hardware class, network zone, or compliance boundary. Multiple labels let a workflow target a combination in runs-on while the underlying capacity remains one managed scale set.

That convenience does not replace isolation. A label is routing metadata, so teams should keep separate pools when workloads require different trust boundaries, credentials, or network access. The useful change is reducing duplication where the distinction is scheduling rather than security.

Upgrade details deserve a canary

ARC can now apply custom labels and annotations to roles, bindings, service accounts, and listener pods. This should help clusters whose policy engines, cost reports, or inventory depend on standard metadata. The new default kubernetes.io/os: linux selector also prevents a listener from landing on an incompatible Windows node in mixed clusters, though operators can override it.

Test 0.14.0 against a noncritical runner group first. Verify existing Helm values, admission policies, pod placement, and every workflow label expression. The new ability to stop autoscaling for outdated configurations depends on a later runner change, according to GitHub, so teams should not treat that protection as immediately complete across an older fleet.