CI runner glossary
Answer in brief
Short, neutral definitions of 25 terms you will meet when comparing CI runners, from basics like self-hosted runner and ephemeral runner to features like Docker layer caching and BYOC.
- Actions Runner Controller (ARC)
- GitHub’s open-source Kubernetes operator that creates and scales self-hosted GitHub Actions runners as pods.
- arm64 runner
- A runner with a 64-bit Arm processor, used to build and test software for Arm servers and devices natively instead of under emulation.
- Bare metal
- Physical servers used directly rather than rented as cloud virtual machines; runner vendors use them to pick faster CPUs and run their own virtualization.
- BYOC (bring your own cloud)
- A deployment model where the runners are virtual machines in the customer’s own cloud account while a vendor’s software orchestrates them.
- CI runner
- The machine that executes the jobs of a continuous-integration pipeline: it checks out code, runs build and test steps and reports the result.
- Compute minutes
- GitLab’s billing unit for jobs on GitLab-hosted runners: job duration multiplied by a cost factor that depends on the runner type and size.
- Concurrency limit
- The maximum number of CI jobs, or total vCPUs, that can run at the same time on a plan or account.
- Dependency cache
- A store that saves downloaded packages and build outputs between CI jobs, keyed by a hash, so later jobs can restore them instead of redoing the work.
- Docker layer caching
- Reusing the unchanged layers of a container image from an earlier build so a CI job rebuilds only what changed.
- Ephemeral runner
- A runner that takes exactly one job and is then destroyed, so every job starts from a clean machine and nothing carries over.
- Firecracker
- An open-source virtual machine monitor from AWS that runs microVMs on KVM, built for short-lived, multi-tenant workloads.
- GitHub-hosted runner
- A virtual machine that GitHub provisions for one GitHub Actions job, with a maintained image of preinstalled tools, and destroys afterwards.
- GitLab Runner executor
- The mechanism a GitLab Runner uses to run a job: a shell, a Docker container, a Kubernetes pod, or an autoscaled virtual machine.
- Just-in-time runner
- A self-hosted GitHub Actions runner created from a one-off configuration issued by the API, valid for a single job.
- Larger runner
- A GitHub-hosted runner with more cores, memory or disk than the standard size, or with a GPU, billed per minute at a higher rate.
- macOS virtualization
- Running macOS as a virtual machine on Apple hardware, which lets a Mac give every CI job a clean, disposable macOS environment.
- MicroVM
- A minimal virtual machine with a stripped-down device model that boots in a fraction of a second and gives each workload its own guest kernel.
- OIDC in CI
- OpenID Connect lets a CI job prove its identity to a cloud provider and receive short-lived credentials, instead of storing long-lived secrets.
- Queue time
- The delay between a CI job being requested and a runner starting to execute it.
- Runner scale set
- A named group of identical, autoscaled self-hosted GitHub Actions runners that a workflow targets by the scale set’s name.
- runs-on label
- The line in a GitHub Actions job that names the kind of runner the job needs; changing it is how a workflow moves to another runner provider.
- Self-hosted runner
- A runner on a machine you (or a vendor acting for you) operate, registered with the CI system to pick up jobs in place of the platform’s hosted runners.
- Spot instance
- Spare cloud capacity sold at a large discount on the condition that the provider can reclaim the machine at short notice.
- Static egress IP
- A fixed, published set of IP addresses that a runner’s outbound traffic comes from, so it can be added to firewall allow-lists.
- Warm pool
- A set of runners started ahead of demand and kept idle so that a queued job begins immediately instead of waiting for a machine to boot.
Check before you rely on it. Products change quickly, and this page reflects public documentation as of November 22, 2026. Confirm anything that matters in the provider’s own documentation.