Moving a job changes one line of its workflow file, runs-on:. The job then runs on a machine of ours rather than one of GitHub's, and this page lists the ways that machine, and GitHub's view of it, are not the same. Each line links to the page that has the whole of it.
No
/dev/kvm. There is no nested virtualisation, so an Android emulator or a nested hypervisor does not start. Docker containers do not need it and work normally. LimitsNo Google Chrome and no Firefox. The x64 images carry Chromium with its ChromeDriver instead, and a job that needs Chrome or Firefox installs it, for example with
browser-actions/setup-chrome. The ARM images carry no browser. The runner imageWindows is Server Core. There is no desktop and no Edge, and Visual Studio is there as its Build Tools, not the full IDE. A Windows runner cannot run Linux containers, as on GitHub's own Windows runners. The runner image
Disk. Each label's root disk is in the labels table, and the operating system and the preinstalled tools share it. A GitHub larger runner has more disk than our label of the same size, and the migration pull request says so beside each move it makes. Runner labels
Memory. At most x64 and Windows sizes our memory is a little under GitHub's, and the migration pull request says so beside each move, so a job that needs all of it can take the next size up. Moving from GitHub-hosted runners
Sticky disks attach to Ubuntu x64 jobs of 1 to 16 vCPU. ARM, Windows and 32 vCPU jobs run with an empty disk that is thrown away when the job ends, even while sticky disks are on. Runner labels
A pull request does not read its base branch's cache. Its job restores from its own ref, then from the default branch. The Actions cache
One cache entry may be up to 10 GiB. A larger save is refused and nothing is kept for it. The Actions cache
github.runner_environmentisself-hosted, and so is therunner_environmentclaim in a job's OIDC token. A condition or a cloud trust policy that requiresgithub-hostedtakes the other branch, or refuses the job. Secrets, OIDC and allowlistsRunner groups. Our runners join your organisation's Default runner group. If your organisation or its enterprise limits which repositories may use that group, or turns self-hosted runners off for a repository, GitHub does not hand that repository's jobs to our runners, and they wait until GitHub times them out. Secrets, OIDC and allowlists
Matrix builds.
runs-onis chosen per job, so a matrix can name ours in one leg and GitHub's in another. The migration pull request moves a${{ matrix.os }}job when the job lists its values as plain labels, and leaves it whole, with the reason, when anything else in the job reads that value or the matrix is built in a way it cannot follow. Moving from another runner providerReusable workflows. A called workflow's jobs run on the labels its own file names. The migration pull request changes the workflow files of the repository it is opened on, so a reusable workflow kept in another repository keeps its labels until you change them there, and a
runs-on:read from a workflow input is listed and left as it is. Moving from another runner providerPull requests from forks. A build from a fork gets no sticky disk, and nothing a fork's build writes is kept on one. Security
Moving from GitHub-hosted runners: the label mapping, and the migration pull request.
Runner labels: every size we sell.