Run Windows jobs on GitHub

Hand a workflow's Windows jobs to GitHub through the repository's push mirror, while the run here waits for them like any other job.

Last updated 18 November 2026View as Markdown

Deutsche Cloud runs no Windows machines. A repository that pushes to GitHub hands its windows-* jobs to GitHub Actions there; the minutes are GitHub's and billed by GitHub.

Prerequisites

  • A push mirror from the repository to a GitHub repository (see Configure a push mirror), with GitHub Actions enabled there.
  • The run's Linux jobs run on a runner pool or your runners, not on the one-machine-per-run hosted machines (give the organization a runner pool with ubuntu-latest; see Create a runner pool).
  • The secrets the Windows job reads, as repository secrets on GitHub.

Hand the job over

Nothing in the workflow changes. When the Windows job's needs are done:

  1. Deutsche Cloud pushes a branch whose name starts with ews-handoff/ to the GitHub repository: the run's commit, with .github/workflows/ replaced by one generated workflow holding only that job, started by a push to exactly that branch, so no other workflow there runs.
  2. The outputs of the jobs it needs arrive as environment values (needs.build.outputs.version reads env.EWS_NEEDS_BUILD_VERSION), and the original commit, ref and event as EWS_ORIGINAL_SHA, EWS_ORIGINAL_REF and EWS_ORIGINAL_EVENT.
  3. The job's log here links GitHub's run, and the commit status github/<job> shows it on the commit, where a branch protection rule can require it.

Result

When GitHub's run ends, its conclusion is the job's, and its outputs: come back for the jobs after it. The branch is then deleted. Cancelling the run here cancels GitHub's run.

  • When GitHub starts no run within 15 minutes, the job fails with "GitHub started no run on … within 15 minutes (Actions disabled, or a ruleset refused the branch?)".
  • A matrix leg is not handed off: "a matrix leg is not handed off yet; give the Windows job its own job key".
  • Without a push mirror, or when the run's Linux jobs use the one-machine-per-run hosted machines, the Windows job is cancelled with the reason and the rest of the run goes on.
Did this page help you?