Cirrus CI is a relatively new hosted CI service that offers several unique features. It's probably the only CI provider to offer full virtualization (KVM) or FreeBSD runners for free. Currently their business model is centered around GitHub Marketplace and only the projects hosted at GitHub are supported.
Fortunately Cirrus CI provides a very capable GraphQL API. Thanks to that I was able to write a simple command line tool to trigger CI builds with custom configuration: cirrus-run. It reads a local YAML file and executes Cirrus CI build with that configuration. You can execute the build against any GitHub repo you're allowed to access.
Since build configuration is not provided by the repo the job may have no relation to it. You can execute any number of jobs for any number of projects against a single dummy holding repo at GitHub - which is the approach I suggest to use for setting up integration with other source code hosting platforms.
Here is how I've set it up with GitLab:
- Project repo is hosted at GitLab
- Each push triggers a new build in GitLab CI.
The only purpose of that build is to execute
cirrus-run triggers the real build in Cirrus CI, waits for it to complete and reports the results.
Cirrus CI build is owned by a dummy GitHub repo that contains only one initial commit. Providing custom clone script allows to skip cloning that dummy repo.
That allows me to continue using GitLab to host my project and GitLab CI to run the jobs it's good at while delegating the more demanding jobs to Cirrus CI. All status reports are gathered by GitLab CI and failure notifications arrive uniformly to my inbox regardless of where the build was executed.
Configuration files that enable the workflow described above: