Image building is inefficient

[Akrog]

If a CI system has more than 1 backend it is inefficient to build the image for each test run.

Ideally we would only build 1 image for all the builders.

[Akrog]

We could use the DependentScheduler to run the test after building the image, but we'll have to make some modifications because dependents don't get triggered if the upstream one (image build) doesn't complete successfully, so they won't report failure. We'll also want to report the results of the image build on the jobs on failure and on success.

[Akrog]

Another alternative is to use GitHub Actions with some custom code (since they don't yet support on-premise runners).

We would have the webhooks defined but disabled in GitHub, and then the action would build the image and publish it to an equivalent of the dockerhub travis-ci repository using secrets.

Then, once the image is published the Action will query the GitHub API (using the GITHUB_TOKEN env variable) to look for the webhooks that match a regex and then send the GITHUB_EVENT_NAME and payload from file GITHUB_EVENT_PATH. It'd need to reproduce the headers X-GitHub-Event and X-Hub-Signature (using the webhook's secret).

Description of env variables: https://developer.github.com/actions/creating-github-actions/accessing-the-runtime-environment/#environment-variables