ci/cd: simplify and refactor workflows #246
Conversation
Simplify the continuous integration workflow by renaming it and making most of the jobs run in parallel. Additionally, the workflow runs on a GitHub hosted runner instead of the Single Node OpenShift runner.
Simplify the CD workflow by invoking a CI build instead of reproducing CI steps on it, as well as splitting the remaining part of the workflow into multiple jobs.
Remove the Tiltfile and Tiltfile-delete files, since Tilt is no longer executed in CI. This is being done in a separate commit so that should Tilt be re-introduced, the files can be restored with a simple revert.
Add an input to the CI workflow to allow skipping sending a Slack notification, as it might not be necessary.
|
It seems every time I read over the documentation and reference for GitHub Actions I find something has changed. I'd love to get some input from everybody on these changes. @rasheedamir @aslafy-z @slauger Can you guys leave a review? The updated actions won't run here until merged, so I tested them on my fork:
And in the |
|
About the failed status check: the Tilt files were removed in this PR, so |
|
Lets do it incrementally; we keep the current workflows as is and add new ones and when the new one has all features of the existing workflow then we can remove We want to test and validate everything i.e. route; grafana dashboard; etc. We use tilt to setup stuff on the SNO |
|
also we have a plan that we should add real integration tests as well i.e. we deploy one sample application and then validate all of the resources are created and are valid with tests; so, we can guarantee 100% quality of this chart |
I see! Does Tilt test those automatically when we run |
|
helm install dry run should be run with |
|
how do we plan on making the missing crds available in kind? |
|
Talked about it at Open Source pruning session today:
|
We discussed switching from running a dry-run install on the Single Node OpenShift clusters as self-hosted runners and instead using kind or k0s.
This PR changes the CI and CD workflows to use kind for a dry-run. The CI workflow is now invoked from the CD workflow to avoid repetition, and Tilt has been removed altogether since I couldn't see why the Tilt setup was necessary.
Bringing back the Tilt files should be as easy as running
git reverton 645165f. If we end up having to use it again, that's how we can restore them.Additionally, linting, unit testing, and the dry run install happen in parallel now. Only the latter needs a cluster, so that's where we bring up kind. For now, since we're not testing specific cluster topologies, no cluster configuration has been added.