I have a GitHub action which contains the following jobs (dependent on each other, running in sequence):
- Compile and deploy.
- Run tests.
Job 1 compiles a C# solution. One project is an ASP.NET Core application that gets deployed to Azure. Another is an xUnit test project that will call the deployed Azure application.
The reason I want them to be two separate jobs is to be able to re-run them independently. Deployment is slow, so if deployment succeeds but tests fail, I want to be able to re-run the tests without re-running the deployment.
Note that in my sequence of jobs above, job 1 creates an artifact-on-disk that job 2 needs, namely the compiled assembly of the test project. This works as long as they are run in sequence, and sometimes it also works if I re-run job 2 much later. But I suspect that my setup is not sane. I suspect that it has just accidentally worked for me so far.
If, for example, I have two feature branches that both trigger the action, I might have a race condition on the test assembly, and one run of the action might accidentally be running a DLL that was compiled from code taken from the other feature branch.
I have read (see this SO thread) that I can do upload-artifact and download-artifact in order to transfer artifacts between jobs. But my test assemblies have dependencies, so I would need to either upload many artifacts or zip the whole folder, upload it, download it and unzip it. That is inconvenient.
Another option would be to check out the source code again and recompile the test assembly at the beginning of job 2. But that might be slow.
What is the best option here? Are there other options?
