Build Pipeline Decision Process : Examples
In the last post, I laid out the decisions that go into a build pipeline. With this post, I want to walk through my decisions on a couple of personal projects.
The code point generator source generator is in a public repository. I wanted the continuous integration (CI) part to compile the code and run the unit tests. I also wanted the continuous deployment (CD) part to deploy just the main branch. So the pipeline also needed to selectively create a NuGet package, create a GitHub release, and publish to NuGet.
The first decision I made was regarding the structure of the pipeline code. I wanted to keep open the option for switching between services, so instead of using the proprietary configuration language of one or other services, I created a Cake script that performed each of the steps.
I wanted to work with a third-party service to get some experience with other options, so I decided not to go with on-prem or custom pipeline management software. This may sound like a contradiction with the previous paragraph, but notice that the pipeline code that I wrote lives in the repository. The service that I chose needed to handle the responsibility of cloning the repository and running the pipeline code. These responsibilities can be custom designed, but I decided against this.
Another concern is that I do not have a budget for build pipeline services, so there needs to be a free ($) tier available. Since this project is open-source and in a public repository, I chose AppVeyor, who offer a free tier for public repositories. I still had to write an AppVeyor specific adapter to run the Cake script, but the proprietary configuration is minimal for that use case.
Private Project (Name TBD)
I am working on another project that is not yet ready for the light of day. It is in a private repository.
As above, I wanted to keep open the option of switching between services (which came in handy), so I used the same model of a Cake script that performed the pipeline steps instead of a proprietary configuration and a proprietary configuration adapter.
Because this project is in a private repository, AppVeyor is not an option. I chose CircleCi because they offer a limited free ($) option for private repositories.
In the next installment, I will go over a recent period where I was able to test the portability of the pipeline code.