In my last post, I mentioned that while upgrading my Heroku tech stack I noticed Codeship was experiencing some kind of outage. This seemed to stop anything appearing in the "checks" part of a pull request (including any kind of error message, which was a long way from helpful) and I decided to investigate Github Actions for my CI/CD needs.
I've been thinking about using Github Actions for a little while, for two reasons. First, I wanted to run my linting in my CI pipeline and I know of a rather good tutorial for getting started doing this using Actions (thanks Dean!). Second, this should move my CI config to the project repository (keeping it together and putting it under version control) and remove a dependency on a third party SaaS product. I can't help feeling that the recent Codeship outage (which I only noticed because the check was missing in my PR, and could easily have missed this) vindicates this last point.
As a side note, Codeship now seems to be fully working again.
Anyway, the Rubocop implementation is actually pretty straightforward, but it took me forever to get the tests running because of a few tricks and gotchas which I thought I'd record for posterity.
Let's start with a timesaver. I read a lot of examples while setting this up, and some had extensive Bundler config in them. However, if you're using the
ruby/setup-ruby@v1 action for setting up Ruby (code here) and you put in:
with: bundler-cache: true
It will just handle everything bundle-related with no more configuration. Hurrah!
Migrating Capybara driver to Apparition
I have no idea if anyone else is still using Capybara Webkit to drive their Capybara tests but I was. It has recently become a pain to install because the underlying library (QtWebKit) has been deprecated. I found this out after quite a while of trying to get the QtWebKit libraries accessible in my Action. That didn't work.
It seems Thoughtbot, the authors, agree with me and have deprecated the thing and recommend a move to Selenium or Apparition. I chose the latter because of claims of backwards compatibility and it was very easy to switch when I finally realised that this was a more sensible way forward. The changes can be seen in this commit along with the inclusion of an Action setting up the Chrome driver in my test workflow.
Configuring the database to work in a containerised world
Good grief this took me forever.
In theory, this is really easy - configure a Postgres database as a service, when the tests run connect it up, and bam. In practice it is also really easy, requiring minimal config to get it working. However, it requires getting a load of options to line up and since it's all running on Github servers, the feedback loop is annoyingly slow so painstakingly iterating through a million tiny variations to get to that simple working config took an eternity.
In the end, there were only two things to note.
First, when configuring the Postgres service one HAS to specify a port (despite it being the default port).
Second, remember to update the test database config in
database.yml to accept some environment variables (and also default to allowing the tests to run locally). It's really easy to do when you actually remember to do it...
It's highly like these are more down to my own incompetence than anything hidden or surprising.
And lo, it works. While it took a while to figure all the details out, the results are actually really simple and easy to duplicate for other Rails projects.
The whole change for implementing Github Actions and implementing the other updates can be seen in this PR.