A collection of docker images that provide a common core for use in continuous integration.
The idea of these images is to balance the following:
- Frequency of updates
- Standard set of tooling
- Common Environment
I encountered a lot of difficult with trying to automate the updating of those docker images. GitLab does not really have the idea of
bots. Instead it would require a personal access token for my account, if I wanted to make commits. I didn’t like, as that would grant access to pretty much every one of my repositories. At the time I was self-hosting Gitlab, so I opted to just create additional users that would be my bots.
Having a bot user worked pretty well, in the sense that a scheduled Gitlab repository would execute the updates. However, at that time the building of each image was a burden on my local self-hosted CI machine. When I opted to use Gitlab.com instead of self-hosted, I ended up abandoning the idea of automated upgrading.
Later on when I was doing some more work in the frontend realm, I found that I didn’t like working with the available docker images. They fit into a couple categories:
- Very big
- Dependencies I did not like/want
- Slow update schedule
That is when I decided to move my docker images into a separate project called
CardboardCI. The hope was a set of small images, with automated updates and a standard set of dependencies (
jq/zip/tar/etc). These images were based on alpine, until ubuntu released minimal images (~35MB). At that time I started the process of moving them over to ubuntu.
When GitHub Actions became available, I noticed that you would be able to use the
GITHUB_TOKEN to commit against the repository itself. This meant you could define a workflow that checked for updates, and if existed, create a pull request with those updates.
In essence, you would be able to roll your own dependabot. I thought this would be ideal for CardboardCI, and moved it from GitLab to GitHub.