However our experience with managing Jenkins via Kubernetes and declarative Jenkins configuration has led us to believe that this cost is small, particularly as the majority of actual building and testing is handled inside docker containers and Kubernetes, alleviating the need for less supported plugins that may make Jenkins administration more difficult.I want to trigger a specific TeamCity (2018.2.4) build when GitLab (Community 11) merge requests are created and pushed but not when any other branch is pushed (for other branches I'll have different build configs).
Obviously as a consequence we take on the indirect costs of maintain Jenkins itself, patching it, upgrading etc. Jenkins offered us the ability to choose one of a number of credentials/secrets management approaches, and models secrets as a more dynamic concept that GitHub Actions provided.Īdditionally, we are operating Jenkins within our development Kubernetes cluster as a kind of system-wide orchestrator, allowing us to use Kubernetes pods as build agents, avoiding the ongoing direct costs associated with GitHub Actions minutes / per-user pricing. Within our deployment pipeline, we have a need to deploy to multiple customer environments, and manage secrets specifically in a way that integrates well with AWS, Kubernetes Secrets, Terraform and our pipelines ourselves. It's been a great balance between a managed service and the flexibility of running it yourself.Īnd while none of my needs have hit the scale of Shopify (I saw one of their engineers speak about it at a conference once, I can't find the video now though 😞) it's good to know I can scale out my worker nodes to hundreds of thousands of workers to reduce the time it takes for my tests to run. When you completely control all of the environment the tests are running in you define those constraints too. It's also proven to be much easier to support than trying to deal with the problems that come with trying to force an app to fit into the nuances and constraints that are imposed by the containers/runtime of a CI service. It means that dev/test/prod parity is simple to achieve and maintain. Having a test environment where the only difference was adding the Buildkite agent was a trivial addition.
Running my app in a production (or production-like) environment was already a solved problem, because everything was already in some form of "everything as code". Eventually it dawned on me why this approach was more ingenious than I realised: It took me a while to get around to Buildkite because at first I didn't see much point given it seemed like you had to run the agent yourself. I think I've tried most of the CI tools out there at some point.
0 kommentar(er)
