[nefasti]

I’ve set up teamcity, Jenkins, CircleCI for past projects and companies. Tried to use gitlab with all their built-in DevOps feature and found it extremely frustrating to get to a good setup.

I dig the idea of all in one but think it really needs some polishing

[throwaway5752]

I am glad you said it. Right now, the setup is so opinionated that it's limited and suffers from the mile wide and inch deep problem. In particular, they are still behind the Jenkins default setup in terms of standardized xunit output parsing, tracking and visualizing test trends over time - and since they have all the git topology information at their disposal there are amazing possibilities there and its a real shame. They are light years away from large scale project management capabilities of JIRA + plugins and I wish they would improve the core source control + build pipeline experience before tackling that space.

[snuxoll]

> Tried to use gitlab with all their built-in DevOps feature and found it extremely frustrating to get to a good setup.

All the AutoDevOps stuff is pretty much just some standard build jobs that use Herokuish and Helm, and the inflexibility of them goes hand-in-hand with Gitlab's "breadth over depth" strategy. Hopefully it continues to mature, but for now it's only appropriate for smaller/less complicated applications.

Fortunately it's pretty easy to ignore the AutoDevOps pipeline and still get the benefits of the various integration points that were added to make it work - if a Kubernetes cluster is tied to a project environment then a kubeconfig will be injected into any CI job with the environment set, the same job will automatically get recognized as the deployment job so rollbacks work seamlessly as well, etc. I've done this myself to deploy Mediawiki on Kubernetes, as the default pipeline was not setup for complex multi-component deployments (memcached, mediawiki, apache trafficserver, service endpoints for an external mysql cluster) and it works well.