← Back to changelog

2026-09-01

Cluster definition v1: a cleaner, standard format

#kubernetes  #yaml 

The cluster definition files in your repository (k8s-clusters/*.yaml) move to a new format: apiVersion: skyscrapers.eu/v1. The old format grew over the years and ended up with mixed naming styles, and with settings for one component spread over several places.

The new format follows the Kubernetes API conventions: every key is camelCase, related settings sit together, and pod resources are set the same way everywhere. A few examples of where things move:

  • spec.eks_version becomes spec.kubernetesVersion.
  • spec.cluster_monitoring, spec.grafana_loki and the other monitoring and logging keys move under spec.observability.
  • spec.traefik and spec.cluster_fqdn move under spec.ingress.
  • spec.karpenter.node_pools becomes the list spec.nodes.pools, where each pool carries its own name.

What this means for you

  • Mistakes get caught early. Every key now lives in a schema, so we validate your cluster definition on every pull request. A misspelled or misplaced key fails that check instead of being silently ignored, and you find out before it reaches your cluster.
  • The file explains itself. Related settings sit together and naming is consistent, so you can read how your cluster is configured without knowing which of our components picks up which key.
  • It is the first step towards self-service. A format this well defined is what we need to safely hand you more of the configuration, so you can change things yourself instead of asking us.

What you need to do

  • Nothing on your side. We rewrite each file for you, keeping your comments and the layout of the file, and we check that the change makes no difference to your infrastructure. You will see it come in as a pull request in your repository.

What’s next

  • Documentation of the format itself: the schema and full examples, so you can look up every available setting.
  • Clear versioning: future changes to the format come as a new version with a migration, instead of one-off key renames.
  • More self-service around cluster configuration changes, now that the format is well defined.