Configurable namingStrategy for KCP, MD and MP in classy Clusters

[sbueringer]

Some context:

• In clusters that are not using ClusterClass users can freely decide on how to name KCP, MD and MP objects.
• This allows users to follow naming conventions for KCP / MD / MS / Machine objects and eventually infrastructure resources
  • Note: In CAPI it is generally undefined how or if a Machine name maps to a specific infrastructure resource name. Some provider might guarantee a certain mapping or even use the Machine name directly for a server.
• With ClusterClass today we generate names as follows:
  • KCP: {cluster-name}-{random-suffix}
  • MD: {cluster-name}-{md-topology-name}-{random-suffix}
  • MP: {cluster-name}-{mp-topology-name}-{random-suffix}

As a management cluster operator I would like to be able to configure how KCP, MD and MP objects are named in classy Clusters.

Note: This is also important to reach feature parity between classy and non-classy Clusters, so folks that have to follow naming conventions can start using ClusterClass.

Proposal

The proposal is now basically to make the currently hard-coded name formats configurable.

Per default the behavior stays the same as today.
Additionally users can customize the names by providing a namingStrategy on the ClusterClass

Example:

apiVersion: cluster.x-k8s.io/v1beta1
kind: ClusterClass
spec:
  controlPlane:
    namingStrategy:
      name: "{{ .cluster.name }}-{{ .randomSuffix }}"
  workers:
    machineDeployments:
    - class: "default-worker"
      namingStrategy:
        name: "{{ .cluster.name }}-{{ .machineDeployment.topologyName }}-{{ .randomSuffix }}"
   # Same for MachinePools

Notes:

• the names are simple go templates
• they will get some data that can be used for rendering that should be ~ aligned to builtin variables for ClusterClass (although we only provide a very small subset of those variables as things like e.g. MD replicas or version don't make sense)
• Existing objects are not affected, we will only generate a name once per object on creation
• If we're getting close to the max name lengths we should follow the same pattern as the SimpleNameGenerator to ensur the last few characters are a random string.
• Required changes: ClusterClass API type, name calculation, e2e test coverage (just set namingStrategies in one of our test ClusterClasses)
• We would like to propose to cherry-pick this back into v1.5 (based on "Enhancements or additions to experimental Cluster API features" (CONTRIBUTIN.md)

[sbueringer]

/triage accepted

/assign @chrischdi
/assign @killianmuldoon

[chrischdi]

@sbueringer this is done 🥳

via:

• ✨ ClusterClass: Introduce NamingStrategy and allow generating names using go templates #9340
• 🐛 topology: fix namingstrategy webhook to not use uppercase characters for testing the template and align unit test to e2e test #9425
• [release-1.5] ✨ ClusterClass: Introduce NamingStrategy and allow generating names using go templates #9428

And docs:

• 📖 add documentation about namingStrategy in ClusterClasses #9433
• [release-1.5] 📖 add documentation about namingStrategy in ClusterClasses #9435

/close

[k8s-ci-robot]

@chrischdi: Closing this issue.

In response to this:

@sbueringer this is done 🥳

via:

• ✨ ClusterClass: Introduce NamingStrategy and allow generating names using go templates #9340
• 🐛 topology: fix namingstrategy webhook to not use uppercase characters for testing the template and align unit test to e2e test #9425
• [release-1.5] ✨ ClusterClass: Introduce NamingStrategy and allow generating names using go templates #9428

And docs:

• 📖 add documentation about namingStrategy in ClusterClasses #9433
• [release-1.5] 📖 add documentation about namingStrategy in ClusterClasses #9435

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.