The smallest useful operator I've ever written.

The kubebuilder scaffolding for a new operator is 47 files and 1,800 lines of code before you write any logic. For two years that fact alone kept me from solving a problem that needed an operator.

Then a colleague pointed me at a tiny helper library, and I solved the problem in 40 lines.

The problem

We had a recurring class of bug: a small set of namespaces needed a specific NetworkPolicy applied to every new pod, but the policy depended on a label that wasn’t always present at create time. Manually fixing it was a one-line kubectl patch. The team did it about three times a week.

Three times a week, for two years, is around 300 patches. We are not, as a species, good at compounding annoyances.

The operator

package main

import (
  "context"
  corev1 "k8s.io/api/core/v1"
  "sigs.k8s.io/controller-runtime/pkg/client"
  ctrl "sigs.k8s.io/controller-runtime"
)

func main() {
  mgr, _ := ctrl.NewManager(ctrl.GetConfigOrDie(), ctrl.Options{})
  ctrl.NewControllerManagedBy(mgr).
    For(&corev1.Pod{}).
    Complete(reconcile.Func(func(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
      // ensure the policy label; create the NetworkPolicy if missing.
      return ctrl.Result{}, nil
    }))
  mgr.Start(ctrl.SetupSignalHandler())
}

The actual logic is boring. The lesson is that the scaffolding was the problem, not the operator. Forty lines of Go and the recurring annoyance has been gone for six months.

The lesson

If your platform team has a list of “small wins we’ll get to later,” it is almost always the scaffolding tax that is keeping them there. Find a smaller hammer.