Designing a Kubernetes Operator

Kubernetes Operators encode operational knowledge into software — automating deployment, scaling, and management of complex stateful applications.

What is an Operator?

An Operator = Custom Resource Definition (CRD) + Custom Controller

Built-in K8s:
  Deployment → Controller ensures N pods running

Operator:
  PostgreSQLCluster → Controller ensures DB cluster healthy
  ElasticsearchCluster → Controller manages shards/replicas

Custom Resource Definition (CRD)

Extend Kubernetes API with custom types:

apiVersion: db.example.com/v1
kind: PostgreSQLCluster
metadata:
  name: my-postgres
spec:
  replicas: 3
  version: "16"
  storage: "50Gi"
  backup:
    schedule: "0 2 * * *"
    destination: "s3://backups/"

Controller Reconciliation Loop

func (r *Reconciler) Reconcile(req Request) (Result, error) {
  // 1. Fetch current state
  cluster := &PostgreSQLCluster{}
  r.Get(ctx, req.NamespacedName, cluster)

  // 2. Calculate desired state
  desired := r.buildDesiredState(cluster)

  // 3. Reconcile diff
  if diff := calculateDiff(current, desired); diff != nil {
    r.applyChanges(diff)
  }

  // 4. Update status
  cluster.Status.Replicas = current.ReadyReplicas
  r.Status().Update(ctx, cluster)

  return Result{RequeueAfter: 30 * time.Second}, nil
}

Operator Capabilities

Level 1: Basic install
Level 2: Upgrades
Level 3: Lifecycle management
Level 4: Deep insights
Level 5: Auto-pilot (self-healing, auto-tuning)

Example Behaviors

PostgreSQL Operator:
  - Automatic leader election
  - Streaming replication setup
  - Automatic failover
  - Scheduled backups
  - Point-in-time recovery
  - Configuration management

Tools

Operator SDK (Red Hat/Mask)
Kubebuilder (official)
COPY (cluster-api)

Conclusion

Operators are the pattern for managing stateful workloads on Kubernetes. They encode operational expertise that would otherwise require manual SRE intervention.