Q18: Deployments vs. ReplicaSets vs. StatefulSets: When to use StatefulSets and how headless services work. - DevOps & Kubernetes (50 Qs)

Workload Controllers Architecture

Kubernetes encapsulates container management into higher-level declarative controllers tailored for stateless versus stateful systems.

scss

Stateless Architecture (Deployment)
[ Deployment ] ---> [ ReplicaSet ] ---> [ Pod-abc1 ] (Interchangeable, random IDs)
                                   ---> [ Pod-xyz2 ] (Dynamic IP, shared PV or stateless)

Stateful Architecture (StatefulSet)
[ StatefulSet ] ---> [ Pod-0 ] <---> [ PVC-0 ] <---> [ Dedicated PV 0 ] (Sticky Storage)
                ---> [ Pod-1 ] <---> [ PVC-1 ] <---> [ Dedicated PV 1 ]
                ---> [ Pod-2 ] <---> [ PVC-2 ] <---> [ Dedicated PV 2 ]
                        ^
                        | (Ordered Deployment: 0 -> 1 -> 2)
              [ Headless Service (clusterIP: None) ]
              DNS: pod-0.db-service.default.svc.cluster.local

Key Differences

Feature	Deployment	StatefulSet
Identity	Ephemeral, random hash suffixes (e.g., `api-7bf89d-4kx2p`).	Deterministic, sequential index (e.g., `redis-0`, `redis-1`).
Storage	Pods share the same PVC or use ephemeral storage.	Each Pod gets its own dedicated PVC via `volumeClaimTemplates`.
Startup / Teardown	Scaled in parallel.	Scaled sequentially (0 → 1 → 2; teardown 2 → 1 → 0).
Networking	Standard load-balanced ClusterIP service.	Headless Service (`clusterIP: None`) providing direct A-records for each Pod.

Headless Service Blueprint for Distributed Clusters (Kafka/ZooKeeper/Cassandra)

yaml

apiVersion: v1
kind: Service
metadata:
  name: postgres-cluster
spec:
  clusterIP: None # Headless: returns Pod IPs directly in DNS queries
  selector:
    app: postgres
  ports:
    - port: 5432
      name: postgres