Kubernetes Ep 9: StatefulSets & Stateful Database Deployments

TL;DR (Quick Summary)

#

• StatefulSet vs Deployment:
  • Deployment: Pods get random hashes (app-75b69bd654-2x9lq), share storage claims, and start/stop in parallel.
  • StatefulSet: Pods get persistent ordinal names (db-0, db-1, db-2), unique DNS hostnames, separate PVs via volumeClaimTemplates, and start/stop sequentially.
• Headless Service (clusterIP: None): Disables load balancing VIP so DNS queries directly resolve individual Pod IPs for direct node-to-node replication.

1. StatefulSet Architecture & Ordinal Indexing

#

graph TD
    subgraph StatefulSet["StatefulSet: redis-cluster (replicas: 3)"]
        P0["Pod: redis-cluster-0
DNS: redis-cluster-0.redis-service"]
        P1["Pod: redis-cluster-1
DNS: redis-cluster-1.redis-service"]
        P2["Pod: redis-cluster-2
DNS: redis-cluster-2.redis-service"]
    end

    subgraph VolumeClaimTemplates["volumeClaimTemplates"]
        PVC0["PVC: data-redis-cluster-0"] -->|Binds| PV0["PV 0 (10Gi)"]
        PVC1["PVC: data-redis-cluster-1"] -->|Binds| PV1["PV 1 (10Gi)"]
        PVC2["PVC: data-redis-cluster-2"] -->|Binds| PV2["PV 2 (10Gi)"]
    end

    P0 -.-> PVC0
    P1 -.-> PVC1
    P2 -.-> PVC2

2. Headless Service Manifest (clusterIP: None)

#

A Headless Service does not assign a single Virtual IP. Instead, it creates A-records in CoreDNS for each individual Pod endpoint matching the selector.

Create headless-service.yaml:

apiVersion: v1
kind: Service
metadata:
  name: redis-headless
  namespace: default
  labels:
    app: redis
spec:
  clusterIP: None # Headless Service
  selector:
    app: redis
  ports:
  - port: 6379
    name: redis-port

3. Complete StatefulSet Manifest

#

Create statefulset.yaml:

apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: redis-cluster
  namespace: default
spec:
  serviceName: "redis-headless"
  replicas: 3
  selector:
    matchLabels:
      app: redis
  template:
    metadata:
      labels:
        app: redis
    spec:
      containers:
      - name: redis
        image: redis:7.0-alpine
        ports:
        - containerPort: 6379
          name: redis-port
        volumeMounts:
        - name: redis-data
          mountPath: /data
  # Dynamically creates a PVC for EACH ordinal Pod replica!
  volumeClaimTemplates:
  - metadata:
      name: redis-data
    spec:
      accessModes: [ "ReadWriteOnce" ]
      resources:
        requests:
          storage: 1Gi

Apply both manifests:

kubectl apply -f headless-service.yaml
kubectl apply -f statefulset.yaml

4. Observing Sequential Deployment & DNS Names

#

Watch the Pod creation process in real-time:

kubectl get pods -l app=redis --watch

Sequential Output:

NAME              READY   STATUS              RESTARTS   AGE
redis-cluster-0   0/1     ContainerCreating   0          2s
redis-cluster-0   1/1     Running             0          8s
redis-cluster-1   0/1     ContainerCreating   0          1s
redis-cluster-1   1/1     Running             0          7s
redis-cluster-2   0/1     ContainerCreating   0          1s
redis-cluster-2   1/1     Running             0          6s

Notice how redis-cluster-1 was NOT started until redis-cluster-0 became fully Ready!

Checking Individual DNS Hostnames

#

Each Pod gets a deterministic, predictable network address:

• redis-cluster-0.redis-headless.default.svc.cluster.local
• redis-cluster-1.redis-headless.default.svc.cluster.local
• redis-cluster-2.redis-headless.default.svc.cluster.local

List dynamically created PVCs:

kubectl get pvc -l app=redis

NAME                   STATUS   VOLUME                                     CAPACITY   ACCESS MODES
redis-data-redis-cluster-0   Bound    pvc-11111111-2222-3333-4444-555555555555   1Gi        RWO
redis-data-redis-cluster-1   Bound    pvc-66666666-7777-8888-9999-000000000000   1Gi        RWO
redis-data-redis-cluster-2   Bound    pvc-aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee   1Gi        RWO

5. Summary & Next Steps

#

StatefulSets guarantee ordering, persistent identities, and dedicated storage volumes per replica for database systems.

In Episode 10: DaemonSets & Cluster Node Agents, we will shift gears to node-level background workloads, learning how DaemonSets deploy log collectors and metric agents automatically across every worker node in your cluster!