Deploying Databases in Kubernetes

# CHAPTER 15

Deploying Databases in Kubernetes #

1. Introduction #

2. Learning Objectives #

• Understand the unique challenges of running Databases in Kubernetes.

• Contrast a standard Deployment with a StatefulSet.

• Understand stable network identities (Headless Services).

• Deploy a MySQL database using a StatefulSet and PVC.

• Discuss the merits of Managed Cloud Databases (RDS) vs. Kubernetes databases.

3. Beginner-Friendly Explanation #

• Deployment (Stateless): You manage a fleet of generic white vans. If Van #4 breaks down, you buy a new white van. You do not care about the license plate or the name of the van, as long as you have 10 vans driving. (This is perfect for web servers).

• StatefulSet (Stateful): You manage a fleet of armored bank trucks. Truck #1 strictly carries gold. Truck #2 strictly carries diamonds. If Truck #1 breaks down, you cannot just replace it with a generic van. You need a highly specific replacement truck that instantly inherits the identity, the security clearance, and the specific cargo (The PVC) of the original Truck #1.

A StatefulSet guarantees strict identity and strict storage mapping.

4. The StatefulSet Controller #

• Ordered Naming: If you deploy a StatefulSet named mysql with 3 replicas, it doesn't create random names like mysql-8f7b5. It creates exactly: mysql-0, mysql-1, and mysql-2.

• Ordered Creation/Deletion: It starts them in order. It waits for mysql-0 to be fully healthy before starting mysql-1.

• Sticky Storage: This is the most critical feature. It dynamically provisions a unique Persistent Volume Claim (PVC) for *each* specific Pod. If mysql-1 crashes, Kubernetes creates a new mysql-1 and rigorously attaches the *exact same hard drive* back to it.

5. Headless Services #

6. Anatomy of a StatefulSet YAML #

apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: mysql
spec:
  serviceName: "mysql-service" # Required: Points to the Headless Service
  replicas: 1
  selector:
    matchLabels:
      app: mysql
  template:
    metadata:
      labels:
        app: mysql
    spec:
      containers:
      - name: mysql
        image: mysql:8.0
        env:
          - name: MYSQL_ROOT_PASSWORD
            value: "supersecret"
        volumeMounts:
        - name: data
          mountPath: /var/lib/mysql
  volumeClaimTemplates:
  - metadata:
      name: data
    spec:
      accessModes: [ "ReadWriteOnce" ]
      resources:
        requests:
          storage: 10Gi

7. Mini Project: Deploy a Persistent MySQL DB #

Step-by-Step Tutorial:

1. 1. Save the YAML from Section 6 into stateful-mysql.yaml.

1. 2. First, we must create the Headless Service required by the StatefulSet. Create mysql-svc.yaml:

apiVersion: v1
kind: Service
metadata:
  name: mysql-service
spec:
  clusterIP: None
  selector:
    app: mysql

1. 3. Apply both files:

kubectl apply -f mysql-svc.yaml
kubectl apply -f stateful-mysql.yaml

1. 4. Watch the Pod creation: kubectl get pods -w. You will notice it boots with the strict name mysql-0.

1. 5. Verify the dynamic storage creation: kubectl get pvc. You will see Kubernetes automatically generated a PVC named data-mysql-0.

1. 6. The Test: Delete the Pod! kubectl delete pod mysql-0.

1. 7. Watch kubectl get pods. Kubernetes will instantly recreate it, name it mysql-0 again, and securely remount the data-mysql-0 hard drive to it. Total data persistence achieved!

8. Real-World Scenarios #

9. Best Practices #

• The Operator Pattern: Manually managing a 3-node PostgreSQL cluster (handling leader election, failovers, and backups) using raw StatefulSets is agonizing. In production, you must use the Operator Pattern. You install a third-party software (like the Zalando Postgres Operator) into your cluster. The Operator acts as a robot Database Administrator, completely automating the complex lifecycle of the database cluster.

10. Common Mistakes #

• Scaling Down a StatefulSet: If you scale a StatefulSet down from 3 replicas to 2, Kubernetes deletes the mysql-2 Pod. Crucially, it DOES NOT delete the PVC (data-mysql-2)! This is a safety mechanism to prevent accidental data loss. You must manually run kubectl delete pvc data-mysql-2 to stop paying your cloud provider for that physical hard drive.

11. Exercises #

1. 1. Contrast the naming conventions and creation ordering of Pods managed by a Deployment versus Pods managed by a StatefulSet.

1. 2. Explain the function of a Headless Service. Why is random load balancing detrimental to a Master-Replica database architecture?

12. FAQs #

13. Interview Questions #

• Q: Detail the architectural necessity of the volumeClaimTemplates specification within a StatefulSet manifest. How does this differ from manually defining a static PVC in a Deployment?

• Q: A CTO asks for your architectural recommendation: Deploy the primary transactional PostgreSQL database inside the company's existing Kubernetes cluster, or utilize a managed cloud service like AWS RDS. Present a balanced argument highlighting the operational overhead of both approaches.

14. Summary #

15. Next Chapter Recommendation #