Cloud-Native Microservices With Kubernetes - 2nd Edition

Having a Minimum Viable PostgreSQL Cluster

We created the PostgreSQL Pods using a StatefulSet; each one of them is distinguished by a stable network identity and can be accessed directly using the Headless Service. Each Pod has its own PersistentVolumeClaim to store its data persistently. Are we done? Not yet. This setup allows us to have multiple PostgreSQL instances running, but they are not yet configured to work together as a primary-replica cluster.

Kubernetes helps us manage the lifecycle of the Pods and their storage, but it doesn't handle the database-level configuration needed for replication, failover, leader election, and backups. We need to set up all of these aspects manually or use additional tools.

Here is a brief overview of what needs to be done next:

1. Pick a primary Pod — usually postgres-0.

2. Initialize its database.

3. Enable streaming replication settings (wal_level, max_wal_senders, etc.).

4. Create a replication user and allow remote replication connections.

5. Prepare the replicas (e.g., postgres-1, postgres-2).

6. Clear their data directories.

7. Take a base backup from the primary.

8. Configure them to start in standby mode and connect to the primary for WAL streaming.

9. Networking setup.

10. Use the headless Service to provide stable DNS names for each Pod.

11. Each replica connects to the primary using its stable hostname (postgres-0...svc.cluster.local).

12. Start all Pods.

13. Verify that replicas are streaming data from the primary and staying in sync.

14. Expose connection endpoints.