A single PostgreSQL node is an outage waiting to happen. Here is how to architect an automated failover cluster using Patroni, etcd, and HAProxy that recovers in under 30 seconds without risking split-brain corruption.