Read Replicas + PgBouncer: Scaling Postgres Without Panic

When a database starts throwing errors under load, the instinct is to make it bigger. But often the bottleneck isn't CPU — it's connections. Thousands of short-lived connections exhaust the limit long before the box is busy.

PgBouncer fixes that by pooling and reusing connections, so a flood of clients maps onto a sane number of real database connections. Exhaustion errors basically disappear.

Read replicas attack the other half: read-heavy traffic no longer competes with writes. You route reads to a replica, keep writes on the primary, and watch replication lag as you shift more load over.

Done incrementally — a slice of read traffic at a time — it's a low-drama change with a big payoff: reads scale by adding replicas, independent of writes.