We scaled PgBouncer to 4x throughput
Summary
The article explains how ClickHouse Managed Postgres scales PgBouncer by running a fleet of processes across CPU cores with so_reuseport, turning multiple cores into parallel poolers behind a single endpoint. Benchmark results show the fleet delivering about 4x higher throughput than a single process, with detailed observations on CPU usage and connection limits. It also covers cancellation behavior across the fleet and practical guidance for sizing and configuration.