We run PgBouncer on Kubernetes at Fresha as the “gatekeeper” in front of Postgres, and it became one of those components that’s essential, though it started rough initially.

This is a practical war story about operating PgBouncer at scale on K8s: what broke, what surprised us, and what we had to adapt to make it reliable in production. I’ll cover the real operational challenges(surviving pod evictions, rollouts, config rollouts) and how we hardened the system.

The second half is about upgrades: how we used PgBouncer to execute Postgres version upgrades with effectively zero downtime, what “zero downtime” actually meant in practice with PgBouncer specifically and what tricks we needed to invent for this.

Anton Borisov is a Principal Data Architect at Fresha (London), working on high-scale data and database infrastructure across Postgres and real-time systems. He spends an unhealthy amount of time turning “this should be fun” components into production-grade systems. He speaks and writes about data, streaming, and modern infrastructure, and in his spare time he also deploys dahlias into pots in his garden.