Running Postgres in Kubernetes without losing your mind

Datadog's engineering team published a detailed walkthrough of running PostgreSQL with high availability on Kubernetes. The post covers the hard parts: keeping the primary and replicas in sync, handling failover when the primary drops, and making sure queries don't get lost when a pod restarts.

It's not a trivial problem. Postgres was built to run on a single machine with direct access to disk. Kubernetes treats each pod as disposable. You have to reconcile those two worlds — which means choosing between managed services, operator-based setups, or rolling your own replication. The article walks through the trade-offs and what Datadog chose.

This is the kind of infrastructure knowledge that tends to live in senior engineers' heads. When the team writes it down, it helps anyone else building the same thing, especially teams that don't have the luxury of a dedicated database ops person.

Why this matters for us: every time a team gets better at running Postgres in production, our apps run faster and the people who depend on them — our customers, our families, our communities — experience less downtime.