Postgres Summit US2026

Over the past decade I've run Postgres at several companies in three different ways. Self-managed on EC2, managed on RDS, and globally replicated on Aurora Global Database. Each model promised to fix the last one's problems, and each brought new ones of its own.

This talk compares all three from the perspective of the person carrying the pager. I'll cover what you give up when you lose superuser, where RDS guardrails help and where they get in the way, how Aurora's storage layer changes failover and replication behavior in practice, and what the cost curves really look like once you're past the marketing.

I'll close with the decision framework we landed on after getting it wrong a few times, including how to defend your choice to leadership and to your on-call rotation.