An inside look at how OpenAI scaled PostgreSQL to millions of queries per second using replicas, caching, rate limiting, and workload isolation.