A flash sale goes live on a popular booking platform. Fifty dollars off the Friday evening flight from New York to London. Within seconds, a thousand Lambdas wake up and each one runs: UPDATE flight_inventory SET seats_available = seats_available - 1 WHERE flight_id = 'NYC-LON-FRI' AND fare_class = 'Y' ; Enter fullscreen mode Exit fullscreen mode Your p99 latency jumps from 50ms to 30 seconds. No code changed. No deploy. CPU is fine. Connections look healthy-ish. What happened? The answer is locks . If your mental model of database locking is fuzzy, serverless will find every gap and turn it into a pager alert. This post is the foundation: pessimistic vs optimistic, what isolation levels actually lock, and why short-lived functions amplify the pain. We'll use Postgres and TypeScript throughout, and the running example is a booking platform — flights, seats, fare classes — because most of the interesting concurrency problems show up there in their natural habitat.…