The bouncer at the door (rate limiting)

The stall finally gets someone on the door. A practical break from the consistency theory of Splitting the ledger (sharding) and The cut phone line (CAP, done properly), and it pays that theory back at the end.

Before you start. You need one fact about a server: it finishes a fixed amount of work per second, and a request is one job handed to it. And one about networks: one clerk asking another “what is the count now?” and waiting for the answer is a round trip. Counting in your own head takes microseconds; waiting for that answer takes milliseconds, and across the world most of a second.