Distributed Rate Limiting

Everything in the previous lesson assumed one machine holding one counter. Real systems run many machines behind a load balancer, and that breaks the assumption in a specific way.

The limit multiplies by the number of servers

Suppose the rule is 100 requests per minute per user, and the service runs on 10 servers, each keeping its own counter in its own memory. The load balancer spreads a user's requests across all 10.

Each server sees roughly 10 percent of the traffic, so each one counts to 10 and lets everything through. **The user sends 1,000 requests per minute and is never rejected