5 articles
A NAT gateway inbound traffic spike of 100MB/s with no deployment changes. The culprit wasn't the network — it was two overfetching bugs hiding in a Kafka consumer and a MongoDB query, exposed by moving to new infrastructure.
In-memory rate limiters break the moment you scale horizontally. In Part 4 we move state into Redis, introduce a Store interface for flexibility, and implement both sliding window and token bucket algorithms that work correctly across multiple instances.
The sliding window isn't the only way to rate limit. In Part 3 we implement the token bucket algorithm — a different mental model that handles bursts more gracefully.
In Part 1 we built a correct sliding window rate limiter. In Part 2 we fix its biggest weakness — a global lock that blocks real traffic during cleanup at scale.
A practical guide to building a correct, concurrent, memory-safe rate limiter in Go using the sliding window algorithm — with honest tradeoffs included.