Distributed
Token Bucket

Make rate limiting atomic.

The limiter is designed for concurrent application instances where a local counter is not enough. Redis becomes the shared state, while a Lua script keeps refill, consume and retry calculation together.

The token-bucket model makes the behaviour intuitive: requests spend tokens, tokens refill over time, and exhausted buckets return a predictable retry path. The design keeps the decision close to the shared state rather than scattering it across instances.