Efficient rate limiting with a growing distributed cluster is hard. These are the challenges we faced when building a horizontally scalable rate limiter.

Ably is a distributed realtime messaging platform. To ensure load and data are distributed evenly and consistently we use consistent hashing algorithms.

In this article we explore the juxtaposition of a Pub/Sub fan-out pattern and a Queue based pattern and why both are sometimes needed.

How Ably's engineers overcome one of the hardest aspects of a globally-distributed realtime messaging service: connection state recovery and stream continuity.