How would you design a real-time chat application capable of serving one million users?

A real-time chat application serving one million users should be designed as a horizontally scalable, event-driven distributed system where stateless Node.js WebSocket gateways handle persistent client connections, while dedicated services manage authentication, message routing, persistence, notifications, and presence. The architecture should rely on distributed messaging (such as Kafka, NATS, or RabbitMQ), fast in-memory data stores (Redis), partitioned databases, and careful handling of reliability concerns like ordering, retries, backpressure, and failure recovery. The key challenge is not handling one million registered users, but sustaining hundreds of thousands of concurrent WebSocket connections while maintaining low latency, high availability, and predictable scalability.

At this scale, the system must satisfy several requirements simultaneously:

A typical high-level architecture looks like:…