How would you design a highly scalable WebSocket architecture for millions of concurrent users?

A scalable WebSocket architecture uses load balancing with sticky sessions, horizontal scaling, and a pub/sub backbone like Redis or Kafka.

WebSockets are stateful connections, so scaling requires careful design. A common architecture includes: (1) Load balancer with session affinity, (2) multiple WebSocket nodes, (3) a message broker (Redis Pub/Sub, Kafka, or NATS) to sync messages across nodes, and (4) optional API gateway for authentication. This ensures messages from one node are broadcast to all relevant clients connected to other nodes.