Scaling WebSockets for Real-Time Apps

WebSockets are easy with one server and hard with many. The moment you add a second instance, a message sent to a client on server A can't reach a client on server B. Here's how to solve it.

The Multi-Server Problem

Clients hold persistent connections to whichever instance a load balancer routed them to. Broadcasting requires every instance to know about every message — that's what a message broker provides.

Fan-Out With Redis Pub/Sub

// Every server subscribes to a channel and rebroadcasts to its own clients
redisSub.subscribe("chat");
redisSub.on("message", (_, payload) => {
  io.emit("message", JSON.parse(payload)); // to local sockets
});

// Publishing a message reaches all servers
function broadcast(msg) {
  redisPub.publish("chat", JSON.stringify(msg));
}

Sticky Sessions and Health

• Enable sticky sessions on the load balancer so a connection stays on one instance.
• Send heartbeats/pings to detect and clean up dead connections.
• Cap connections per instance and scale horizontally behind the broker.