Slow Clients and Backpressure

A client on a slow network can't drain messages fast enough, so the server's send buffer grows. Without limits, a few slow clients exhaust memory. Monitor per-connection buffered bytes (bufferedAmount/socket write buffer), apply policies: drop non-critical updates (send only the latest state — conflation), disconnect clients exceeding limits (they reconnect and resync), and keep messages small. For high-frequency data (prices, positions), conflate to the latest value per key.

Slow consumer policies

TypeScript (ws)

const MAX_BUFFER = 1 * 1024 * 1024;   // 1 MB
function sendOrDrop(ws: WebSocket, msg: Message) {
  if (ws.bufferedAmount > MAX_BUFFER) {
    if (msg.critical) return ws.close(4008, "slow consumer");   // client will reconnect and resync
    return metrics.increment("ws.dropped");                    // drop non-critical update
  }
  ws.send(JSON.stringify(msg));
}

// Conflation: keep only the latest price per symbol and flush every 100 ms
const latest = new Map<string, Price>();
setInterval(() => { for (const [, p] of latest) broadcast(p); latest.clear(); }, 100);

Common pitfalls

• Ignoring bufferedAmount.
• Sending every intermediate value of fast-changing data.
• Large JSON messages for frequent updates.

Key points

• Bound per-connection buffers
• Conflate high-frequency data
• Disconnect and resync when needed

Further reading: MDN: WebSocket.bufferedAmount