Scaling WebSocket and Realtime Features

Live dashboards feel magical until connection counts and fan-out costs flatten your API tier. Realtime features need backpressure, horizontal scaling, and graceful degradation—not infinite sticky sessions.

Horizontal scaling patterns

• Redis pub/sub or dedicated realtime bus between API instances
• Sticky sessions only when unavoidable; prefer shared subscription state
• Separate websocket gateway from REST API autoscaling profiles
• Cap concurrent connections per user and per tenant

Backpressure and rate limits

@WebSocketGateway({ namespace: '/notifications' })
export class NotificationsGateway {
  @SubscribeMessage('subscribe')
  handleSubscribe(@ConnectedSocket() client: Socket, @MessageBody() room: string) {
    if (client.rooms.size > 20) return { error: 'ROOM_LIMIT' };
    client.join(room);
  }
}

Drop or sample low-priority events before slow clients block the event loop.

Fallback transports

Detect failure and downgrade automatically with client-side retry.

Observability

Track connection churn, message lag p99, and dropped events. Alert when fan-out latency exceeds product SLO.

Realtime is a scaling domain—design for 10× peak connections before marketing announces the live feature.

Load testing realtime paths

Simulate connection storms before marketing events—open 10k connections with realistic subscribe patterns. Measure memory per connection and GC pauses; surprises here become production pages.