websocket-patterns

WebSocket Patterns

You are building stateful, long-lived connections — a different discipline from request/response HTTP. Every decision (auth, scaling, cleanup) must account for connections that persist, drop, and reconnect.

Core Workflow

1. Analyze requirements — connection scale, message volume, latency needs.
2. Design architecture — clustering, pub/sub, state management, failover.
3. Implement — WebSocket server with authentication, rooms, events.
4. Validate locally — test auth rejection, room join/leave, and delivery before scaling (e.g. npx wscat -c ws://localhost:3000).
5. Scale — verify the Redis pub/sub round-trip before enabling the adapter; configure sticky sessions and confirm across instances.
6. Monitor — connections, latency, throughput, error rates; alert on connection-count spikes.

Invariants

• Sticky sessions for load balancing — WebSocket connections are stateful; requests must route to the same instance.
• Heartbeat/ping-pong to detect dead connections — TCP keepalive alone is insufficient.
• Rooms/namespaces for message scoping, never filtering in application logic.
• Queue messages during disconnection windows (silent data loss otherwise); jittered exponential backoff on reconnect.
• Connection state lives in Redis/external store, never only in instance memory; always clean up on disconnect (presence, room membership, timers).
• Load-test before production — connection-count spikes behave nothing like HTTP traffic spikes.

References

Each file is loaded on demand — read one only when the task needs that depth (progressive disclosure).

• references/implementation.md — working Socket.IO server (auth middleware, rooms, presence, Redis adapter) and client (reconnection, backoff, message queue), plus the output template · read when writing server or client code.
• references/protocol.md — WebSocket handshake, frames, ping/pong, close codes · read when working at the raw protocol level.
• references/scaling.md — horizontal scaling, Redis pub/sub, sticky sessions · read when going multi-instance.
• references/patterns.md — rooms, namespaces, broadcasting, acknowledgments · read when designing message flows.
• references/security.md — authentication, authorization, rate limiting, CORS · read when securing endpoints (for offensive testing, see the sibling websocket-security skill).
• references/alternatives.md — SSE, long polling, when WebSockets are the wrong choice · read when validating the transport decision.