Scaling WebSockets to 100K Concurrent Connections

WebSockets are essential for real-time features but create unique scaling challenges. Unlike HTTP, each connection is persistent and stateful. At 100K concurrent connections, every inefficiency multiplies.

Challenges We Solved

• Memory — Each connection consumes memory; we optimized per-connection overhead to under 4KB
• Load balancing — Sticky sessions with graceful connection draining during deploys
• Fan-out — Publishing to thousands of connections efficiently using Redis pub/sub
• Reconnection — Clients reconnect with backoff and resume from last-seen event ID
• Health checks — Heartbeat pings to detect and clean up zombie connections

Our WebSocket infrastructure now handles 100K concurrent connections per node with p99 message delivery latency under 50 milliseconds.