We built a real-time notification system for a client dashboard. WebSockets on localhost: instant updates, zero issues. Deployed to production: connections established, then died after exactly 60 seconds.

What We Ruled Out First

Not a code bug—the same build worked locally. Not authentication—the handshake completed successfully. Not SSL—wss:// connected fine.

The Real Problem

Our load balancer had a default idle timeout of 60 seconds. WebSocket connections are long-lived HTTP upgrades. The LB treated them as idle HTTP connections and cut them off.

The Fix

We increased the load balancer idle timeout for WebSocket routes and enabled sticky sessions so reconnections hit the same backend instance. We also added client-side heartbeat pings every 30 seconds as a safety net.

Lesson

Infrastructure defaults are designed for request/response traffic, not persistent connections. Always check proxy, load balancer, and CDN timeout settings when deploying WebSockets or SSE.