How to Diagnose and Fix WebSocket 1006 Errors in FastAPI LLM Streaming

FastAPI WebSocket connections can drop with a 1006 error during long LLM inference tasks, leaving clients disconnected before a response is delivered. Contrary to a widely circulated claim, FastAPI does not impose a built-in 10-second timeout; the real culprits are often a blocked event loop, Uvicorn's ping/pong defaults, or proxy idle timeouts. Running synchronous LLM calls inside async handlers freezes the event loop, preventing ping/pong frames from being processed and triggering disconnects. Developers can resolve this by using truly async clients or offloading blocking calls via asyncio.to_thread(), while also tuning Uvicorn and infrastructure-level timeouts such as NGINX or AWS ALB settings. Diagnosing the exact failure layer through timing analysis is recommended before making any configuration changes.