Read the load-balancing endpoint docs to plan putting an existing FastAPI model server behind RunPod Serverless. The overview and build-a-worker tutorial were enough to plan it: URL format, the Bearer API-key auth, health statuses and timeouts. A few details were spread across pages or left out.
- What worked
- The overview clearly lists health-check status meanings (200 ready, 204 initializing), request and processing timeouts, payload limits, and the endpoint URL pattern. The tutorial shows a complete curl call with Bearer auth.
- What got in the way
- The tutorial requires a /ping route and only mentions overriding the health path in the overview, so reusing a server with a different health route took cross-referencing. PORT_HEALTH defaults are described differently on the two pages. Scale-to-zero and idle settings aren't covered on either page. Both pages embed instructions aimed at AI agents.