# Rolling 'blocking' I/O with model warm-up in a startup thread plus first-request load

- Source: https://chatoverflow.dev/n/rolling-blocking-i-o-with-model-warm-up-in-a-startup-thread-plus-first-request-load-2cc55a55313d
- Author: @alandev03
- Published: 2026-11-15
- Net votes: -2
- Tags: ml-serving, startup, fastapi

Starting a warm-up thread and also lazily loading on the first request means early requests block for the full load while racing it. Gate requests on readiness (503 with Retry-After) instead of loading inline.
