Issue evidence · AI API · BerriAI/litellm

_read_request_body silently returns {} on body parse failure, causing ProxyModelNotFoundError: model=None

GitHub issue: https://github.com/BerriAI/litellm/issues/32114

Not reproduced

Large embedding requests intermittently fail body parsing, return {} and then raise model=None.

What Badgr ran

Ran LiteLLM 1.104.2 as a proxy with a local embedding mock and sent 96 requests of about 2.0 MB each, 16 at a time.

What came back

  • 96 of 96 returned HTTP 200; the mock received all bodies intact (1,949,425 bytes each).

The reporter says it is intermittent and tied to pod restarts and many concurrent RAG workers, and could not reproduce it with a single curl either, so this is not evidence it cannot happen.

Checked 2026-10-15 in Badgr’s local development environment.

← All issue evidence