mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-08 08:08:37 +00:00
openai SDK 3 reads over httpx2 and, from 3.25, wraps transport errors in APIConnectionError or APITimeoutError. The stream wrapper only knew httpx, so a connection drop after finish_reason raised MidStreamFallbackError and a read timeout fell back as a 500. Transport errors are now classified across httpx, httpx2 and the SDK's explicit cause chain, a drop after finish_reason ends the stream on both the sync and async paths, and file_content(stream=True) raises litellm.Timeout or litellm.APIConnectionError instead of raw transport errors |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_aiohttp_cleanup_closed.py | ||
| test_aiohttp_handler.py | ||
| test_aiohttp_so_keepalive.py | ||
| test_aiohttp_transport.py | ||
| test_asgi_handler.py | ||
| test_async_client_cleanup.py | ||
| test_container_handler.py | ||
| test_credential_leak_prevention.py | ||
| test_gemini_session_leak.py | ||
| test_http_handler.py | ||
| test_llm_http_handler.py | ||
| test_mock_transport.py | ||
| test_transport_errors.py | ||