scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-10 22:41:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 82

refactor(proxy): tighten deferred stream logging discard docstring

Browse source 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
jesus 2026-09-08 01:25:36 +00:00
parent 0f9dc5cbcb
commit af60bda196
1 changed files with 6 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
litellm/proxy/utils.py
View file

@ -3279,17 +3279,12 @@ class ProxyLogging:
@staticmethod
def _discard_deferred_stream_logging_for_failure(request_data: dict) -> None:
"""Discard the deferred stream-complete dispatch when the stream ends in
a failure (e.g. an end-of-stream guardrail block raising out of the
callback chain). The deferred dispatch is the success logging path —
firing it here would record the blocked request as a success callback
and a ``status=success`` spend row before the outer generator's
``post_call_failure_hook`` writes the failure row. The CSW shape parks
``(assembled ModelResponse, cache_hit)``; record its partial usage so
the failure row bills what the stream consumed instead of zero. The
native /v1/messages and responses shapes park ``(coroutine,)`` and
still need the flush (no success row is produced without it), so they
keep the existing fire behaviour.
"""Drop the parked success dispatch when the stream ends in an exception.
The CSW shape parks ``(assembled ModelResponse, cache_hit)``: its usage is
carried onto the logging object so the failure row bills what the stream
consumed. The native /v1/messages and responses shapes park a logging
coroutine with no recoverable usage, so they keep firing as before.
"""
logging_obj: Final = request_data.get("litellm_logging_obj")
if logging_obj is None: