From d3dab8e294b06badb2d34f31ce6aade9380a49ea Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Tue, 1 Sep 2026 12:46:04 -0700 Subject: [PATCH 1/2] fix(rerank): map provider errors with the resolved provider on sync and async paths --- litellm/rerank_api/main.py | 22 ++++++-- tests/test_litellm/rerank_api/test_main.py | 61 ++++++++++++++++++++++ 2 files changed, 80 insertions(+), 3 deletions(-) diff --git a/litellm/rerank_api/main.py b/litellm/rerank_api/main.py index c8f7842aebf..597d1cfb863 100644 --- a/litellm/rerank_api/main.py +++ b/litellm/rerank_api/main.py @@ -43,10 +43,17 @@ async def arerank( """ Async: Reranks a list of documents based on their relevance to the query """ + _custom_llm_provider: str | None = None # rebind-ok: set by the get_llm_provider unpack; read in the except try: loop: Final = asyncio.get_event_loop() kwargs["arerank"] = True + _, _custom_llm_provider, _, _ = litellm.get_llm_provider( # rebind-ok: see pre-declaration above + model=model, + custom_llm_provider=custom_llm_provider, + api_base=kwargs.get("api_base", None), + ) + func: Final = partial( rerank, model, @@ -70,7 +77,11 @@ async def arerank( response = init_response return response except Exception as e: - raise e + raise exception_type( + model=model, + custom_llm_provider=_custom_llm_provider or custom_llm_provider, + original_exception=e, + ) @client @@ -115,6 +126,7 @@ def rerank( model_info: Final = kwargs.get("model_info", None) user: Final = kwargs.get("user", None) client: Final = kwargs.get("client", None) + _custom_llm_provider: str | None = None # rebind-ok: set by the get_llm_provider unpack; read in the except try: _is_async: Final = kwargs.pop("arerank", False) is True optional_params: Final = GenericLiteLLMParams(**kwargs) @@ -127,7 +139,7 @@ def rerank( ( model, - _custom_llm_provider, + _custom_llm_provider, # rebind-ok: see pre-declaration above dynamic_api_key, dynamic_api_base, ) = litellm.get_llm_provider( @@ -538,4 +550,8 @@ def rerank( return response except Exception as e: verbose_logger.error("Error in rerank: %s", e) - raise exception_type(model=model, custom_llm_provider=custom_llm_provider, original_exception=e) + raise exception_type( + model=model, + custom_llm_provider=_custom_llm_provider or custom_llm_provider, + original_exception=e, + ) diff --git a/tests/test_litellm/rerank_api/test_main.py b/tests/test_litellm/rerank_api/test_main.py index 587be59c550..62149c742d6 100644 --- a/tests/test_litellm/rerank_api/test_main.py +++ b/tests/test_litellm/rerank_api/test_main.py @@ -111,6 +111,67 @@ def test_together_rerank_honors_api_base(respx_mock: respx.MockRouter): assert mock_route.calls[0].request.headers["authorization"] == "Bearer fake-together-key" +DASHSCOPE_404_BODY = { + "error": { + "message": "The model `does-not-exist` does not exist or you do not have access to it.", + "type": "invalid_request_error", + "param": None, + "code": "model_not_found", + }, + "request_id": "mock-request-id", +} + + +def test_rerank_error_names_provider_and_keeps_body(respx_mock: respx.MockRouter, monkeypatch): + """Regression for the rerank error path mapping with the unresolved provider param: + a provider 404 surfaced as 'None - ' instead of naming the provider and its error body.""" + monkeypatch.delenv("DASHSCOPE_API_BASE", raising=False) + monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False) + + mock_route = respx_mock.post("https://dashscope.example/v1/reranks") + mock_route.return_value = httpx.Response(404, json=DASHSCOPE_404_BODY) + + with pytest.raises(litellm.NotFoundError) as exc_info: + litellm.rerank( + model="dashscope/does-not-exist", + query=MARKER_QUERY, + documents=[MARKER_DOC], + api_key="fake-dashscope-key", + api_base="https://dashscope.example/v1", + ) + + assert mock_route.called + assert "DashscopeException" in str(exc_info.value) + assert "does not exist or you do not have access to it" in str(exc_info.value) + assert "None - " not in str(exc_info.value) + + +@pytest.mark.asyncio +async def test_arerank_error_is_mapped_to_litellm_exception(respx_mock: respx.MockRouter, monkeypatch): + """Regression for arerank's bare re-raise: provider errors escaped as raw + provider exception classes instead of the mapped litellm exception contract.""" + monkeypatch.delenv("DASHSCOPE_API_BASE", raising=False) + monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False) + monkeypatch.setenv("DISABLE_AIOHTTP_TRANSPORT", "True") + + mock_route = respx_mock.post("https://dashscope.example/v1/reranks") + mock_route.return_value = httpx.Response(404, json=DASHSCOPE_404_BODY) + + with pytest.raises(litellm.NotFoundError) as exc_info: + await litellm.arerank( + model="dashscope/does-not-exist", + query=MARKER_QUERY, + documents=[MARKER_DOC], + api_key="fake-dashscope-key", + api_base="https://dashscope.example/v1", + ) + + assert mock_route.called + assert "DashscopeException" in str(exc_info.value) + assert "does not exist or you do not have access to it" in str(exc_info.value) + assert "None - " not in str(exc_info.value) + + @pytest.mark.asyncio async def test_together_rerank_async_honors_env_api_base(respx_mock: respx.MockRouter, monkeypatch): """Regression: TOGETHER_AI_API_BASE was honored by chat but ignored by rerank.""" From d59fcda8af69f5545a8e7c29b29d12362e2bffad Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Tue, 1 Sep 2026 14:47:59 -0700 Subject: [PATCH 2/2] fix(rerank): adopt declared authenticating providers in arerank instead of resolving them get_llm_provider runs the OAuth device flow for github_copilot and chatgpt, so calling it on the event loop before the executor dispatch let an authenticated caller block the loop for the length of the polling window. Adopt the declared provider via declared_authenticating_provider, matching the metadata callers in utils.py, and only resolve for everything else. --- litellm/rerank_api/main.py | 19 +++++++++---- tests/test_litellm/rerank_api/test_main.py | 32 ++++++++++++++++++++++ 2 files changed, 45 insertions(+), 6 deletions(-) diff --git a/litellm/rerank_api/main.py b/litellm/rerank_api/main.py index 597d1cfb863..37ca989b8d3 100644 --- a/litellm/rerank_api/main.py +++ b/litellm/rerank_api/main.py @@ -6,6 +6,7 @@ from typing import Any, Final, Literal import litellm from litellm._logging import verbose_logger +from litellm.litellm_core_utils.get_llm_provider_logic import declared_authenticating_provider from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.base_llm.rerank.transformation import BaseRerankConfig from litellm.llms.bedrock.rerank.handler import BedrockRerankHandler @@ -43,16 +44,22 @@ async def arerank( """ Async: Reranks a list of documents based on their relevance to the query """ - _custom_llm_provider: str | None = None # rebind-ok: set by the get_llm_provider unpack; read in the except + _custom_llm_provider: str | None = ( + None # rebind-ok: set by the declared-provider guard or the get_llm_provider unpack; read in the except + ) try: loop: Final = asyncio.get_event_loop() kwargs["arerank"] = True - _, _custom_llm_provider, _, _ = litellm.get_llm_provider( # rebind-ok: see pre-declaration above - model=model, - custom_llm_provider=custom_llm_provider, - api_base=kwargs.get("api_base", None), - ) + declared_provider: Final = declared_authenticating_provider(model, custom_llm_provider) + if declared_provider is not None: + _custom_llm_provider = declared_provider # rebind-ok: see pre-declaration above + else: + _, _custom_llm_provider, _, _ = litellm.get_llm_provider( # rebind-ok: see pre-declaration above + model=model, + custom_llm_provider=custom_llm_provider, + api_base=kwargs.get("api_base", None), + ) func: Final = partial( rerank, diff --git a/tests/test_litellm/rerank_api/test_main.py b/tests/test_litellm/rerank_api/test_main.py index 62149c742d6..2b6cfeda2c2 100644 --- a/tests/test_litellm/rerank_api/test_main.py +++ b/tests/test_litellm/rerank_api/test_main.py @@ -172,6 +172,38 @@ async def test_arerank_error_is_mapped_to_litellm_exception(respx_mock: respx.Mo assert "None - " not in str(exc_info.value) +@pytest.mark.asyncio +async def test_arerank_declared_authenticating_provider_skips_resolution(monkeypatch): + """Regression for the event-loop hazard in arerank's provider pre-resolution: + get_llm_provider runs the blocking OAuth device flow for github_copilot/chatgpt, + so arerank must adopt the declared provider instead of resolving it, while the + except path still maps with that declared provider.""" + from litellm.llms.base_llm.chat.transformation import BaseLLMException + + resolution_calls = [] + + def record_resolution(*args, **kwargs): + resolution_calls.append((args, kwargs)) + return "gpt-4o", "github_copilot", None, None + + def rerank_raises_provider_error(*args, **kwargs): + raise BaseLLMException(status_code=401, message='{"error":"bad key"}') + + monkeypatch.setattr(litellm, "get_llm_provider", record_resolution) + monkeypatch.setattr("litellm.rerank_api.main.rerank", rerank_raises_provider_error) + + with pytest.raises(litellm.AuthenticationError) as exc_info: + await litellm.arerank( + model="github_copilot/gpt-4o", + query=MARKER_QUERY, + documents=[MARKER_DOC], + ) + + assert resolution_calls == [] + assert "Github_copilotException" in str(exc_info.value) + assert "None - " not in str(exc_info.value) + + @pytest.mark.asyncio async def test_together_rerank_async_honors_env_api_base(respx_mock: respx.MockRouter, monkeypatch): """Regression: TOGETHER_AI_API_BASE was honored by chat but ignored by rerank."""