fix(responses): forward client headers on response management routes

litellm/responses/main.py

@@ -653,6 +653,18 @@ def _pop_use_chat_completions_api_kw(kwargs: dict[str, Any]) -> bool:
     return bool(use_cc)
 
 
+def _merge_forwarded_client_headers(
+    extra_headers: dict[str, Any] | None,
+    kwargs: dict[str, Any],
+) -> dict[str, Any] | None:
+    """Merge the proxy's forwarded client headers (`headers` kwarg) into `extra_headers`."""
+    client_headers = kwargs.get("headers")
+    return ResponsesAPIRequestUtils.merge_client_forwarded_headers(
+        extra_headers=extra_headers,
+        client_headers=client_headers if isinstance(client_headers, dict) else None,
+    )
+
+
 def _resolve_model_provider_for_responses(
     model: str,
     custom_llm_provider: str | None,
@@ -918,11 +930,7 @@ def responses(
         _is_async = kwargs.pop("aresponses", False) is True
         use_chat_completions_api = _pop_use_chat_completions_api_kw(kwargs)
 
-        client_headers = kwargs.get("headers")
-        extra_headers = ResponsesAPIRequestUtils.merge_client_forwarded_headers(
-            extra_headers=extra_headers,
-            client_headers=client_headers if isinstance(client_headers, dict) else None,
-        )
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
         local_vars["extra_headers"] = extra_headers
 
         # Convert text_format to text parameter if provided
@@ -1241,6 +1249,8 @@ def delete_responses(
         litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj")  # type: ignore
         litellm_call_id: str | None = kwargs.get("litellm_call_id", None)
         _is_async = kwargs.pop("adelete_responses", False) is True
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
+        local_vars["extra_headers"] = extra_headers
 
         # get llm provider logic
         litellm_params = GenericLiteLLMParams(**kwargs)
@@ -1412,6 +1422,8 @@ def get_responses(
         litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj")  # type: ignore
         litellm_call_id: str | None = kwargs.get("litellm_call_id", None)
         _is_async = kwargs.pop("aget_responses", False) is True
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
+        local_vars["extra_headers"] = extra_headers
 
         # get llm provider logic
         litellm_params = GenericLiteLLMParams(**kwargs)
@@ -1561,6 +1573,8 @@ def list_input_items(
         litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj")  # type: ignore
         litellm_call_id: str | None = kwargs.get("litellm_call_id", None)
         _is_async = kwargs.pop("alist_input_items", False) is True
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
+        local_vars["extra_headers"] = extra_headers
 
         litellm_params = GenericLiteLLMParams(**kwargs)
 
@@ -1705,6 +1719,8 @@ def cancel_responses(
         litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj")  # type: ignore
         litellm_call_id: str | None = kwargs.get("litellm_call_id", None)
         _is_async = kwargs.pop("acancel_responses", False) is True
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
+        local_vars["extra_headers"] = extra_headers
 
         # get llm provider logic
         litellm_params = GenericLiteLLMParams(**kwargs)
@@ -1877,6 +1893,8 @@ def compact_responses(
         litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj")  # type: ignore
         litellm_call_id: str | None = kwargs.get("litellm_call_id", None)
         _is_async = kwargs.pop("acompact_responses", False) is True
+        extra_headers = _merge_forwarded_client_headers(extra_headers, kwargs)
+        local_vars["extra_headers"] = extra_headers
 
         # get llm provider logic
         litellm_params = GenericLiteLLMParams(**kwargs)

tests/test_litellm/responses/test_responses_api_request_body.py

@@ -77,6 +77,11 @@ def __init__(self, json_data, status_code=200):
     def json(self):
         return self._json_data
 
+    def raise_for_status(self) -> None:
+        if self.status_code >= 400:
+            request = httpx.Request("GET", "http://mock")
+            raise httpx.HTTPStatusError(self.text, request=request, response=httpx.Response(self.status_code))
+
 
 def _assert_request_body_matches(request_body: dict, expected_body: dict) -> None:
     for key, expected_value in expected_body.items():
@@ -367,3 +372,64 @@ async def test_aresponses_client_header_conflict_is_case_insensitive():
 
     assert [name for name in request_headers if name.lower() == "x-shared"] == ["x-shared"]
     assert request_headers["x-shared"] == "from-caller"
+
+
+_MANAGEMENT_ROUTES = (
+    (litellm.aget_responses, "get", _minimal_responses_api_payload("resp_123", "gpt-4o")),
+    (litellm.adelete_responses, "delete", {"id": "resp_123", "object": "response", "deleted": True}),
+    (litellm.acancel_responses, "post", _minimal_responses_api_payload("resp_123", "gpt-4o")),
+    (litellm.alist_input_items, "get", {"object": "list", "data": [], "has_more": False}),
+)
+
+
+@pytest.mark.parametrize("route,http_method,payload", _MANAGEMENT_ROUTES)
+@pytest.mark.asyncio
+async def test_responses_management_routes_forward_client_headers_to_provider(route, http_method, payload):
+    """
+    The response management routes must forward the proxy's `headers` kwarg
+    (`forward_client_headers_to_llm_api`) to the provider, like creating a response does.
+    """
+    with patch(
+        f"litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.{http_method}",
+        new_callable=AsyncMock,
+    ) as mock_request:
+        mock_request.return_value = MockResponse(payload, 200)
+
+        await route(
+            response_id="resp_123",
+            custom_llm_provider="openai",
+            api_key="fake-api-key",
+            headers={"x-my-new-header": "hello-from-client"},
+        )
+
+        mock_request.assert_called_once()
+        request_headers = dict(mock_request.call_args.kwargs["headers"])
+
+    assert request_headers["x-my-new-header"] == "hello-from-client"
+
+
+@pytest.mark.parametrize("route,http_method,payload", _MANAGEMENT_ROUTES)
+@pytest.mark.asyncio
+async def test_responses_management_routes_extra_headers_win_over_client_headers(route, http_method, payload):
+    """
+    Explicit `extra_headers` beat the forwarded client headers on the management routes too,
+    case-insensitively.
+    """
+    with patch(
+        f"litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.{http_method}",
+        new_callable=AsyncMock,
+    ) as mock_request:
+        mock_request.return_value = MockResponse(payload, 200)
+
+        await route(
+            response_id="resp_123",
+            custom_llm_provider="openai",
+            api_key="fake-api-key",
+            headers={"X-Shared": "from-client"},
+            extra_headers={"x-shared": "from-caller"},
+        )
+
+        request_headers = dict(mock_request.call_args.kwargs["headers"])
+
+    assert [name for name in request_headers if name.lower() == "x-shared"] == ["x-shared"]
+    assert request_headers["x-shared"] == "from-caller"