feat: add timeout to passthrough route

litellm/passthrough/main.py

@@ -25,7 +25,7 @@ from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
 from litellm.llms.custom_httpx.llm_http_handler import BaseLLMHTTPHandler
 from litellm.passthrough.utils import CommonUtils
-from litellm.utils import client
+from litellm.utils import client, supports_httpx_timeout
 
 base_llm_http_handler = BaseLLMHTTPHandler()
 from .utils import BasePassthroughUtils
@@ -52,6 +52,7 @@ async def allm_passthrough_route(
     json: Optional[Any] = None,
     params: Optional[QueryParamTypes] = None,
     cookies: Optional[CookieTypes] = None,
+    timeout: Optional[Union[float, int, httpx.Timeout]] = None,
     client: Optional[Union[HTTPHandler, AsyncHTTPHandler]] = None,
     **kwargs,
 ) -> Union[httpx.Response, AsyncGenerator[Any, Any]]:
@@ -98,6 +99,7 @@ async def allm_passthrough_route(
             json=json,
             params=params,
             cookies=cookies,
+            timeout=timeout,
             client=client,
             **kwargs,
         )
@@ -178,6 +180,7 @@ def llm_passthrough_route(
     json: Optional[Any] = None,
     params: Optional[QueryParamTypes] = None,
     cookies: Optional[CookieTypes] = None,
+    timeout: Optional[Union[float, int, httpx.Timeout]] = None,
     client: Optional[Union[HTTPHandler, AsyncHTTPHandler]] = None,
     **kwargs,
 ) -> Union[
@@ -286,6 +289,17 @@ def llm_passthrough_route(
     if json and isinstance(json, dict) and "model" in json:
         json["model"] = model
 
+    ### TIMEOUT LOGIC ###
+    timeout = timeout or kwargs.get("request_timeout", 600) or 600
+    if (
+        custom_llm_provider is not None
+        and isinstance(timeout, httpx.Timeout)
+        and not supports_httpx_timeout(custom_llm_provider)
+    ):
+        timeout = timeout.read or 600
+    elif not isinstance(timeout, httpx.Timeout):
+        timeout = float(timeout)
+
     request = client.client.build_request(
         method=method,
         url=updated_url,
@@ -296,6 +310,7 @@ def llm_passthrough_route(
         params=params,
         headers=headers,
         cookies=cookies,
+        timeout=timeout,
     )
 
     ## IS STREAMING REQUEST

tests/test_litellm/passthrough/test_passthrough_main.py

@@ -510,6 +510,159 @@ def test_azure_with_custom_api_base_and_key():
         assert response.status_code == 200  # type: ignore[union-attr]
 
 
+def test_timeout_param_forwarded_to_build_request():
+    """
+    Verify that the `timeout` parameter is resolved and forwarded to
+    build_request so httpx enforces it on the underlying HTTP call.
+    The @client decorator reads kwargs["timeout"] on exception and attaches
+    it to the exception for the router's fallback logic.
+    """
+    client = HTTPHandler()
+
+    mock_provider_config = MagicMock()
+    mock_provider_config.get_complete_url.return_value = (
+        httpx.URL("https://api.example.com/v1/chat/completions"),
+        "https://api.example.com",
+    )
+    mock_provider_config.get_api_key.return_value = "test-key"
+    mock_provider_config.validate_environment.return_value = {}
+    mock_provider_config.sign_request.return_value = ({}, None)
+    mock_provider_config.is_streaming_request.return_value = False
+
+    with patch(
+        "litellm.utils.ProviderConfigManager.get_provider_passthrough_config",
+        return_value=mock_provider_config,
+    ), patch(
+        "litellm.litellm_core_utils.get_litellm_params.get_litellm_params",
+        return_value={},
+    ), patch(
+        "litellm.litellm_core_utils.get_llm_provider_logic.get_llm_provider",
+        return_value=("my-model", "openai", "test-key", "https://api.example.com"),
+    ), patch.object(
+        client.client, "send", return_value=MagicMock(status_code=200)
+    ), patch.object(
+        client.client, "build_request"
+    ) as mock_build_request:
+        mock_logging_obj = MagicMock()
+        mock_logging_obj.update_environment_variables = MagicMock()
+
+        llm_passthrough_route(
+            model="openai/my-model",
+            endpoint="v1/chat/completions",
+            method="POST",
+            custom_llm_provider="openai",
+            json={"model": "my-model", "messages": [{"role": "user", "content": "Hi"}]},
+            timeout=30,
+            client=client,
+            litellm_logging_obj=mock_logging_obj,
+        )
+
+        mock_build_request.assert_called_once()
+        call_kwargs = mock_build_request.call_args.kwargs
+        assert call_kwargs["timeout"] == 30.0
+
+
+def test_timeout_defaults_to_600_when_not_provided():
+    """
+    When no timeout is passed, the passthrough should default to 600s
+    (matching the completion endpoint behaviour).
+    """
+    client = HTTPHandler()
+
+    mock_provider_config = MagicMock()
+    mock_provider_config.get_complete_url.return_value = (
+        httpx.URL("https://api.example.com/v1/chat/completions"),
+        "https://api.example.com",
+    )
+    mock_provider_config.get_api_key.return_value = "test-key"
+    mock_provider_config.validate_environment.return_value = {}
+    mock_provider_config.sign_request.return_value = ({}, None)
+    mock_provider_config.is_streaming_request.return_value = False
+
+    with patch(
+        "litellm.utils.ProviderConfigManager.get_provider_passthrough_config",
+        return_value=mock_provider_config,
+    ), patch(
+        "litellm.litellm_core_utils.get_litellm_params.get_litellm_params",
+        return_value={},
+    ), patch(
+        "litellm.litellm_core_utils.get_llm_provider_logic.get_llm_provider",
+        return_value=("my-model", "openai", "test-key", "https://api.example.com"),
+    ), patch.object(
+        client.client, "send", return_value=MagicMock(status_code=200)
+    ), patch.object(
+        client.client, "build_request"
+    ) as mock_build_request:
+        mock_logging_obj = MagicMock()
+        mock_logging_obj.update_environment_variables = MagicMock()
+
+        llm_passthrough_route(
+            model="openai/my-model",
+            endpoint="v1/chat/completions",
+            method="POST",
+            custom_llm_provider="openai",
+            json={"model": "my-model", "messages": [{"role": "user", "content": "Hi"}]},
+            client=client,
+            litellm_logging_obj=mock_logging_obj,
+        )
+
+        mock_build_request.assert_called_once()
+        call_kwargs = mock_build_request.call_args.kwargs
+        assert call_kwargs["timeout"] == 600.0
+
+
+def test_timeout_httpx_timeout_object_forwarded():
+    """
+    When an httpx.Timeout object is passed, it should be forwarded
+    directly to build_request without conversion.
+    """
+    client = HTTPHandler()
+
+    mock_provider_config = MagicMock()
+    mock_provider_config.get_complete_url.return_value = (
+        httpx.URL("https://api.example.com/v1/chat/completions"),
+        "https://api.example.com",
+    )
+    mock_provider_config.get_api_key.return_value = "test-key"
+    mock_provider_config.validate_environment.return_value = {}
+    mock_provider_config.sign_request.return_value = ({}, None)
+    mock_provider_config.is_streaming_request.return_value = False
+
+    custom_timeout = httpx.Timeout(timeout=45.0, connect=10.0)
+
+    with patch(
+        "litellm.utils.ProviderConfigManager.get_provider_passthrough_config",
+        return_value=mock_provider_config,
+    ), patch(
+        "litellm.litellm_core_utils.get_litellm_params.get_litellm_params",
+        return_value={},
+    ), patch(
+        "litellm.litellm_core_utils.get_llm_provider_logic.get_llm_provider",
+        return_value=("my-model", "openai", "test-key", "https://api.example.com"),
+    ), patch.object(
+        client.client, "send", return_value=MagicMock(status_code=200)
+    ), patch.object(
+        client.client, "build_request"
+    ) as mock_build_request:
+        mock_logging_obj = MagicMock()
+        mock_logging_obj.update_environment_variables = MagicMock()
+
+        llm_passthrough_route(
+            model="openai/my-model",
+            endpoint="v1/chat/completions",
+            method="POST",
+            custom_llm_provider="openai",
+            json={"model": "my-model", "messages": [{"role": "user", "content": "Hi"}]},
+            timeout=custom_timeout,
+            client=client,
+            litellm_logging_obj=mock_logging_obj,
+        )
+
+        mock_build_request.assert_called_once()
+        call_kwargs = mock_build_request.call_args.kwargs
+        assert call_kwargs["timeout"] is custom_timeout
+
+
 def test_content_param_forwarded_to_build_request():
     """
     Regression test: the `content` parameter passed to llm_passthrough_route