Merge e24a7b9b94 into b4a8bc715f

litellm/llms/custom_httpx/http_handler.py

@@ -214,6 +214,12 @@ def _default_cached_client_timeout() -> httpx.Timeout:
     return httpx.Timeout(timeout=configured, connect=HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS)
 
 
+def _with_bounded_connect(timeout: float | httpx.Timeout | None) -> httpx.Timeout | None:
+    if timeout is None or isinstance(timeout, httpx.Timeout):
+        return timeout
+    return httpx.Timeout(timeout=timeout, connect=min(timeout, HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS))
+
+
 _CLIENT_REFCOUNT_WHEN_HANDLER_IS_SOLE_REFERRER: Final = 2
 
 
@@ -959,7 +965,7 @@ class AsyncHTTPHandler:
                 json=json,
                 params=params,
                 headers=headers,
-                timeout=timeout,
+                timeout=_with_bounded_connect(timeout),
                 files=files,
                 content=request_content,
             )
@@ -970,7 +976,7 @@ class AsyncHTTPHandler:
             return response
         except httpx.ConnectError:
             # Retry the request with a new session if there is a connection error
-            new_client: Final = self.create_client(timeout=timeout, event_hooks=self.event_hooks)
+            new_client: Final = self.create_client(timeout=_with_bounded_connect(timeout), event_hooks=self.event_hooks)
             try:
                 return await self.single_connection_request(
                     url=url,

tests/unit/a2a_protocol/providers/pydantic_ai_agents/test_pydantic_ai_agent_headers.py

@@ -14,6 +14,7 @@ from litellm.a2a_protocol.providers.pydantic_ai_agents.config import PydanticAIP
 from litellm.a2a_protocol.providers.pydantic_ai_agents.transformation import (
     PydanticAITransformation,
 )
+from litellm.constants import HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS
 
 
 def _build_mock_client(response_payload):
@@ -240,7 +241,12 @@ async def test_provider_config_non_streaming_defaults_to_sixty_second_timeout(ag
         "id": "req-5",
         "result": {"kind": "message", "role": "agent", "parts": [{"kind": "text", "text": "ok"}], "messageId": ANY},
     }
-    assert request.extensions["timeout"] == {"connect": 60.0, "read": 60.0, "write": 60.0, "pool": 60.0}
+    assert request.extensions["timeout"] == {
+        "connect": min(60.0, HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS),
+        "read": 60.0,
+        "write": 60.0,
+        "pool": 60.0,
+    }
     assert request.headers["content-type"] == "application/json"
 
 
@@ -259,7 +265,12 @@ async def test_provider_config_non_streaming_forwards_timeout_and_ignores_unrela
 
     request: Final = agent_route.calls.last.request
     assert response["id"] == "req-5"
-    assert request.extensions["timeout"] == {"connect": 5.5, "read": 5.5, "write": 5.5, "pool": 5.5}
+    assert request.extensions["timeout"] == {
+        "connect": min(5.5, HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS),
+        "read": 5.5,
+        "write": 5.5,
+        "pool": 5.5,
+    }
     assert request.headers["x-trace-id"] == "abc-123"
 
 

tests/unit/llms/custom_httpx/test_http_handler.py

@@ -25,6 +25,7 @@ from yarl import URL
 
 import litellm
 from litellm.caching.llm_caching_handler import LLMClientCache
+from litellm.constants import HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS
 from litellm.llms.custom_httpx.aiohttp_transport import LiteLLMAiohttpTransport
 from litellm.llms.custom_httpx.http_handler import (
     _CLIENT_REFCOUNT_WHEN_HANDLER_IS_SOLE_REFERRER,
@@ -141,6 +142,52 @@ def test_sync_post_streaming_status_error_should_not_wait_forever_for_body(
         litellm_handler.close()
 
 
+def _echo_timeout(request: httpx.Request) -> httpx.Response:
+    return httpx.Response(200, json=request.extensions["timeout"])
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("timeout", "expected"),
+    [
+        (600.0, {"connect": HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS, "read": 600.0, "write": 600.0, "pool": 600.0}),
+        (2.0, {"connect": 2.0, "read": 2.0, "write": 2.0, "pool": 2.0}),
+        (httpx.Timeout(600.0), {"connect": 600.0, "read": 600.0, "write": 600.0, "pool": 600.0}),
+    ],
+)
+async def test_async_post_numeric_timeout_keeps_bounded_connect_timeout(
+    timeout: float | httpx.Timeout, expected: Mapping[str, float]
+):
+    handler: Final = AsyncHTTPHandler(transport=httpx.MockTransport(_echo_timeout))
+    try:
+        response: Final = await handler.post("http://upstream.invalid/v1/chat/completions", json={}, timeout=timeout)
+    finally:
+        await handler.close()
+
+    assert response.json() == expected
+
+
+@pytest.mark.asyncio
+async def test_async_post_replay_after_connect_error_keeps_bounded_connect_timeout():
+    def refuse(request: httpx.Request) -> httpx.Response:
+        raise httpx.ConnectError("connection refused", request=request)
+
+    handler: Final = AsyncHTTPHandler(transport=httpx.MockTransport(_echo_timeout))
+    await handler.client.aclose()
+    handler.client = httpx.AsyncClient(transport=httpx.MockTransport(refuse))
+    try:
+        response: Final = await handler.post("http://upstream.invalid/v1/chat/completions", json={}, timeout=600.0)
+    finally:
+        await handler.client.aclose()
+
+    assert response.json() == {
+        "connect": HTTP_HANDLER_CONNECT_TIMEOUT_SECONDS,
+        "read": 600.0,
+        "write": 600.0,
+        "pool": 600.0,
+    }
+
+
 @pytest.mark.asyncio
 async def test_ssl_security_level(monkeypatch):
     # Ensure aiohttp transport is enabled for this test