openai[patch]: fix client caching when request_timeout is specified via httpx.Timeout (#31698)

Resolves #31697

Author: ccurme

libs/partners/openai/langchain_openai/chat_models/_client_utils.py

@@ -41,8 +41,7 @@ def __del__(self) -> None:
             pass
 
 
-@lru_cache
-def _get_default_httpx_client(
+def _build_sync_httpx_client(
     base_url: Optional[str], timeout: Any
 ) -> _SyncHttpxClientWrapper:
     return _SyncHttpxClientWrapper(
@@ -53,8 +52,7 @@ def _get_default_httpx_client(
     )
 
 
-@lru_cache
-def _get_default_async_httpx_client(
+def _build_async_httpx_client(
     base_url: Optional[str], timeout: Any
 ) -> _AsyncHttpxClientWrapper:
     return _AsyncHttpxClientWrapper(
@@ -63,3 +61,47 @@ def _get_default_async_httpx_client(
         or "https://api.openai.com/v1",
         timeout=timeout,
     )
+
+
+@lru_cache
+def _cached_sync_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _SyncHttpxClientWrapper:
+    return _build_sync_httpx_client(base_url, timeout)
+
+
+@lru_cache
+def _cached_async_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _AsyncHttpxClientWrapper:
+    return _build_async_httpx_client(base_url, timeout)
+
+
+def _get_default_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _SyncHttpxClientWrapper:
+    """Get default httpx client.
+
+    Uses cached client unless timeout is ``httpx.Timeout``, which is not hashable.
+    """
+    try:
+        hash(timeout)
+    except TypeError:
+        return _build_sync_httpx_client(base_url, timeout)
+    else:
+        return _cached_sync_httpx_client(base_url, timeout)
+
+
+def _get_default_async_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _AsyncHttpxClientWrapper:
+    """Get default httpx client.
+
+    Uses cached client unless timeout is ``httpx.Timeout``, which is not hashable.
+    """
+    try:
+        hash(timeout)
+    except TypeError:
+        return _build_async_httpx_client(base_url, timeout)
+    else:
+        return _cached_async_httpx_client(base_url, timeout)

libs/partners/openai/tests/unit_tests/chat_models/test_base.py

@@ -6,6 +6,7 @@
 from typing import Any, Literal, Optional, Union, cast
 from unittest.mock import AsyncMock, MagicMock, patch
 
+import httpx
 import pytest
 from langchain_core.load import dumps, loads
 from langchain_core.messages import (
@@ -74,6 +75,30 @@ def test_openai_model_param() -> None:
     assert llm.max_tokens == 10
 
 
+def test_openai_client_caching() -> None:
+    """Test that the OpenAI client is cached."""
+    llm1 = ChatOpenAI(model="gpt-4.1-mini")
+    llm2 = ChatOpenAI(model="gpt-4.1-mini")
+    assert llm1.root_client._client is llm2.root_client._client
+
+    llm3 = ChatOpenAI(model="gpt-4.1-mini", base_url="foo")
+    assert llm1.root_client._client is not llm3.root_client._client
+
+    llm4 = ChatOpenAI(model="gpt-4.1-mini", timeout=None)
+    assert llm1.root_client._client is llm4.root_client._client
+
+    llm5 = ChatOpenAI(model="gpt-4.1-mini", timeout=3)
+    assert llm1.root_client._client is not llm5.root_client._client
+
+    llm6 = ChatOpenAI(
+        model="gpt-4.1-mini", timeout=httpx.Timeout(timeout=60.0, connect=5.0)
+    )
+    assert llm1.root_client._client is not llm6.root_client._client
+
+    llm7 = ChatOpenAI(model="gpt-4.1-mini", timeout=(5, 1))
+    assert llm1.root_client._client is not llm7.root_client._client
+
+
 def test_openai_o1_temperature() -> None:
     llm = ChatOpenAI(model="o1-preview")
     assert llm.temperature == 1