openai: cache httpx client (#31260)

![Screenshot 2025-05-16 at 3 49
54 PM](https://github.com/user-attachments/assets/4b377384-a769-4487-b801-bd1aa0ed66c1)

Co-authored-by: Sydney Runkle <[email protected]>

Author: ccurme

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

@@ -0,0 +1,65 @@
+"""Helpers for creating OpenAI API clients.
+
+This module allows for the caching of httpx clients to avoid creating new instances
+for each instance of ChatOpenAI.
+
+Logic is largely replicated from openai._base_client.
+"""
+
+import asyncio
+import os
+from functools import lru_cache
+from typing import Any, Optional
+
+import openai
+
+
+class _SyncHttpxClientWrapper(openai.DefaultHttpxClient):
+    """Borrowed from openai._base_client"""
+
+    def __del__(self) -> None:
+        if self.is_closed:
+            return
+
+        try:
+            self.close()
+        except Exception:
+            pass
+
+
+class _AsyncHttpxClientWrapper(openai.DefaultAsyncHttpxClient):
+    """Borrowed from openai._base_client"""
+
+    def __del__(self) -> None:
+        if self.is_closed:
+            return
+
+        try:
+            # TODO(someday): support non asyncio runtimes here
+            asyncio.get_running_loop().create_task(self.aclose())
+        except Exception:
+            pass
+
+
+@lru_cache
+def _get_default_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _SyncHttpxClientWrapper:
+    return _SyncHttpxClientWrapper(
+        base_url=base_url
+        or os.environ.get("OPENAI_BASE_URL")
+        or "https://api.openai.com/v1",
+        timeout=timeout,
+    )
+
+
+@lru_cache
+def _get_default_async_httpx_client(
+    base_url: Optional[str], timeout: Any
+) -> _AsyncHttpxClientWrapper:
+    return _AsyncHttpxClientWrapper(
+        base_url=base_url
+        or os.environ.get("OPENAI_BASE_URL")
+        or "https://api.openai.com/v1",
+        timeout=timeout,
+    )

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

@@ -102,6 +102,11 @@
 from pydantic.v1 import BaseModel as BaseModelV1
 from typing_extensions import Self
 
+from langchain_openai.chat_models._client_utils import (
+    _get_default_async_httpx_client,
+    _get_default_httpx_client,
+)
+
 if TYPE_CHECKING:
     from openai.types.responses import Response
 
@@ -621,7 +626,10 @@ def validate_environment(self) -> Self:
                 self.http_client = httpx.Client(
                     proxy=self.openai_proxy, verify=global_ssl_context
                 )
-            sync_specific = {"http_client": self.http_client}
+            sync_specific = {
+                "http_client": self.http_client
+                or _get_default_httpx_client(self.openai_api_base, self.request_timeout)
+            }
             self.root_client = openai.OpenAI(**client_params, **sync_specific)  # type: ignore[arg-type]
             self.client = self.root_client.chat.completions
         if not self.async_client:
@@ -636,7 +644,12 @@ def validate_environment(self) -> Self:
                 self.http_async_client = httpx.AsyncClient(
                     proxy=self.openai_proxy, verify=global_ssl_context
                 )
-            async_specific = {"http_client": self.http_async_client}
+            async_specific = {
+                "http_client": self.http_async_client
+                or _get_default_async_httpx_client(
+                    self.openai_api_base, self.request_timeout
+                )
+            }
             self.root_async_client = openai.AsyncOpenAI(
                 **client_params,
                 **async_specific,  # type: ignore[arg-type]