fix(http): honor request timeout for parameterized sync clients

Signed-off-by: amarrtech <272048731+amarrtech@users.noreply.github.com>

litellm/llms/custom_httpx/http_handler.py

@@ -1796,7 +1796,11 @@ def _get_httpx_client(params: dict | None = None) -> HTTPHandler:
 
     if params is not None:
         # Filter out params that are only used for cache key, not for HTTPHandler.__init__
-        handler_params: Final = {k: v for k, v in params.items() if k != "disable_aiohttp_transport"}
+        configured_timeout: Final = params.get("timeout")
+        handler_params: Final = {
+            **{k: v for k, v in params.items() if k not in ("disable_aiohttp_transport", "timeout")},
+            "timeout": configured_timeout if configured_timeout is not None else _default_cached_client_timeout(),
+        }
         _new_client = HTTPHandler(**handler_params)
     else:
         _new_client = HTTPHandler(timeout=_default_cached_client_timeout())

tests/unit/llms/custom_httpx/test_http_handler.py

@@ -741,6 +741,18 @@ class TestDefaultCachedClientTimeoutHonorsRequestTimeout:
         client = get_async_httpx_client(llm_provider=LlmProviders.BEDROCK)
         assert client.timeout.read == 300.0
 
+    def test_cached_sync_client_with_none_timeout_uses_explicit_request_timeout(self, monkeypatch: pytest.MonkeyPatch):
+        from litellm.caching.llm_caching_handler import LLMClientCache
+
+        monkeypatch.setattr(litellm, "request_timeout", 300)
+        monkeypatch.setattr(litellm, "request_timeout_explicitly_set", True)
+        litellm.in_memory_llm_clients_cache = LLMClientCache()
+
+        client = _get_httpx_client(params={"timeout": None})
+        request = client.client.build_request("GET", "https://example.com")
+
+        assert request.extensions["timeout"]["read"] == 300.0
+
 
 async def _read_http_request(reader: asyncio.StreamReader) -> None:
     raw = b""