[Fix proxy perf] Use correct cache key when reading from redis cache (#5928)

* fix parallel request limiter use correct user id

* async def get_user_object(
fix

* use safe get_internal_user_object

* fix store internal users in redis correctly

litellm/proxy/hooks/parallel_request_limiter.py

@@ -322,9 +322,9 @@ class _PROXY_MaxParallelRequestsHandler(CustomLogger):
         # check if REQUEST ALLOWED for user_id
         user_id = user_api_key_dict.user_id
         if user_id is not None:
-            _user_id_rate_limits = await self.internal_usage_cache.async_get_cache(
-                key=user_id,
-                litellm_parent_otel_span=user_api_key_dict.parent_otel_span,
+            _user_id_rate_limits = await self.get_internal_user_object(
+                user_id=user_id,
+                user_api_key_dict=user_api_key_dict,
             )
             # get user tpm/rpm limits
             if _user_id_rate_limits is not None and isinstance(
@@ -741,3 +741,39 @@ class _PROXY_MaxParallelRequestsHandler(CustomLogger):
                     str(e)
                 )
             )
+
+    async def get_internal_user_object(
+        self,
+        user_id: str,
+        user_api_key_dict: UserAPIKeyAuth,
+    ) -> Optional[dict]:
+        """
+        Helper to get the 'Internal User Object'
+
+        It uses the `get_user_object` function from `litellm.proxy.auth.auth_checks`
+
+        We need this because the UserApiKeyAuth object does not contain the rpm/tpm limits for a User AND there could be a perf impact by additionally reading the UserTable.
+        """
+        from litellm._logging import verbose_proxy_logger
+        from litellm.proxy.auth.auth_checks import get_user_object
+        from litellm.proxy.proxy_server import prisma_client
+
+        try:
+            _user_id_rate_limits = await get_user_object(
+                user_id=user_id,
+                prisma_client=prisma_client,
+                user_api_key_cache=self.internal_usage_cache.dual_cache,
+                user_id_upsert=False,
+                parent_otel_span=user_api_key_dict.parent_otel_span,
+                proxy_logging_obj=None,
+            )
+
+            if _user_id_rate_limits is None:
+                return None
+
+            return _user_id_rate_limits.model_dump()
+        except Exception as e:
+            verbose_proxy_logger.exception(
+                "Parallel Request Limiter: Error getting user object", str(e)
+            )
+            return None