From 100cfc11acf73e6c8bd92f7f856568f7623f0d4f Mon Sep 17 00:00:00 2001
From: Ishaan Jaff <ishaanjaffer0324@gmail.com>
Date: Wed, 3 Dec 2025 14:13:28 -0800
Subject: [PATCH] [Bug Fix] Parallel Request Limiter with /messages  (#17426)

* fix: use standard_logging_object for parallel request limiter

* fix test parallel request limtier
---
 .../hooks/parallel_request_limiter_v3.py      | 34 ++++++++++---------
 .../hooks/test_parallel_request_limiter_v3.py | 10 +++---
 2 files changed, 24 insertions(+), 20 deletions(-)

diff --git a/litellm/proxy/hooks/parallel_request_limiter_v3.py b/litellm/proxy/hooks/parallel_request_limiter_v3.py
index 9436df585a..2abba1d497 100644
--- a/litellm/proxy/hooks/parallel_request_limiter_v3.py
+++ b/litellm/proxy/hooks/parallel_request_limiter_v3.py
@@ -1343,19 +1343,19 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
                 "INSIDE parallel request limiter ASYNC SUCCESS LOGGING"
             )
 
-            # Get metadata from kwargs
-            litellm_metadata = kwargs["litellm_params"].get(
-                get_metadata_variable_name_from_kwargs(kwargs), {}
+            # Get metadata from standard_logging_object - this correctly handles both
+            # 'metadata' and 'litellm_metadata' fields from litellm_params
+            standard_logging_object = kwargs.get("standard_logging_object") or {}
+            standard_logging_metadata = standard_logging_object.get("metadata") or {}
+
+            # user_api_key_hash is the same as user_api_key (it's the hash)
+            user_api_key = standard_logging_metadata.get("user_api_key_hash")
+            user_api_key_user_id = standard_logging_metadata.get("user_api_key_user_id")
+            user_api_key_team_id = standard_logging_metadata.get("user_api_key_team_id")
+            user_api_key_organization_id = standard_logging_metadata.get(
+                "user_api_key_org_id"
             )
-            if litellm_metadata is None:
-                return
-            user_api_key = litellm_metadata.get("user_api_key")
-            user_api_key_user_id = litellm_metadata.get("user_api_key_user_id")
-            user_api_key_team_id = litellm_metadata.get("user_api_key_team_id")
-            user_api_key_organization_id = litellm_metadata.get(
-                "user_api_key_organization_id"
-            )
-            user_api_key_end_user_id = kwargs.get("user") or litellm_metadata.get(
+            user_api_key_end_user_id = kwargs.get("user") or standard_logging_metadata.get(
                 "user_api_key_end_user_id"
             )
             model_group = get_model_group_from_litellm_kwargs(kwargs)
@@ -1501,10 +1501,12 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
             litellm_parent_otel_span: Union[
                 Span, None
             ] = _get_parent_otel_span_from_kwargs(kwargs)
-            litellm_metadata = kwargs["litellm_params"]["metadata"]
-            user_api_key = (
-                litellm_metadata.get("user_api_key") if litellm_metadata else None
-            )
+            # Get metadata from standard_logging_object - this correctly handles both
+            # 'metadata' and 'litellm_metadata' fields from litellm_params
+            standard_logging_object = kwargs.get("standard_logging_object") or {}
+            standard_logging_metadata = standard_logging_object.get("metadata") or {}
+            user_api_key = standard_logging_metadata.get("user_api_key_hash")
+
             pipeline_operations: List[RedisPipelineIncrementOperation] = []
 
             if user_api_key:
diff --git a/tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py b/tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py
index 6574c500fc..c8c30d41b5 100644
--- a/tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py
+++ b/tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py
@@ -22,6 +22,7 @@ from litellm.proxy.hooks.parallel_request_limiter_v3 import (
 from litellm.proxy.utils import InternalUsageCache, ProxyLogging, hash_token
 from litellm.types.utils import ModelResponse, Usage
 
+
 class TimeController:
     def __init__(self):
         self._current = datetime.utcnow()
@@ -461,10 +462,11 @@ async def test_token_rate_limit_type_respected_v3(monkeypatch, token_rate_limit_
     )
 
     # Create mock kwargs for the success event
+    # Use standard_logging_object which is the canonical source for metadata
     mock_kwargs = {
-        "litellm_params": {
+        "standard_logging_object": {
             "metadata": {
-                "user_api_key": _api_key,
+                "user_api_key_hash": _api_key,
                 "user_api_key_user_id": None,
                 "user_api_key_team_id": None,
                 "user_api_key_end_user_id": None,
@@ -532,8 +534,8 @@ async def test_async_log_failure_event_v3():
         internal_usage_cache=InternalUsageCache(local_cache)
     )
 
-    # Mock kwargs with user_api_key
-    mock_kwargs = {"litellm_params": {"metadata": {"user_api_key": _api_key}}}
+    # Mock kwargs with user_api_key via standard_logging_object
+    mock_kwargs = {"standard_logging_object": {"metadata": {"user_api_key_hash": _api_key}}}
 
     # Capture pipeline operations
     captured_ops = []