From 8cad7cb80ba03e44e90e73d0dc9f28d619466cf0 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Sun, 26 Jul 2026 13:10:14 +0000
Subject: [PATCH] fix(alerting): read hanging-request aliases from pre-call
 metadata location

---
 .../SlackAlerting/hanging_request_check.py    |  4 +-
 litellm/litellm_core_utils/core_helpers.py    | 21 +++++++
 .../test_hanging_request_check.py             | 61 +++++++++++++++++++
 3 files changed, 84 insertions(+), 2 deletions(-)

diff --git a/litellm/integrations/SlackAlerting/hanging_request_check.py b/litellm/integrations/SlackAlerting/hanging_request_check.py
index 136b6583f38..620738a1d67 100644
--- a/litellm/integrations/SlackAlerting/hanging_request_check.py
+++ b/litellm/integrations/SlackAlerting/hanging_request_check.py
@@ -14,7 +14,7 @@ from typing import TYPE_CHECKING, Any, Optional
 import litellm
 from litellm._logging import verbose_proxy_logger
 from litellm.caching.in_memory_cache import InMemoryCache
-from litellm.litellm_core_utils.core_helpers import get_litellm_metadata_from_kwargs
+from litellm.litellm_core_utils.core_helpers import get_request_metadata
 from litellm.types.integrations.slack_alerting import (
     HANGING_ALERT_BUFFER_TIME_SECONDS,
     MAX_OLDEST_HANGING_REQUESTS_TO_CHECK,
@@ -57,7 +57,7 @@ class AlertingHangingRequestCheck:
         if request_data is None:
             return
 
-        request_metadata = get_litellm_metadata_from_kwargs(kwargs=request_data)
+        request_metadata = get_request_metadata(request_data=request_data)
         model = request_data.get("model", "")
         api_base: Optional[str] = None
 
diff --git a/litellm/litellm_core_utils/core_helpers.py b/litellm/litellm_core_utils/core_helpers.py
index cecc35ee1c1..bd129ab6921 100644
--- a/litellm/litellm_core_utils/core_helpers.py
+++ b/litellm/litellm_core_utils/core_helpers.py
@@ -234,6 +234,27 @@ def get_litellm_metadata_from_kwargs(kwargs: dict):
     return {}
 
 
+def get_request_metadata(request_data: dict) -> dict:
+    """
+    Resolve proxy-internal metadata from request data at any point in the request lifecycle.
+
+    Proxy pre-call hooks see metadata at the top level of the request body; `litellm_params`
+    only exists once the SDK completion call has built it, after which the same metadata is
+    nested under it. Readers that can run on either side of that boundary must check both.
+    """
+    nested_metadata = get_litellm_metadata_from_kwargs(kwargs=request_data)
+    if nested_metadata:
+        return nested_metadata
+
+    litellm_metadata = request_data.get("litellm_metadata")
+    metadata = request_data.get("metadata")
+    if isinstance(litellm_metadata, dict) and litellm_metadata:
+        if isinstance(metadata, dict) and metadata:
+            return add_missing_spend_metadata_to_litellm_metadata({**litellm_metadata}, metadata)
+        return litellm_metadata
+    return metadata if isinstance(metadata, dict) else {}
+
+
 def reconstruct_model_name(
     model_name: str,
     custom_llm_provider: Optional[str],
diff --git a/tests/test_litellm/integrations/SlackAlerting/test_hanging_request_check.py b/tests/test_litellm/integrations/SlackAlerting/test_hanging_request_check.py
index 063aabd309b..c0595b9189c 100644
--- a/tests/test_litellm/integrations/SlackAlerting/test_hanging_request_check.py
+++ b/tests/test_litellm/integrations/SlackAlerting/test_hanging_request_check.py
@@ -81,6 +81,67 @@ class TestAlertingHangingRequestCheck:
         assert cached_data.request_id == "test_request_123"
         assert cached_data.model == "gpt-4"
         assert cached_data.api_base == "https://api.openai.com/v1"
+        assert cached_data.key_alias == "test_key"
+        assert cached_data.team_alias == "test_team"
+
+    @pytest.mark.asyncio
+    async def test_add_request_reads_aliases_from_top_level_litellm_metadata(
+        self, hanging_request_checker
+    ):
+        """
+        Batch/file routes keep proxy metadata in top-level `litellm_metadata`, which is
+        also the only place it lives at pre-call time.
+        """
+        request_data = {
+            "litellm_call_id": "litellm_metadata_request",
+            "model": "gpt-4",
+            "litellm_metadata": {
+                "user_api_key_alias": "batch_key",
+                "user_api_key_team_alias": "batch_team",
+            },
+        }
+
+        await hanging_request_checker.add_request_to_hanging_request_check(request_data)
+
+        cached_data = (
+            await hanging_request_checker.hanging_request_cache.async_get_cache(
+                key="litellm_metadata_request"
+            )
+        )
+
+        assert cached_data.key_alias == "batch_key"
+        assert cached_data.team_alias == "batch_team"
+
+    @pytest.mark.asyncio
+    async def test_add_request_prefers_litellm_params_metadata_when_present(
+        self, hanging_request_checker
+    ):
+        """
+        Post-call kwargs carry metadata under `litellm_params`; that copy wins over a
+        stale top-level one.
+        """
+        request_data = {
+            "litellm_call_id": "litellm_params_request",
+            "model": "gpt-4",
+            "metadata": {"user_api_key_alias": "stale_key"},
+            "litellm_params": {
+                "metadata": {
+                    "user_api_key_alias": "nested_key",
+                    "user_api_key_team_alias": "nested_team",
+                }
+            },
+        }
+
+        await hanging_request_checker.add_request_to_hanging_request_check(request_data)
+
+        cached_data = (
+            await hanging_request_checker.hanging_request_cache.async_get_cache(
+                key="litellm_params_request"
+            )
+        )
+
+        assert cached_data.key_alias == "nested_key"
+        assert cached_data.team_alias == "nested_team"
 
     @pytest.mark.asyncio
     async def test_add_request_to_hanging_request_check_none_request_data(