fix(alerting): read key/team alias from top-level metadata for hanging-request alert

At proxy pre-call time litellm_params is not yet populated, so
get_litellm_metadata_from_kwargs returns {} and the hanging-request Slack
alert always showed empty Key Alias / Team Alias. Fall back to the top-level
request metadata where the alias actually lives at that point, and add a
regression test.

Fixes #34708

litellm/integrations/SlackAlerting/hanging_request_check.py

@@ -58,6 +58,17 @@ class AlertingHangingRequestCheck:
             return
 
         request_metadata = get_litellm_metadata_from_kwargs(kwargs=request_data)
+        if not request_metadata:
+            # At proxy pre-call time, `litellm_params` is not populated yet, so
+            # `get_litellm_metadata_from_kwargs` returns {}. The key/team alias is
+            # present at the top level of the request at this point, so fall back
+            # to it - otherwise the hanging-request alert cannot identify which
+            # key/team is responsible (Key Alias / Team Alias show up empty).
+            request_metadata = (
+                request_data.get("metadata")
+                or request_data.get("litellm_metadata")
+                or {}
+            )
         model = request_data.get("model", "")
         api_base: Optional[str] = None
 

tests/test_litellm/integrations/SlackAlerting/test_hanging_request_check.py

@@ -123,6 +123,41 @@ class TestAlertingHangingRequestCheck:
         assert cached_data.key_alias == ""
         assert cached_data.team_alias == ""
 
+    @pytest.mark.asyncio
+    async def test_add_request_populates_alias_from_top_level_metadata(
+        self, hanging_request_checker
+    ):
+        """
+        At proxy pre-call time, ``litellm_params`` is not yet populated and the
+        key/team alias lives at the top level of the request. The hanging-request
+        entry must still capture the alias so the alert can attribute the hang.
+
+        Regression test for empty Key Alias / Team Alias (#34708).
+        """
+        request_data = {
+            "litellm_call_id": "precall_request_001",
+            "model": "gpt-4",
+            # no top-level "litellm_params" at pre-call time
+            "metadata": {
+                "user_api_key_alias": "precall_key",
+                "user_api_key_team_alias": "precall_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="precall_request_001"
+            )
+        )
+
+        assert cached_data is not None
+        assert cached_data.key_alias == "precall_key"
+        assert cached_data.team_alias == "precall_team"
+
     @pytest.mark.asyncio
     async def test_send_hanging_request_alert(self, hanging_request_checker):
         """