fix(rate-limiting): read success-event tags from the resolved metadata bucket, not raw kwargs

Bugbot finding: async_log_success_event passed raw kwargs (Logging.model_call_details)
to _get_tags_from_request_kwargs, which checks a top-level key matching
metadata_variable_name before falling back to kwargs["litellm_params"]. Some
call paths populate that top-level key present but None, so it won it over
the real, populated dict nested under litellm_params, silently dropping
token/dollar accounting for those routes. key_hash/key_alias/order_tags_for_identity_resolution
already read from litellm_params_for_metadata; tag extraction now does too.
Regression test confirmed to fail on the pre-fix code.

litellm/proxy/hooks/global_tag_rate_limits_hook.py

@@ -725,7 +725,7 @@ class _PROXY_GlobalTagRateLimitsHook(  # pyright: ignore[reportUnusedClass]  # o
         key_alias: Final = _extract_key_alias(litellm_params_for_metadata, metadata_variable_name)
 
         tags: Final = _order_tags_for_identity_resolution(
-            _get_tags_from_request_kwargs(kwargs, metadata_variable_name=metadata_variable_name),
+            _get_tags_from_request_kwargs(litellm_params_for_metadata, metadata_variable_name=metadata_variable_name),
             litellm_params_for_metadata,
             metadata_variable_name,
         )

tests/test_litellm/proxy/hooks/test_global_tag_rate_limits_hook.py

@@ -1181,6 +1181,50 @@ async def test_log_success_event_accounts_when_litellm_params_carries_a_null_lit
         )
 
 
+@pytest.mark.asyncio
+async def test_log_success_event_reads_tags_when_top_level_kwargs_carries_a_null_metadata_key(
+    time_controller, monkeypatch
+):
+    """
+    Bugbot finding: some call paths populate a top-level "metadata" (or
+    "litellm_metadata") key on kwargs (Logging.model_call_details) set to
+    None, alongside the real, populated dict nested under
+    kwargs["litellm_params"]. _get_tags_from_request_kwargs checks the
+    top-level key first; passing it raw kwargs made a present-but-None
+    top-level key win, reading no tags at all even though
+    metadata_variable_name correctly resolved to "metadata".
+    """
+    monkeypatch.setattr(
+        litellm,
+        "global_tag_rate_limits",
+        {
+            "dollar_limits": {
+                "limits": [{"name": "daily_spend", "tag_id": "end_user_id", "limit": 10.0, "period_seconds": 86400}]
+            }
+        },
+    )
+    hook = _make_hook(time_controller)
+
+    kwargs = {
+        "litellm_call_id": "call-1",
+        "metadata": None,
+        "litellm_params": {
+            "metadata": {"tags": ["end_user_id:u1"], "user_api_key": "hash"},
+        },
+        "standard_logging_object": {"total_tokens": 0, "response_cost": 12.0},
+    }
+    await hook.async_log_success_event(kwargs=kwargs, response_obj=None, start_time=0, end_time=0)
+    await asyncio.sleep(0)
+
+    with pytest.raises(ProxyRateLimitError):
+        await hook.async_pre_call_hook(
+            user_api_key_dict=_key(),
+            cache=DualCache(),
+            data=_data(["end_user_id:u1"], call_id="call-2"),
+            call_type="completion",
+        )
+
+
 @pytest.mark.asyncio
 async def test_dollar_limit_respects_apply_to_key_alias_at_accounting_time(time_controller, monkeypatch):
     """The entry only applies to `premium-key`; a non-listed key's spend must