From 9974cf4bf817331053649fccc1f1b9e00d61d570 Mon Sep 17 00:00:00 2001
From: yucheng <yucheng@berri.ai>
Date: Wed, 16 Sep 2026 08:17:58 +0000
Subject: [PATCH] fix(proxy): honor key-level disable_fallbacks after first
 pre-call pass

Key metadata disable_fallbacks only lands on data during add_key_level_controls,
so the local rate-limit fallback retry now rechecks it post pre-call. Also use a
real UserAPIKeyAuth in the skip pre-call test since the path reads router_settings

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/proxy/common_request_processing.py    |  7 ++++-
 .../test_response_polling_pre_call_checks.py  |  2 +-
 .../proxy/test_common_request_processing.py   | 30 +++++++++++++++++--
 3 files changed, 35 insertions(+), 4 deletions(-)

diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py
index d18fab1c9f0..36b49ef064c 100644
--- a/litellm/proxy/common_request_processing.py
+++ b/litellm/proxy/common_request_processing.py
@@ -2093,7 +2093,12 @@ class ProxyBaseLLMRequestProcessing:
         except ProxyRateLimitError as original_exc:
             rate_limited_data: Final = self.data
             original_model: Final = rate_limited_data.get("model")
-            if pristine is None or not configured_fallbacks or not isinstance(original_model, str):
+            if (
+                pristine is None
+                or not configured_fallbacks
+                or rate_limited_data.get("disable_fallbacks")
+                or not isinstance(original_model, str)
+            ):
                 raise
 
             fallback_models: Final = self._resolve_fallback_models(
diff --git a/tests/proxy_unit_tests/test_response_polling_pre_call_checks.py b/tests/proxy_unit_tests/test_response_polling_pre_call_checks.py
index 459834d0fd2..9f1a228855e 100644
--- a/tests/proxy_unit_tests/test_response_polling_pre_call_checks.py
+++ b/tests/proxy_unit_tests/test_response_polling_pre_call_checks.py
@@ -87,7 +87,7 @@ class TestSkipPreCallLogic:
                 await processor.base_process_llm_request(
                     request=MagicMock(spec=Request),
                     fastapi_response=MagicMock(spec=Response),
-                    user_api_key_dict=MagicMock(spec=UserAPIKeyAuth),
+                    user_api_key_dict=UserAPIKeyAuth(),
                     route_type="aresponses",
                     proxy_logging_obj=mock_proxy_logging,
                     llm_router=MagicMock(),
diff --git a/tests/test_litellm/proxy/test_common_request_processing.py b/tests/test_litellm/proxy/test_common_request_processing.py
index 7d9a78fd981..e9d74a46daa 100644
--- a/tests/test_litellm/proxy/test_common_request_processing.py
+++ b/tests/test_litellm/proxy/test_common_request_processing.py
@@ -6410,7 +6410,9 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
 
     @staticmethod
     def _otel_key(
-        rpm_limit: int | None = None, model_rpm_limit: dict[str, int] | None = None
+        rpm_limit: int | None = None,
+        model_rpm_limit: dict[str, int] | None = None,
+        disable_fallbacks: bool = False,
     ) -> ProxyUserAPIKeyAuth:
         from opentelemetry.sdk.trace import TracerProvider
 
@@ -6419,7 +6421,10 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
             api_key="hashed-key",
             parent_otel_span=span,
             rpm_limit=rpm_limit,
-            metadata={"model_rpm_limit": model_rpm_limit} if model_rpm_limit else {},
+            metadata={
+                **({"model_rpm_limit": model_rpm_limit} if model_rpm_limit else {}),
+                **({"disable_fallbacks": True} if disable_fallbacks else {}),
+            },
         )
 
     @staticmethod
@@ -6540,6 +6545,27 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
         assert data["model"] == fallback_model
         assert rig[3] == [primary_model, primary_model, fallback_model]
 
+    @pytest.mark.asyncio
+    async def test_key_metadata_disable_fallbacks_returns_429_instead_of_retrying(
+        self, monkeypatch: pytest.MonkeyPatch
+    ):
+        """``disable_fallbacks`` set in key metadata only lands on ``data`` during the first
+        pre-call pass (``add_key_level_controls``), so it must be honored after that pass."""
+        from litellm.proxy.common_utils.proxy_rate_limit_error import ProxyRateLimitError
+
+        primary_model = "gpt-4.1"
+        fallback_model = "gpt-4.1-mini"
+        key = self._otel_key(model_rpm_limit={primary_model: 1}, disable_fallbacks=True)
+        rig = self._v3_limiter_rig(monkeypatch, key, [{primary_model: [fallback_model]}])
+        request = {"model": primary_model, "messages": [{"role": "user", "content": "hi"}]}
+
+        await self._pre_call(dict(request), key, rig)
+        with pytest.raises(ProxyRateLimitError) as exc_info:
+            await self._pre_call(dict(request), key, rig)
+
+        assert exc_info.value.status_code == 429
+        assert rig[3] == [primary_model, primary_model]
+
 
 class _RecordingSuccessLogger(CustomLogger):
     def __init__(self):