fix(proxy): apply DB-persisted safe litellm settings on every worker's config reload

Peer workers previously kept their startup value for block_requests_for_models_without_pricing
until a restart, so a toggle from the UI only took effect on the worker that served the request.

litellm/proxy/proxy_server.py

@@ -6823,6 +6823,19 @@ class ProxyConfig:
 
         if self._should_load_db_object(object_type="config_overrides"):
             await self._init_hashicorp_vault_config_override(prisma_client=prisma_client)
+            await self._apply_safe_litellm_settings_overrides_from_db(prisma_client=prisma_client)
+
+    async def _apply_safe_litellm_settings_overrides_from_db(self, prisma_client: PrismaClient) -> None:
+        config_record: Final = await get_config_param(prisma_client, "litellm_settings")
+        if config_record is None or config_record.param_value is None:
+            return
+        raw_settings: Final = config_record.param_value
+        litellm_settings: Final = json.loads(raw_settings) if isinstance(raw_settings, str) else raw_settings
+        if not isinstance(litellm_settings, dict):
+            return
+        for key, value in litellm_settings.items():
+            if key in LITELLM_SETTINGS_SAFE_DB_OVERRIDES:
+                setattr(litellm, key, value)
 
     async def _init_semantic_filter_settings_in_db(self, prisma_client: PrismaClient):
         """

tests/test_litellm/proxy/management_endpoints/test_cost_tracking_settings.py

@@ -740,6 +740,32 @@ class TestBlockRequestsForModelsWithoutPricing:
 
             assert litellm.block_requests_for_models_without_pricing is True
 
+    @pytest.mark.asyncio
+    async def test_periodic_db_sync_applies_flag_to_peer_worker(self):
+        """The ~10s reconcile loop runs _init_non_llm_objects_in_db on every worker; it must apply
+        the persisted flag so peers converge without a restart."""
+        from types import SimpleNamespace
+
+        from litellm.proxy.proxy_server import ProxyConfig
+
+        config_record = SimpleNamespace(
+            param_value={"block_requests_for_models_without_pricing": True, "unsafe_key": "x"}
+        )
+        with (
+            patch.object(litellm, "block_requests_for_models_without_pricing", False),
+            patch.object(
+                ProxyConfig,
+                "_should_load_db_object",
+                side_effect=lambda object_type: object_type == "config_overrides",
+            ),
+            patch.object(ProxyConfig, "_init_hashicorp_vault_config_override", AsyncMock()),
+            patch("litellm.proxy.proxy_server.get_config_param", AsyncMock(return_value=config_record)),
+        ):
+            await ProxyConfig()._init_non_llm_objects_in_db(prisma_client=MagicMock())
+
+            assert litellm.block_requests_for_models_without_pricing is True
+            assert not hasattr(litellm, "unsafe_key")
+
     @pytest.mark.asyncio
     async def test_patch_requires_store_model_in_db(self):
         with (