fix(auth): apply temp_budget_increase for cache-hit keys

litellm/proxy/auth/user_api_key_auth.py

@@ -1673,10 +1673,9 @@ async def _user_api_key_auth_builder(
             valid_token.end_user_tpm_limit = end_user_params.get("end_user_tpm_limit")
             valid_token.end_user_rpm_limit = end_user_params.get("end_user_rpm_limit")
             valid_token.allowed_model_region = end_user_params.get("allowed_model_region")
-            # update key budget with temp budget increase
-            valid_token = _update_key_budget_with_temp_budget_increase(
-                valid_token
-            )  # updating it here, allows all downstream reporting / checks to use the updated budget
+
+        if valid_token is not None:
+            valid_token = _update_key_budget_with_temp_budget_increase(valid_token)
 
         user_obj: Optional[LiteLLM_UserTable] = None
         valid_token_dict: dict = {}

tests/test_litellm/proxy/auth/test_user_api_key_auth.py

@@ -4515,3 +4515,73 @@ async def test_raises_when_model_from_nested_field(self):
 
         assert exc_info.value is original_error
         assert "model" not in request_data
+
+
+@pytest.mark.asyncio
+async def test_temp_budget_increase_applied_for_cached_key():
+    """
+    Regression for https://github.com/BerriAI/litellm/issues/25760
+
+    temp_budget_increase used to be applied only on the DB-fetch path, so a key
+    served from cache kept its original max_budget and was wrongly blocked once
+    spend crossed the original budget (but stayed under the effective budget).
+
+    Seed the auth cache with a key whose spend (5.0) exceeds its original
+    max_budget (2.0) but is under the effective budget (2.0 + 100.0). The cache-hit
+    request must not raise and the resolved token must carry max_budget == 102.0.
+    """
+    from datetime import datetime, timedelta
+
+    from litellm.proxy.utils import hash_token
+
+    api_key = "sk-temp-budget-cache-regression"
+    hashed_token = hash_token(api_key)
+    expiry = (datetime.now() + timedelta(days=1)).isoformat()
+
+    cached_key = UserAPIKeyAuth(
+        token=hashed_token,
+        max_budget=2.0,
+        spend=5.0,
+        metadata={"temp_budget_increase": 100.0, "temp_budget_expiry": expiry},
+    )
+
+    user_api_key_cache = DualCache()
+    await _cache_key_object(
+        hashed_token=hashed_token,
+        user_api_key_obj=cached_key,
+        user_api_key_cache=user_api_key_cache,
+        proxy_logging_obj=None,
+    )
+
+    mock_request = MagicMock()
+    mock_request.url.path = "/v1/chat/completions"
+    mock_request.method = "POST"
+    mock_request.headers = {"authorization": f"Bearer {api_key}"}
+    mock_request.query_params = {}
+    mock_request.state = SimpleNamespace()
+
+    proxy_logging_obj = MagicMock()
+    proxy_logging_obj.budget_alerts = AsyncMock()
+
+    with (
+        patch("litellm.proxy.proxy_server.general_settings", {}),
+        patch("litellm.proxy.proxy_server.master_key", "sk-master"),
+        patch("litellm.proxy.proxy_server.prisma_client", MagicMock()),
+        patch("litellm.proxy.proxy_server.user_api_key_cache", user_api_key_cache),
+        patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging_obj),
+        patch(
+            "litellm.proxy.auth.user_api_key_auth._virtual_key_max_budget_alert_check",
+            new_callable=AsyncMock,
+        ),
+    ):
+        result = await _user_api_key_auth_builder(
+            request=mock_request,
+            api_key=f"Bearer {api_key}",
+            azure_api_key_header="",
+            anthropic_api_key_header=None,
+            google_ai_studio_api_key_header=None,
+            azure_apim_header=None,
+            request_data={"model": "gpt-4o-mini"},
+        )
+
+    assert result.max_budget == 102.0