fix(health)!: let configured deployment parameters win over request overrides

When a connection test names a model that resolves to a configured deployment,
that deployment's routing and credential parameters are authoritative. A request
supplying a complete connection of its own is unaffected.

BREAKING CHANGE: /health/test_connection no longer lets a request replace the
routing or credential parameters of a configured model it names. Supply the full
connection parameters instead of naming a configured model.

(cherry picked from commit e5effcb861)

litellm/proxy/health_endpoints/_health_endpoints.py

@@ -6,7 +6,17 @@ import secrets
 import time
 import traceback
 from datetime import datetime, timedelta
-from typing import Any, Dict, Iterable, Literal, Optional, TypedDict, Union, cast
+from typing import (
+    Any,
+    Dict,
+    Iterable,
+    Literal,
+    Mapping,
+    Optional,
+    TypedDict,
+    Union,
+    cast,
+)
 
 import fastapi
 from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
@@ -28,6 +38,9 @@ from litellm.proxy._types import (
     UserAPIKeyAuth,
     WebhookEvent,
 )
+from litellm.proxy.auth.auth_utils import (
+    _BANNED_REQUEST_BODY_PARAMS,  # pyright: ignore[reportPrivateUsage]  # one canonical list, shared with the request-body check
+)
 from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
 from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler
 from litellm.proxy.health_check import (
@@ -79,6 +92,27 @@ def _reject_os_environ_references(params: dict) -> None:
                 stack.append(value)
 
 
+def _reject_banned_param_overrides(request_params: Mapping[str, object]) -> None:
+    """Reject request params that would replace a configured deployment's routing or credentials.
+
+    Applied only when a configured deployment supplies the base parameters. The
+    request may still adjust benign fields; routing and credential fields come
+    from the configuration. A caller who wants a fully custom connection supplies
+    the complete parameter set instead of naming a configured model.
+    """
+    for param in _BANNED_REQUEST_BODY_PARAMS:
+        if param in request_params:
+            raise HTTPException(
+                status_code=400,
+                detail={
+                    "error": (
+                        f"{param} cannot be overridden when testing a configured model. "
+                        "Provide the full connection parameters instead of naming a configured model."
+                    )
+                },
+            )
+
+
 def get_callback_identifier(callback):
     """
     Get the callback identifier string, handling both strings and objects.
@@ -1860,7 +1894,6 @@ async def test_model_connection(
                 )
 
         # Merge: config params (from proxy config) as base, request params override
-        # This allows users to override specific params while using config for credentials
         litellm_params = {**config_litellm_params, **request_litellm_params}
 
         ## Auth check
@@ -1875,6 +1908,8 @@ async def test_model_connection(
             prisma_client=prisma_client,
             premium_user=premium_user,
         )
+        if config_litellm_params:
+            _reject_banned_param_overrides(request_litellm_params)
         # Include health_check_params if provided
         litellm_params = _update_litellm_params_for_health_check(
             model_info={},

tests/test_litellm/proxy/health_endpoints/test_health_endpoints.py

@@ -2364,3 +2364,32 @@ def test_clean_endpoint_data_strips_credentials_keeps_routing_fields():
     assert "aws_access_key_id" not in cleaned
     assert cleaned.get("api_base") == "https://example.test/v1"
     assert cleaned.get("api_version") == "2024-10-21"
+
+
+class TestRejectBannedParamOverrides:
+    """Routing and credential fields come from the deployment configuration when a
+    request names a configured model; a request that wants its own connection
+    supplies the whole parameter set instead."""
+
+    def test_banned_param_is_refused(self):
+        from fastapi import HTTPException
+
+        from litellm.proxy.health_endpoints._health_endpoints import (
+            _reject_banned_param_overrides,
+        )
+
+        for param in ("api_base", "base_url", "vertex_credentials", "aws_web_identity_token"):
+            with pytest.raises(HTTPException) as exc_info:
+                _reject_banned_param_overrides({"model": "gpt-4o", param: "caller-supplied"})
+            assert exc_info.value.status_code == 400
+            assert param in str(exc_info.value.detail)
+
+    def test_benign_params_are_allowed(self):
+        from litellm.proxy.health_endpoints._health_endpoints import (
+            _reject_banned_param_overrides,
+        )
+
+        _reject_banned_param_overrides({})
+        _reject_banned_param_overrides({"model": "gpt-4o"})
+        _reject_banned_param_overrides({"model": "gpt-4o", "api_key": "sk-caller-owned"})
+        _reject_banned_param_overrides({"mode": "chat", "timeout": 30})