fix(proxy): strip tool config from health check probes (#31008)

Health checks reuse the deployment's litellm_params, so a deployment with extra_body.tools (e.g. openrouter:web_search) sent those server tools on the trivial probe and OpenRouter 500'd it, while real completions on the same deployment succeeded. _update_litellm_params_for_health_check now works on a copy and strips tools and tool_choice from both the top level and extra_body before probing, since a trivial probe never needs tool configuration

litellm/proxy/health_check.py

@@ -47,6 +47,13 @@ _MAX_TOKEN_SUPPORT_MODES: frozenset[str] = frozenset(
     {"chat", "completion", "responses"}
 )
 
+# Tool config is meaningless on a trivial health probe and makes some providers
+# fail it outright (OpenRouter server tools like `openrouter:web_search` 500 the
+# probe). Stripped from both the top-level params and `extra_body` before probing.
+_HEALTH_CHECK_STRIPPED_REQUEST_KEYS: frozenset[str] = frozenset(
+    {"tools", "tool_choice"}
+)
+
 
 def _resolve_health_check_mode(
     model_info: Mapping[str, object], litellm_params: Mapping[str, object]
@@ -452,7 +459,20 @@ def _update_litellm_params_for_health_check(
     - updates the `model` param with the `health_check_model` if it exists Doc: https://docs.litellm.ai/docs/proxy/health#wildcard-routes
     - updates the `voice` param with the `health_check_voice` for `audio_speech` mode if it exists Doc: https://docs.litellm.ai/docs/proxy/health#text-to-speech-models
     - for Bedrock models with region routing (bedrock/region/model), strips the litellm routing prefix but preserves the model ID, and pins `custom_llm_provider` to `bedrock` (only when the deployment hasn't already set one, so an explicit `bedrock_converse` survives) so the bare model id still resolves to the provider (e.g. cross-region ids like `us.cohere.embed-v4:0`)
+    - works on a copy and strips `tools`/`tool_choice` from the top level and `extra_body`; a trivial probe never needs tool config and some providers (e.g. OpenRouter server tools) 500 when it is present
     """
+    litellm_params = {
+        k: v
+        for k, v in litellm_params.items()
+        if k not in _HEALTH_CHECK_STRIPPED_REQUEST_KEYS
+    }
+    _extra_body = litellm_params.get("extra_body")
+    if isinstance(_extra_body, dict):
+        litellm_params["extra_body"] = {
+            k: v
+            for k, v in _extra_body.items()
+            if k not in _HEALTH_CHECK_STRIPPED_REQUEST_KEYS
+        }
     mode = _resolve_health_check_mode(
         model_info, litellm_params  # any-ok: untyped router config dict
     )

tests/litellm_utils_tests/test_health_check.py

@@ -447,6 +447,62 @@ def test_update_litellm_params_for_health_check():
     )
 
 
+def test_health_check_strips_tool_config_from_probe():
+    """
+    Regression for #31008: health probes must not carry tool configuration.
+
+    OpenRouter server tools configured via `extra_body.tools` (e.g.
+    `openrouter:web_search`) made the trivial probe 500. `tools`/`tool_choice`
+    are stripped from both the top level and `extra_body`, while other
+    `extra_body` fields (e.g. provider routing) are preserved, and the caller's
+    params are never mutated.
+    """
+    from litellm.proxy.health_check import _update_litellm_params_for_health_check
+
+    extra_body = {
+        "tools": [{"type": "openrouter:web_search"}],
+        "tool_choice": "auto",
+        "provider": {"order": ["OpenAI"]},
+    }
+    litellm_params = {
+        "model": "openrouter/openai/gpt-4o-mini",
+        "api_key": "fake_key",
+        "tools": [{"type": "function", "function": {"name": "f"}}],
+        "tool_choice": "auto",
+        "extra_body": extra_body,
+    }
+
+    updated = _update_litellm_params_for_health_check({}, litellm_params)
+
+    # probe is stripped of all tool config, top-level and nested
+    assert "tools" not in updated and "tool_choice" not in updated
+    assert "tools" not in updated["extra_body"]
+    assert "tool_choice" not in updated["extra_body"]
+    assert updated["extra_body"]["provider"] == {"order": ["OpenAI"]}
+    assert isinstance(updated["messages"], list)
+
+    # caller's params are not mutated: a fresh object, no probe fields leaked in,
+    # tool config intact both top-level and nested
+    assert updated is not litellm_params
+    assert "messages" not in litellm_params
+    assert "tools" in litellm_params and "tool_choice" in litellm_params
+    assert extra_body["tools"] == [{"type": "openrouter:web_search"}]
+    assert extra_body["tool_choice"] == "auto"
+    assert updated["extra_body"] is not extra_body
+
+
+def test_health_check_probe_without_extra_body_does_not_crash():
+    """A deployment with no `extra_body` must still build a valid probe."""
+    from litellm.proxy.health_check import _update_litellm_params_for_health_check
+
+    updated = _update_litellm_params_for_health_check(
+        {}, {"model": "openai/gpt-4o-mini", "api_key": "fake_key"}
+    )
+
+    assert "extra_body" not in updated
+    assert isinstance(updated["messages"], list)
+
+
 @pytest.mark.asyncio
 async def test_perform_health_check_filters_by_model_id():
     """