Merge 41b9e1e09c into 2dccc0dc79

litellm/litellm_core_utils/health_check_helpers.py

@@ -82,7 +82,16 @@ class HealthCheckHelpers:
         from litellm.proxy.litellm_pre_call_utils import LiteLLMProxyRequestSetup
 
         _metadata_variable_name: Final = "litellm_metadata"
-        litellm_metadata: Final = HealthCheckHelpers._get_metadata_for_health_check_call()
+        existing_metadata: Final = model_params.get(_metadata_variable_name)
+        health_check_metadata: Final = HealthCheckHelpers._get_metadata_for_health_check_call()
+        existing_tags: Final = existing_metadata.get("tags", ()) if isinstance(existing_metadata, dict) else ()
+        health_check_tags: Final = health_check_metadata["tags"]
+        litellm_metadata: Final = health_check_metadata.copy()
+        if isinstance(existing_metadata, dict):
+            litellm_metadata.update(existing_metadata)
+        litellm_metadata.update(health_check_metadata)
+        health_check_tags[:] = dict.fromkeys((*existing_tags, *health_check_tags))
+        litellm_metadata["tags"] = health_check_tags
         model_params[_metadata_variable_name] = litellm_metadata
         model_params = LiteLLMProxyRequestSetup.add_user_api_key_auth_to_request_metadata(
             data=model_params,

litellm/proxy/health_check.py

@@ -527,6 +527,14 @@ async def _run_model_health_check(model: dict):
         litellm_params,  # any-ok: untyped router config dict
     )
     litellm_params = _update_litellm_params_for_health_check(model_info, litellm_params)
+    existing_metadata: Final = litellm_params.get("litellm_metadata")
+    health_check_metadata: Final = (
+        existing_metadata.copy()
+        if isinstance(existing_metadata, dict)
+        else {}  # mutable-ok: health-check metadata is enriched in place
+    )
+    health_check_metadata["model_info"] = model_info
+    litellm_params["litellm_metadata"] = health_check_metadata
     timeout: Final = model_info.get("health_check_timeout") or HEALTH_CHECK_TIMEOUT_SECONDS
 
     return await run_with_timeout(

tests/test_litellm/proxy/test_health_check_max_tokens.py

@@ -481,6 +481,42 @@ async def test_run_model_health_check_threads_resolved_mode_to_ahealth_check():
     assert probed_params["model"] == "amazon.titan-embed-text-v2:0"
 
 
+@pytest.mark.asyncio
+async def test_run_model_health_check_preserves_deployment_identity_for_spend_tracking():
+    fake_ahealth_check = AsyncMock(return_value={})
+    model_info = {
+        "id": "bedrock-health-check-deployment",
+        "input_cost_per_token": 1e-06,
+        "output_cost_per_token": 5e-06,
+        "health_check_params": {"litellm_metadata": {"operator_probe_flag": True, "tags": ["operator-health-tag"]}},
+    }
+    model = {
+        "litellm_params": {"model": "bedrock/claude_platform/claude-haiku-4-5"},
+        "model_info": model_info,
+    }
+
+    with patch.object(  # test-quality-ok: observes probe payload before provider execution
+        hc_module.litellm, "ahealth_check", fake_ahealth_check
+    ):
+        await hc_module._run_model_health_check(model)
+
+    probed_params = fake_ahealth_check.call_args.args[0]
+    assert probed_params["model"] == "claude_platform/claude-haiku-4-5"
+    assert probed_params["litellm_metadata"]["operator_probe_flag"] is True
+    assert probed_params["litellm_metadata"]["model_info"] == model_info
+
+
+def test_health_check_tracking_preserves_deployment_identity():
+    model_info = {"id": "bedrock-health-check-deployment"}
+
+    updated = HealthCheckHelpers._update_model_params_with_health_check_tracking_information(
+        {"litellm_metadata": {"model_info": model_info, "tags": ["operator-health-tag"]}}
+    )
+
+    assert updated["litellm_metadata"]["model_info"] == model_info
+    assert "operator-health-tag" in updated["litellm_metadata"]["tags"]
+
+
 def test_autodetected_embedding_skips_reasoning_effort():
     """reasoning_effort must not leak into an embedding probe whose mode is auto-detected.