diff --git a/fastdeploy/entrypoints/openai/api_server.py b/fastdeploy/entrypoints/openai/api_server.py index 75494c372f6..11d8e651733 100644 --- a/fastdeploy/entrypoints/openai/api_server.py +++ b/fastdeploy/entrypoints/openai/api_server.py @@ -14,7 +14,6 @@ """ import asyncio -import json import os import signal import threading @@ -770,30 +769,6 @@ async def metrics(): return Response(metrics_text, media_type="text/plain") -@metrics_app.get("/config-info") -@tracing.trace_span("config-info") -def config_info() -> Response: - """ - Get the current configuration of the API server. - """ - global llm_engine - if llm_engine is None: - return Response("Engine not loaded", status_code=500) - cfg = llm_engine.cfg - - def process_object(obj): - if hasattr(obj, "__dict__"): - # 处理有__dict__属性的对象 - return obj.__dict__ - return None # 或其他默认处理 - - cfg_dict = {k: v for k, v in cfg.__dict__.items()} - env_dict = {k: v() for k, v in environment_variables.items()} - cfg_dict["env_config"] = env_dict - result_content = json.dumps(cfg_dict, default=process_object, ensure_ascii=False) - return Response(result_content, media_type="application/json") - - def run_metrics_server(): """ run metrics server diff --git a/tests/e2e/test_ernie_03b_pd_router_v0_ipc.py b/tests/e2e/test_ernie_03b_pd_router_v0_ipc.py index edb6eafdb38..d0c069b760b 100644 --- a/tests/e2e/test_ernie_03b_pd_router_v0_ipc.py +++ b/tests/e2e/test_ernie_03b_pd_router_v0_ipc.py @@ -261,13 +261,6 @@ def headers(): return {"Content-Type": "application/json"} -def test_metrics_config(metrics_url): - timeout = 600 - url = metrics_url.replace("metrics", "config-info") - res = requests.get(url, timeout=timeout) - assert res.status_code == 200 - - def send_request(url, payload, timeout=60): """ 发送请求到指定的URL，并返回响应结果。 diff --git a/tests/e2e/test_ernie_03b_pd_router_v1_ipc.py b/tests/e2e/test_ernie_03b_pd_router_v1_ipc.py index 55074c97481..646115e5cc2 100644 --- a/tests/e2e/test_ernie_03b_pd_router_v1_ipc.py +++ b/tests/e2e/test_ernie_03b_pd_router_v1_ipc.py @@ -262,13 +262,6 @@ def headers(): return {"Content-Type": "application/json"} -def test_metrics_config(metrics_url): - timeout = 600 - url = metrics_url.replace("metrics", "config-info") - res = requests.get(url, timeout=timeout) - assert res.status_code == 200 - - def send_request(url, payload, timeout=60): """ 发送请求到指定的URL，并返回响应结果。 diff --git a/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp1.py b/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp1.py index 130f39ab15a..afee279a7fe 100644 --- a/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp1.py +++ b/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp1.py @@ -266,13 +266,6 @@ def headers(): return {"Content-Type": "application/json"} -def test_metrics_config(metrics_url): - timeout = 600 - url = metrics_url.replace("metrics", "config-info") - res = requests.get(url, timeout=timeout) - assert res.status_code == 200 - - def send_request(url, payload, timeout=60): """ 发送请求到指定的URL，并返回响应结果。 diff --git a/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp2.py b/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp2.py index 05482792e44..1b0a75ad2a1 100644 --- a/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp2.py +++ b/tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp2.py @@ -270,13 +270,6 @@ def headers(): return {"Content-Type": "application/json"} -def test_metrics_config(metrics_url): - timeout = 600 - url = metrics_url.replace("metrics", "config-info") - res = requests.get(url, timeout=timeout) - assert res.status_code == 200 - - def send_request(url, payload, timeout=60): """ 发送请求到指定的URL，并返回响应结果。 diff --git a/tests/e2e/test_ernie_03b_router.py b/tests/e2e/test_ernie_03b_router.py index da9b1aae8ed..704955911c1 100644 --- a/tests/e2e/test_ernie_03b_router.py +++ b/tests/e2e/test_ernie_03b_router.py @@ -267,13 +267,6 @@ def headers(): return {"Content-Type": "application/json"} -def test_metrics_config(metrics_url): - timeout = 600 - url = metrics_url.replace("metrics", "config-info") - res = requests.get(url, timeout=timeout) - assert res.status_code == 200 - - def send_request(url, payload, timeout=60): """ 发送请求到指定的URL，并返回响应结果。 diff --git a/tests/entrypoints/openai/test_api_server.py b/tests/entrypoints/openai/test_api_server.py index 48704e026b6..de781d64734 100644 --- a/tests/entrypoints/openai/test_api_server.py +++ b/tests/entrypoints/openai/test_api_server.py @@ -803,14 +803,6 @@ def test_control_scheduler(): scheduler2.update_config.assert_not_called() -def test_config_info(): - args = _build_args() - with _patch_common_imports(args): - api_server = _reload_api_server(args) - api_server.llm_engine = None - assert api_server.config_info().status_code == 500 - - # ── /v1/abort_requests ────────────────────────────────────────────── diff --git a/tests/entrypoints/openai/test_metrics_routes.py b/tests/entrypoints/openai/test_metrics_routes.py index 8a486987e28..bb3237d64f0 100644 --- a/tests/entrypoints/openai/test_metrics_routes.py +++ b/tests/entrypoints/openai/test_metrics_routes.py @@ -21,7 +21,6 @@ import asyncio import importlib -import json import os import tempfile from types import SimpleNamespace @@ -83,7 +82,7 @@ def _get_route(app, path: str): return None -def test_metrics_and_config_routes(): +def test_metrics_route(): with ( patch("fastdeploy.utils.FlexibleArgumentParser.parse_args") as mock_parse_args, patch("fastdeploy.utils.retrive_model_from_server") as mock_retrive_model, @@ -100,7 +99,6 @@ def test_metrics_and_config_routes(): api_server = importlib.reload(api_server_mod) - # 1) /metrics from fastdeploy.metrics import metrics as metrics_mod if not hasattr(metrics_mod.main_process_metrics, "cache_config_info"): @@ -116,89 +114,6 @@ def test_metrics_and_config_routes(): ) assert "fastdeploy:" in metrics_text - # 2) /config-info - # Inject a fake engine so /config-info returns 200 - from types import SimpleNamespace as NS - - api_server.llm_engine = NS(cfg=NS(dummy="value")) - - cfg_route = _get_route(api_server.app, "/config-info") - assert cfg_route is not None - - cfg_resp = cfg_route.endpoint() - assert cfg_resp.status_code == 200 - assert getattr(cfg_resp, "media_type", "").startswith("application/json") - cfg_text = ( - cfg_resp.body.decode("utf-8") if isinstance(cfg_resp.body, (bytes, bytearray)) else str(cfg_resp.body) - ) - data = json.loads(cfg_text) - assert isinstance(data, dict) - assert "env_config" in data - - -def test_config_info_engine_not_loaded_returns_500(): - # Ensure we take the branch where llm_engine is None - with ( - patch("fastdeploy.utils.FlexibleArgumentParser.parse_args") as mock_parse_args, - patch("fastdeploy.utils.retrive_model_from_server") as mock_retrive_model, - patch("fastdeploy.entrypoints.chat_utils.load_chat_template") as mock_load_template, - ): - mock_parse_args.return_value = _build_mock_args() - mock_retrive_model.return_value = "test-model" - mock_load_template.return_value = None - - from fastdeploy.entrypoints.openai import api_server as api_server_mod - - api_server = importlib.reload(api_server_mod) - - # Fresh import sets llm_engine to None - cfg_route = _get_route(api_server.app, "/config-info") - assert cfg_route is not None - - resp = cfg_route.endpoint() - assert resp.status_code == 500 - # message body is simple text - assert b"Engine not loaded" in getattr(resp, "body", b"") - - -def test_config_info_process_object_branches(): - # Cover forcing json default() to handle - # both an object with __dict__ and one without. - with ( - patch("fastdeploy.utils.FlexibleArgumentParser.parse_args") as mock_parse_args, - patch("fastdeploy.utils.retrive_model_from_server") as mock_retrive_model, - patch("fastdeploy.entrypoints.chat_utils.load_chat_template") as mock_load_template, - ): - mock_parse_args.return_value = _build_mock_args() - mock_retrive_model.return_value = "test-model" - mock_load_template.return_value = None - - from fastdeploy.entrypoints.openai import api_server as api_server_mod - - api_server = importlib.reload(api_server_mod) - - # Build a cfg with values that exercise both branches of process_object() - class WithDict: - pass - - has_dict = WithDict() - has_dict.a = 1 - no_dict = object() - - from types import SimpleNamespace as NS - - api_server.llm_engine = NS(cfg=NS(with_dict=has_dict, without_dict=no_dict)) - - cfg_route = _get_route(api_server.app, "/config-info") - assert cfg_route is not None - - resp = cfg_route.endpoint() - assert resp.status_code == 200 - data = json.loads(resp.body.decode("utf-8")) - # The object with __dict__ becomes its dict; the one without becomes null - assert data.get("with_dict") == {"a": 1} - assert "without_dict" in data and data["without_dict"] is None - def test_metrics_app_routes_when_metrics_port_diff(): # Cover metrics_app '/metrics' @@ -224,45 +139,3 @@ def test_metrics_app_routes_when_metrics_port_diff(): assert getattr(resp, "media_type", "").startswith("text/plain") text = resp.body.decode("utf-8") if isinstance(resp.body, (bytes, bytearray)) else str(resp.body) assert "fastdeploy:" in text - - -def test_metrics_app_config_info_branches(): - # Cover metrics_app '/config-info' 500 branch and success path - # including process_object branches and response - with ( - patch("fastdeploy.utils.FlexibleArgumentParser.parse_args") as mock_parse_args, - patch("fastdeploy.utils.retrive_model_from_server") as mock_retrive_model, - patch("fastdeploy.entrypoints.chat_utils.load_chat_template") as mock_load_template, - ): - mock_parse_args.return_value = _build_mock_args_with_side_metrics() - mock_retrive_model.return_value = "test-model" - mock_load_template.return_value = None - - from fastdeploy.entrypoints.openai import api_server as api_server_mod - - api_server = importlib.reload(api_server_mod) - - # First, llm_engine is None -> 500 - cfg_route = _get_route(api_server.metrics_app, "/config-info") - assert cfg_route is not None - resp = cfg_route.endpoint() - assert resp.status_code == 500 - - # Then set a fake engine with cfg carrying both serializable and non-serializable objects - class WithDict: - pass - - has_dict = WithDict() - has_dict.x = 42 - no_dict = object() - - from types import SimpleNamespace as NS - - api_server.llm_engine = NS(cfg=NS(with_dict=has_dict, without_dict=no_dict)) - - resp2 = cfg_route.endpoint() - assert resp2.status_code == 200 - data = json.loads(resp2.body.decode("utf-8")) - assert data.get("with_dict") == {"x": 42} - assert "without_dict" in data and data["without_dict"] is None - assert "env_config" in data