[BugFix] fix work metrics not returned by metrics api (#4912)

* [BugFix] fix work metrics not returned by metrics api

* [fix] fix conflict

* [fix] fix ci

Author: liyonghua0910

fastdeploy/__init__.py

@@ -17,11 +17,20 @@
 import os
 import subprocess
 import sys
+import uuid
 
 # suppress warning log from paddlepaddle
 os.environ["GLOG_minloglevel"] = "2"
 # suppress log from aistudio
 os.environ["AISTUDIO_LOG"] = "critical"
+# set prometheus dir
+if os.getenv("PROMETHEUS_MULTIPROC_DIR", "") == "":
+    prom_dir = f"/tmp/fd_prom_{str(uuid.uuid4())}"
+    os.environ["PROMETHEUS_MULTIPROC_DIR"] = prom_dir
+    if os.path.exists(prom_dir):
+        os.rmdir(prom_dir)
+    os.mkdir(prom_dir)
+
 import typing
 
 from paddleformers.utils.log import logger as pf_logger

fastdeploy/entrypoints/openai/api_server.py

@@ -60,7 +60,6 @@
 from fastdeploy.envs import environment_variables
 from fastdeploy.metrics.metrics import (
     EXCLUDE_LABELS,
-    cleanup_prometheus_files,
     get_filtered_metrics,
     main_process_metrics,
 )
@@ -630,17 +629,19 @@ def launch_metrics_server():
     if not is_port_available(args.host, args.metrics_port):
         raise Exception(f"The parameter `metrics_port`:{args.metrics_port} is already in use.")
 
-    prom_dir = cleanup_prometheus_files(True)
-    os.environ["PROMETHEUS_MULTIPROC_DIR"] = prom_dir
+    # Move setting prometheus directory to fastdeploy/__init__.py
+    # prom_dir = cleanup_prometheus_files(True)
+    # os.environ["PROMETHEUS_MULTIPROC_DIR"] = prom_dir
     metrics_server_thread = threading.Thread(target=run_metrics_server, daemon=True)
     metrics_server_thread.start()
     time.sleep(1)
 
 
-def setup_metrics_environment():
-    """Prepare Prometheus multiprocess directory before starting API workers."""
-    prom_dir = cleanup_prometheus_files(True)
-    os.environ["PROMETHEUS_MULTIPROC_DIR"] = prom_dir
+# NOTE: This is commented out since PROMETHEUS_MULTIPROC_DIR is already set up in fastdeploy/__init__.py
+# def setup_metrics_environment():
+#     """Prepare Prometheus multiprocess directory before starting API workers."""
+#     prom_dir = cleanup_prometheus_files(True)
+#     os.environ["PROMETHEUS_MULTIPROC_DIR"] = prom_dir
 
 
 controller_app = FastAPI()
@@ -755,7 +756,7 @@ def main():
         launch_metrics_server()
         console_logger.info(f"Launching metrics service at http://{args.host}:{args.metrics_port}/metrics")
     else:
-        setup_metrics_environment()
+        # setup_metrics_environment()
         console_logger.info(f"Launching metrics service at http://{args.host}:{args.port}/metrics")
     console_logger.info(f"Launching chat completion service at http://{args.host}:{args.port}/v1/chat/completions")
     console_logger.info(f"Launching completion service at http://{args.host}:{args.port}/v1/completions")

tests/entrypoints/openai/test_metrics_routes.py

@@ -184,27 +184,17 @@ class WithDict:
         assert "without_dict" in data and data["without_dict"] is None
 
 
-def test_setup_metrics_environment_sets_env_var(tmp_path):
-    # Cover calling setup_metrics_environment()
-    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)
-
-        desired_dir = str(tmp_path / "prom_multiproc")
-
-        # Patch the name imported into api_server so we don't touch real FS
-        with patch("fastdeploy.entrypoints.openai.api_server.cleanup_prometheus_files", return_value=desired_dir):
-            api_server.setup_metrics_environment()
-            assert os.environ.get("PROMETHEUS_MULTIPROC_DIR") == desired_dir
+def test_setup_metrics_environment_sets_env_var():
+    # Cover 'fastdeploy/__init__.py'
+    import fastdeploy
+
+    # clear PROMETHEUS_MULTIPROC_DIR
+    os.environ.pop("PROMETHEUS_MULTIPROC_DIR")
+    # reload fastdeploy
+    assert os.environ.get("PROMETHEUS_MULTIPROC_DIR") is None
+    importlib.reload(fastdeploy)
+    # check if PROMETHEUS_MULTIPROC_DIR is reinitialized
+    assert os.environ.get("PROMETHEUS_MULTIPROC_DIR").startswith("/tmp/fd_prom")
 
 
 def test_metrics_app_routes_when_metrics_port_diff():