feat: wire agent runtime metrics

tests/test_agent_tokens_events.py

@@ -94,6 +94,36 @@ def test_initialize_llm_uses_chain_event_selection(monkeypatch) -> None:
     assert agent._llm_provider_selection["selected_provider_id"] == "provider-1"
 
 
+def test_record_usage_emits_low_cardinality_token_metrics() -> None:
+    """Token 指标只暴露归一供应商类别与输入输出方向。"""
+    agent = MoviePilotAgent(session_id="usage-metrics", user_id="user-1")
+    agent._llm_provider_selection = {"provider": "private-provider-name"}
+
+    with patch("app.agent.orchestrator.record_metric") as record_metric:
+        agent._record_usage(
+            {
+                "has_usage": True,
+                "input_usage_available": True,
+                "input_tokens": 120,
+                "output_tokens": 30,
+                "total_tokens": 150,
+            }
+        )
+
+    record_metric.assert_any_call(
+        "agent.token_usage",
+        120,
+        provider_type="custom",
+        direction="input",
+    )
+    record_metric.assert_any_call(
+        "agent.token_usage",
+        30,
+        provider_type="custom",
+        direction="output",
+    )
+
+
 def test_execute_agent_broadcasts_usage_on_success() -> None:
     """Agent 执行成功后应广播聚合 token 用量事件。"""
     agent = MoviePilotAgent(session_id="usage-success", user_id="user-1")

tests/test_async_db_pooling.py

@@ -14,6 +14,7 @@ Connection 与 aiosqlite 的线程都绑定在创建它的循环上。因此池
 """
 import asyncio
 import threading
+from unittest.mock import patch
 
 import pytest
 
@@ -163,8 +164,16 @@ def test_fallback_slot_times_out_when_exhausted(monkeypatch):
 
     async def run():
         db_module._fallback_slots.acquire()  # 占满唯一名额
-        with pytest.raises(TimeoutError):
-            await db_module._acquire_fallback_slot()
+        with patch("app.db.session.record_metric") as record_metric:
+            with pytest.raises(TimeoutError):
+                await db_module._acquire_fallback_slot()
+        record_metric.assert_any_call("db.pool.timeout", backend="sqlite")
+        record_metric.assert_any_call(
+            "db.pool.wait",
+            pytest.approx(0.05, abs=0.03),
+            backend="sqlite",
+            outcome="timeout",
+        )
 
     asyncio.run(run())
 

tests/test_observability.py

@@ -5,12 +5,16 @@ from typing import Mapping
 
 import httpx
 import pytest
+from sqlalchemy import create_engine
 from starlette.applications import Starlette
 from starlette.responses import PlainTextResponse
 from starlette.routing import Route
 
 from app.adapters.observability import otel
 from app.adapters.web.metrics import HttpMetricsMiddleware
+from app.db.engine import _register_database_pool_metrics
+from app.runtime.extensions.plugin.lifecycle import observe_plugin_lifecycle
+from app.schemas.plugin import PluginRuntimeStatus
 from app.runtime.observability import (
     METRIC_SPECS,
     MetricSpec,
@@ -137,3 +141,36 @@ def test_optional_otel_adapter_falls_back_to_noop(monkeypatch) -> None:
     monkeypatch.setattr(otel.importlib, "import_module", missing)
 
     assert isinstance(otel.build_observation_port(), NoopObservationPort)
+
+
+def test_database_pool_checkout_updates_gauge() -> None:
+    """真实 SQLAlchemy checkout/checkin 应成对维护连接借出量。"""
+    port = RecordingObservationPort()
+    configure_observation(port)
+    engine = create_engine("sqlite://")
+    _register_database_pool_metrics(engine)
+
+    with engine.connect():
+        pass
+    engine.dispose()
+
+    records = [record for record in port.records if record[0].name == "db.pool.checked_out"]
+    assert [record[1] for record in records] == [1.0, -1.0]
+    assert all(record[2] == {"backend": "sqlite"} for record in records)
+
+
+def test_plugin_lifecycle_failed_status_records_error_outcome() -> None:
+    """被插件生命周期内部收敛的加载失败仍应记录 error。"""
+    port = RecordingObservationPort()
+    configure_observation(port)
+
+    @observe_plugin_lifecycle("start")
+    def load_plugin() -> dict[str, PluginRuntimeStatus]:
+        """返回插件加载失败状态。"""
+        return {"Example": PluginRuntimeStatus.LOAD_FAILED}
+
+    load_plugin()
+
+    spec, _, labels = port.records[-1]
+    assert spec.name == "plugin.lifecycle.duration"
+    assert labels == {"operation": "start", "outcome": "error"}