完善系统健康面板并接入AI运行态观测

admin/dashboard/blueprints/system.py

@@ -15,6 +15,7 @@ import toml
 from utils.markdown_to_image import get_md2img_health_snapshot, warmup_md2img_browser_sync
 from utils.ai.llm_registry import LLMRegistry
 from base.plugin_common.plugin_interface import PluginStatus
+from utils.ai.unified_llm import UnifiedLLMClient
 
 # 创建系统信息蓝图
 system_bp = Blueprint('system', __name__)
@@ -403,6 +404,33 @@ def api_system_health_summary():
         # 错误数量直接复用现有统计库，避免为了首页卡片再单独写一套 SQL。
         _, recent_error_count = server.stats_db.get_error_logs(days=1, page=1, limit=1)
 
+        # 基础设施健康：
+        # 1. MySQL 用最轻量的 SELECT 1 做可用性探测；
+        # 2. Redis 用 PING 验证连接池当前是否可拿到可用连接；
+        # 3. 即使探测失败也只反馈到看板，不影响主接口整体返回。
+        mysql_status = "healthy"
+        mysql_summary = "连接正常"
+        try:
+            mysql_conn = server.db_manager.get_mysql_connection()
+            try:
+                with mysql_conn.cursor() as cursor:
+                    cursor.execute("SELECT 1")
+                    cursor.fetchone()
+            finally:
+                mysql_conn.close()
+        except Exception as mysql_error:
+            mysql_status = "danger"
+            mysql_summary = f"MySQL 探测失败: {mysql_error}"
+
+        redis_status = "healthy"
+        redis_summary = "连接正常"
+        try:
+            redis_conn = server.db_manager.get_redis_connection()
+            redis_conn.ping()
+        except Exception as redis_error:
+            redis_status = "danger"
+            redis_summary = f"Redis 探测失败: {redis_error}"
+
         # md2img 健康快照已经有现成实现，这里只做聚合，不主动预热运行时。
         md2img_snapshot = get_md2img_health_snapshot(ensure_runtime=False) or {}
         browser_ready = bool(
@@ -452,6 +480,28 @@ def api_system_health_summary():
             md2img_status = "danger"
             md2img_summary = "运行时未就绪，相关转图能力可能不可用"
 
+        # AI 运行态：
+        # 1. 统一从 UnifiedLLMClient 最近调用窗口读取，避免各插件单独维护监控数据；
+        # 2. 若当前窗口还没有调用记录，就明确返回“暂无调用”，避免误判成异常。
+        ai_runtime = UnifiedLLMClient.get_runtime_snapshot()
+        ai_total_calls = int(ai_runtime.get("total_calls") or 0)
+        ai_failed_calls = int(ai_runtime.get("failed_calls") or 0)
+        if ai_total_calls <= 0:
+            ai_status = "warning"
+            ai_summary = "最近窗口内暂无统一 LLM 调用记录"
+        elif ai_failed_calls > 0:
+            ai_status = "warning"
+            ai_summary = (
+                f"最近 {ai_total_calls} 次调用中失败 {ai_failed_calls} 次，"
+                f"平均耗时 {ai_runtime.get('avg_latency_ms', 0)}ms"
+            )
+        else:
+            ai_status = "healthy"
+            ai_summary = (
+                f"最近 {ai_total_calls} 次调用全部成功，"
+                f"平均耗时 {ai_runtime.get('avg_latency_ms', 0)}ms"
+            )
+
         return jsonify({
             "success": True,
             "data": {
@@ -473,6 +523,27 @@ def api_system_health_summary():
                     "recent_24h_count": recent_error_count,
                     "summary": error_summary,
                 },
+                "infrastructure": {
+                    "status": "healthy" if mysql_status == "healthy" and redis_status == "healthy" else "danger",
+                    "summary": (
+                        "MySQL / Redis 均正常"
+                        if mysql_status == "healthy" and redis_status == "healthy"
+                        else "存在基础设施连接异常"
+                    ),
+                    "mysql": {
+                        "status": mysql_status,
+                        "summary": mysql_summary,
+                    },
+                    "redis": {
+                        "status": redis_status,
+                        "summary": redis_summary,
+                    },
+                },
+                "ai_runtime": {
+                    "status": ai_status,
+                    "summary": ai_summary,
+                    **ai_runtime,
+                },
                 "md2img": {
                     "status": md2img_status,
                     "healthy": md2img_healthy,