fix(proxy-node): 修复服务重启后 tunnel 连接状态不一致的问题

- 启动时重置 DB 中残留的 tunnel_connected=True 状态
- health_scheduler 以 TunnelManager 内存实际连接为准判断节点状态
- tunnel 模式注册时初始状态设为 UNHEALTHY，等 tunnel 连接后再上线

src/modules/proxy_nodes/__init__.py

@@ -20,6 +20,51 @@ if TYPE_CHECKING:
     from sqlalchemy.orm import Session
 
 
+def _reset_tunnel_connected_on_startup() -> None:
+    """服务端启动时将所有 tunnel_connected=True 的节点重置为 False/UNHEALTHY。
+
+    服务端重启后 TunnelManager 内存状态丢失，但 DB 中可能残留
+    tunnel_connected=True 的记录。如果不重置，health_scheduler 会错误地
+    将这些节点标记为 ONLINE，而实际上 tunnel 并未连接。
+    """
+    from datetime import datetime, timezone
+
+    from src.core.logger import logger
+    from src.database import create_session
+    from src.models.database import ProxyNode, ProxyNodeStatus
+
+    db = create_session()
+    try:
+        now = datetime.now(timezone.utc)
+        stale_nodes = (
+            db.query(ProxyNode)
+            .filter(
+                ProxyNode.tunnel_connected == True,  # noqa: E712
+                ProxyNode.is_manual == False,  # noqa: E712
+            )
+            .all()
+        )
+        if stale_nodes:
+            for node in stale_nodes:
+                node.tunnel_connected = False
+                node.tunnel_connected_at = now
+                node.status = ProxyNodeStatus.UNHEALTHY
+                node.updated_at = now
+            db.commit()
+            logger.info(
+                "重置 {} 个节点的残留 tunnel 连接状态 (tunnel_connected -> False)",
+                len(stale_nodes),
+            )
+    except Exception as e:
+        try:
+            db.rollback()
+        except Exception:
+            pass
+        logger.warning("重置 tunnel 连接状态失败: {}", e)
+    finally:
+        db.close()
+
+
 def _get_router() -> Any:
     """延迟导入路由"""
     from src.api.admin.proxy_nodes import router
@@ -36,6 +81,11 @@ async def _on_startup() -> None:
 
     logger = logging.getLogger("aether.modules.proxy_nodes")
 
+    # 服务端启动时，TunnelManager 内存为空，所有 tunnel 连接都需要重新建立。
+    # 重置 DB 中残留的 tunnel_connected=True 状态，避免 health_scheduler
+    # 误将未连接的节点标记为 ONLINE。
+    _reset_tunnel_connected_on_startup()
+
     from src.clients import get_redis_client
 
     redis_client = await get_redis_client()

src/services/proxy_node/health_scheduler.py

@@ -53,6 +53,9 @@ class ProxyNodeHealthScheduler:
         await self._check_heartbeats()
 
     async def _check_heartbeats(self) -> None:
+        from src.services.proxy_node.tunnel_manager import get_tunnel_manager
+
+        manager = get_tunnel_manager()
         db = create_session()
         try:
             now = datetime.now(timezone.utc)
@@ -71,7 +74,20 @@ class ProxyNodeHealthScheduler:
 
             changed = 0
             for node in nodes:
-                if node.tunnel_connected:
+                # 以 TunnelManager 内存中的实际连接状态为准，
+                # 而非仅依赖 DB 的 tunnel_connected 字段。
+                # 服务端重启后 DB 可能残留 tunnel_connected=True，
+                # 但 TunnelManager 内存中已无连接。
+                actually_connected = manager.has_tunnel(node.id)
+
+                # 同步修正 DB 中不一致的 tunnel_connected 字段
+                if node.tunnel_connected != actually_connected:
+                    node.tunnel_connected = actually_connected
+                    if not actually_connected:
+                        node.tunnel_connected_at = now
+                    changed += 1
+
+                if actually_connected:
                     new_status = ProxyNodeStatus.ONLINE
                 elif node.tunnel_connected_at:
                     # tunnel 刚断开：给 60s 缓冲期标记为 UNHEALTHY

src/services/proxy_node/service.py

@@ -210,10 +210,14 @@ class ProxyNodeService:
             )
         else:
             node = db.query(ProxyNode).filter(ProxyNode.ip == ip, ProxyNode.port == port).first()
+        # tunnel 模式注册时设为 UNHEALTHY，等 WebSocket tunnel 真正连接后
+        # 由 _update_tunnel_status 设为 ONLINE；非 tunnel 模式保持原逻辑
+        initial_status = ProxyNodeStatus.UNHEALTHY if tunnel_mode else ProxyNodeStatus.ONLINE
+
         if node:
             node.name = name
             node.region = region
-            node.status = ProxyNodeStatus.ONLINE
+            node.status = initial_status
             node.last_heartbeat_at = now
             node.heartbeat_interval = heartbeat_interval
             node.tunnel_mode = tunnel_mode
@@ -234,7 +238,7 @@ class ProxyNodeService:
                 ip=ip,
                 port=port,
                 region=region,
-                status=ProxyNodeStatus.ONLINE,
+                status=initial_status,
                 registered_by=registered_by,
                 last_heartbeat_at=now,
                 heartbeat_interval=heartbeat_interval,