fix： agent自动更新逻辑

backend/apps/engine/services/worker_service.py

@@ -134,5 +134,57 @@ class WorkerService:
             logger.warning(f"[卸载] Worker {worker_id} 远程卸载异常: {e}")
             return False, f"远程卸载异常: {str(e)}"
 
+    def execute_remote_command(
+        self,
+        ip_address: str,
+        ssh_port: int,
+        username: str,
+        password: str | None,
+        command: str
+    ) -> tuple[bool, str]:
+        """
+        在远程主机上执行命令
+        
+        Args:
+            ip_address: SSH 主机地址
+            ssh_port: SSH 端口
+            username: SSH 用户名
+            password: SSH 密码
+            command: 要执行的命令
+            
+        Returns:
+            (success, message) 元组
+        """
+        if not password:
+            return False, "未配置 SSH 密码"
+        
+        try:
+            import paramiko
+            
+            ssh = paramiko.SSHClient()
+            ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
+            
+            ssh.connect(
+                ip_address,
+                port=ssh_port,
+                username=username,
+                password=password,
+                timeout=30
+            )
+            
+            stdin, stdout, stderr = ssh.exec_command(command, timeout=120)
+            exit_status = stdout.channel.recv_exit_status()
+            
+            ssh.close()
+            
+            if exit_status == 0:
+                return True, stdout.read().decode().strip()
+            else:
+                error = stderr.read().decode().strip()
+                return False, error
+                
+        except Exception as e:
+            return False, str(e)
+
 
 __all__ = ["WorkerService"]

backend/apps/engine/views/worker_views.py

@@ -163,6 +163,11 @@ class WorkerNodeViewSet(viewsets.ModelViewSet):
                 logger.info(
                     f"Worker {worker.name} 版本不匹配: agent={agent_version}, server={server_version}"
                 )
+                
+                # 远程 Worker：服务端主动通过 SSH 触发更新
+                # 旧版 agent 不会解析 need_update，所以需要服务端主动推送
+                if not worker.is_local and worker.ip_address:
+                    self._trigger_remote_agent_update(worker, server_version)
         
         return Response({
             'status': 'ok',
@@ -170,6 +175,73 @@ class WorkerNodeViewSet(viewsets.ModelViewSet):
             'server_version': server_version
         })
     
+    def _trigger_remote_agent_update(self, worker, target_version: str):
+        """
+        通过 SSH 触发远程 agent 更新（后台执行，不阻塞心跳响应）
+        
+        使用 Redis 锁防止重复触发（同一 worker 60秒内只触发一次）
+        """
+        import redis
+        from django.conf import settings as django_settings
+        
+        redis_client = redis.from_url(django_settings.REDIS_URL)
+        lock_key = f"agent_update_lock:{worker.id}"
+        
+        # 尝试获取锁（60秒过期，防止重复触发）
+        if not redis_client.set(lock_key, "1", nx=True, ex=60):
+            logger.debug(f"Worker {worker.name} 更新已在进行中，跳过")
+            return
+        
+        # 提取数据避免后台线程访问 ORM
+        worker_id = worker.id
+        worker_name = worker.name
+        ip_address = worker.ip_address
+        ssh_port = worker.ssh_port
+        username = worker.username
+        password = worker.password
+        
+        def _async_update():
+            try:
+                logger.info(f"开始远程更新 Worker {worker_name} 到 {target_version}")
+                
+                # 构建更新命令：拉取新镜像并重启 agent
+                docker_user = getattr(django_settings, 'DOCKER_USER', 'yyhuni')
+                update_cmd = f'''
+                    docker pull {docker_user}/xingrin-agent:{target_version} && \
+                    docker stop xingrin-agent 2>/dev/null || true && \
+                    docker rm xingrin-agent 2>/dev/null || true && \
+                    docker run -d --pull=always \
+                        --name xingrin-agent \
+                        --restart always \
+                        -e HEARTBEAT_API_URL="https://{django_settings.PUBLIC_HOST}" \
+                        -e WORKER_ID="{worker_id}" \
+                        -e IMAGE_TAG="{target_version}" \
+                        -v /proc:/host/proc:ro \
+                        {docker_user}/xingrin-agent:{target_version}
+                '''
+                
+                success, message = self.worker_service.execute_remote_command(
+                    ip_address=ip_address,
+                    ssh_port=ssh_port,
+                    username=username,
+                    password=password,
+                    command=update_cmd
+                )
+                
+                if success:
+                    logger.info(f"Worker {worker_name} 远程更新成功")
+                else:
+                    logger.warning(f"Worker {worker_name} 远程更新失败: {message}")
+                    
+            except Exception as e:
+                logger.error(f"Worker {worker_name} 远程更新异常: {e}")
+            finally:
+                # 释放锁
+                redis_client.delete(lock_key)
+        
+        # 后台执行，不阻塞心跳响应
+        threading.Thread(target=_async_update, daemon=True).start()
+    
     @action(detail=False, methods=['post'])
     def register(self, request):
         """

backend/scripts/worker-deploy/start-agent.sh

@@ -60,12 +60,14 @@ start_agent() {
     log_info "=========================================="
     
     log_info "启动 agent 容器..."
-    # --pull=missing 只在本地没有镜像时才拉取，避免意外更新
+    # --pull=missing: 本地没有镜像时才拉取
+    # 版本更新由服务端通过 SSH 显式 docker pull 触发
     docker run -d --pull=missing \
         --name ${CONTAINER_NAME} \
         --restart always \
         -e SERVER_URL="${PRESET_SERVER_URL}" \
         -e WORKER_ID="${PRESET_WORKER_ID}" \
+        -e IMAGE_TAG="${IMAGE_TAG}" \
         -v /proc:/host/proc:ro \
         ${IMAGE}
     

docker/start.sh

@@ -135,6 +135,7 @@ if [ "$DEV_MODE" = true ]; then
     fi
 else
     # 生产模式：拉取 Docker Hub 镜像
+    # pull 后 up -d 会自动检测镜像变化并重建容器
     if [ "$WITH_FRONTEND" = true ]; then
         echo -e "${CYAN}[PULL]${NC} 拉取最新镜像..."
         ${COMPOSE_CMD} ${COMPOSE_ARGS} pull