功能：完善云管助手与GPU故障恢复核对

platform/internal/server/exclusive_gpu.go

@@ -291,8 +291,12 @@ func (s *Server) prepareGPUDesktop(c *gin.Context) {
 	var previous model.Task
 	err = tx.GetContext(ctx, &previous, `SELECT * FROM tasks WHERE kind='gpu-prepare' AND status IN ('queued','running','timeout') ORDER BY created_at DESC LIMIT 1`)
 	if err == nil {
-		if previous.Actor == "device:"+terminal && previous.TargetID == d.ID && previous.Status != "timeout" {
-			s.deviceJSON(c, 202, gin.H{"taskId": previous.ID, "status": previous.Status, "message": previous.Message})
+		if previous.Actor == "device:"+terminal && previous.TargetID == d.ID && (previous.Status != "timeout" || s.gpuRecoveryPending(previous)) {
+			status := previous.Status
+			if status == "timeout" {
+				status = "running"
+			}
+			s.deviceJSON(c, 202, gin.H{"taskId": previous.ID, "status": status, "message": previous.Message})
 			return
 		}
 		s.fail(c, 409, "gpu_switch_busy", "已有显卡切换任务或未确认结果，请稍后重试或联系管理员")
@@ -343,7 +347,11 @@ func (s *Server) gpuPrepareStatus(c *gin.Context) {
 		s.dbError(c, err)
 		return
 	}
-	s.deviceJSON(c, 200, gin.H{"taskId": task.ID, "status": task.Status, "message": task.Message})
+	status := task.Status
+	if s.gpuRecoveryPending(task) {
+		status = "running"
+	}
+	s.deviceJSON(c, 200, gin.H{"taskId": task.ID, "status": status, "message": task.Message})
 }
 
 func (s *Server) saveGPUStage(ctx context.Context, t model.Task, state gpuPrepareState, message string) error {
@@ -359,15 +367,11 @@ func (s *Server) saveGPUStage(ctx context.Context, t model.Task, state gpuPrepar
 }
 
 func sunshineReady(ctx context.Context, endpoint string) bool {
-	host := endpoint
-	if parsed, _, err := net.SplitHostPort(endpoint); err == nil {
-		host = parsed
-	}
-	// Platform endpoints are hostnames/IPs, never arbitrary URLs or paths.
-	if strings.ContainsAny(host, "/\\@?#") || host == "" {
-		return false
-	}
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://"+net.JoinHostPort(host, "47989")+"/serverinfo", nil)
+	return sunshineReadyAt(ctx, endpoint, "47989")
+}
+
+func sunshineHTTPProbe(ctx context.Context, host, port string) bool {
+	req, err := http.NewRequestWithContext(ctx, "GET", "http://"+net.JoinHostPort(host, port)+"/serverinfo", nil)
 	if err != nil {
 		return false
 	}
@@ -532,6 +536,7 @@ func (s *Server) executeGPUPrepare(parent context.Context, t model.Task) {
 				expected = "running"
 			}
 			if e != nil || power != expected {
+				uncertain = true
 				err = errors.New("final GPU ownership check failed")
 				break
 			}
@@ -546,9 +551,12 @@ func (s *Server) executeGPUPrepare(parent context.Context, t model.Task) {
 		log.Printf("GPU prepare task=%s phase=%s error=%v", t.ID, state.Phase, err)
 		status = "failed"
 		message = "桌面准备失败，请检查桌面授权、占用状态及 PVE"
-		if uncertain {
+		if uncertain || (state.Phase == "service-waiting" && s.knownGPURecovery(state)) {
 			status = "timeout"
-			message = "电源操作或服务状态尚未确认，已阻止再次切换；请管理员核对 PVE 原任务"
+			message = "电源操作尚未确认，平台正在自动核对原任务；不会重复发送电源命令"
+			if state.UPID == "" {
+				message = "未收到 PVE 原任务编号，请管理员核查；不会重复发送电源命令"
+			}
 		}
 	}
 	finalCtx, finish := context.WithTimeout(context.Background(), 5*time.Second)