功能：完善云管助手与GPU故障恢复核对

platform/internal/server/admin.go

@@ -43,7 +43,7 @@ func (s *Server) settings(c *gin.Context) {
 	if pve != nil {
 		provider = "proxmox"
 	}
-	c.JSON(200, gin.H{"mode": s.Config.Mode, "computeProvider": provider, "pveConfigured": pve != nil, "terminalTransport": "https-poll", "capabilities": []string{"diagnostics-center", "admin-mfa", "administrator-sessions", "task-history", "audit-query-export", "terminal-metric-alerts", "terminal-system-metrics", "gpu-inventory", "desktop-batches", "desktop-snapshots", "pve-connection", "desktop-hardware", "login-rate-limit", "csv-imports", "session-history", "monitor-alerts", "administrators", "password", "auth", "resources", "desktops", "templates", "pools", "clone", "desktop-connection", "terminals", "terminal-lifecycle", "terminal-policies", "users", "batch-bindings", "bindings", "tasks", "audit"}})
+	c.JSON(200, gin.H{"mode": s.Config.Mode, "computeProvider": provider, "pveConfigured": pve != nil, "terminalTransport": "https-poll", "capabilities": []string{"common-assistant", "diagnostics-center", "admin-mfa", "administrator-sessions", "task-history", "audit-query-export", "terminal-metric-alerts", "terminal-system-metrics", "gpu-inventory", "gpu-health-agent", "desktop-batches", "desktop-snapshots", "pve-connection", "desktop-hardware", "login-rate-limit", "csv-imports", "session-history", "monitor-alerts", "administrators", "password", "auth", "resources", "desktops", "templates", "pools", "clone", "desktop-connection", "terminals", "terminal-lifecycle", "terminal-policies", "users", "batch-bindings", "bindings", "tasks", "audit"}})
 }
 func (s *Server) resources(c *gin.Context) {
 	items := []model.Resource{}
@@ -514,6 +514,7 @@ func (s *Server) RunWorker(ctx context.Context) {
 	workerCtx, stop := context.WithCancel(ctx)
 	provisionDone := make(chan struct{}, 1)
 	provisionBusy := false
+	var nextGPUCheck time.Time
 	defer func() {
 		stop()
 		if provisionBusy {
@@ -532,7 +533,8 @@ func (s *Server) RunWorker(ctx context.Context) {
 			if err = conn.PingContext(ctx); err != nil {
 				return
 			}
-			// Unknown outcomes are never automatically retried after a worker restart.
+			// Unknown commands are never replayed. Known GPU tasks are observed
+			// separately below so a transient PVE read failure can recover.
 			var expired []model.Task
 			if s.DB.SelectContext(ctx, &expired, "SELECT * FROM tasks WHERE status IN ('queued','running') AND ((kind NOT IN ('provision','hardware','snapshot','batch-desktop','gpu-prepare') AND created_at<now()-interval '2 minutes') OR (kind='gpu-prepare' AND created_at<now()-interval '5 minutes') OR (kind='batch-desktop' AND ((status='queued' AND created_at<now()-interval '2 hours') OR (status='running' AND updated_at<now()-interval '2 minutes'))) OR (kind IN ('provision','hardware','snapshot') AND updated_at<now()-interval '35 minutes'))") == nil {
 				for _, t := range expired {
@@ -563,6 +565,14 @@ func (s *Server) RunWorker(ctx context.Context) {
 					_ = s.complete(ctx, t, "timeout", "结果未确认，请核对设备或 PVE 实际状态；不会自动重放")
 				}
 			}
+			if !provisionBusy && !time.Now().Before(nextGPUCheck) {
+				nextGPUCheck = time.Now().Add(15 * time.Second)
+				if recovery, readErr := s.nextGPURecovery(ctx); readErr == nil {
+					provisionBusy = true
+					go func(task model.Task) { s.recoverGPUPrepare(workerCtx, task); provisionDone <- struct{}{} }(recovery)
+					continue
+				}
+			}
 			var t model.Task
 			err := s.DB.GetContext(ctx, &t, `UPDATE tasks SET status='running',updated_at=now(),message='正在执行，等待最终回执' WHERE id=(SELECT id FROM tasks WHERE ((status='queued' AND kind!='terminal') OR (kind='provision' AND status='running' AND provider_ref!='') OR (kind IN ('hardware','snapshot','batch-desktop') AND status='running')) AND ($1 OR kind NOT IN ('provision','hardware','snapshot','batch-desktop','gpu-prepare')) ORDER BY created_at FOR UPDATE SKIP LOCKED LIMIT 1) RETURNING *`, !provisionBusy)
 			if err != nil {