功能：补充固定场景助手与GPU健康诊断能力

2026-11-12 17:33:07 +08:00
1 parent 9f05aa20ea
commit b6026f9759
@@ -35,6 +35,129 @@ type gpuPrepareState struct {
 	UPID     string `json:"upid,omitempty"`
 }

+type gpuExclusiveMember struct {
+	DesktopID      string   `json:"desktopId"`
+	Name           string   `json:"name"`
+	VMID           int      `json:"vmid"`
+	PowerState     string   `json:"powerState"`
+	PowerObserved  bool     `json:"powerObserved"`
+	ActiveSessions int      `json:"activeSessions"`
+	ActiveTasks    int      `json:"activeTasks"`
+	Conflicts      []string `json:"conflicts"`
+	Eligible       bool     `json:"eligible"`
+}
+type gpuExclusiveStatus struct {
+	Enabled         bool                 `json:"enabled"`
+	Node            string               `json:"node"`
+	VMIDs           []int                `json:"vmids"`
+	ObservedAt      time.Time            `json:"observedAt"`
+	LockAvailable   bool                 `json:"lockAvailable"`
+	GroupComplete   bool                 `json:"groupComplete"`
+	Members         []gpuExclusiveMember `json:"members"`
+	Conflicts       []string             `json:"conflicts"`
+	AllocationReady bool                 `json:"allocationReady"`
+	ReleaseReady    bool                 `json:"releaseReady"`
+	ReleasePolicy   string               `json:"releasePolicy"`
+	ActiveTasks     []model.Task         `json:"activeTasks"`
+}
+
+// exclusiveGPUStatus is a read-only preflight view. It never changes VM power
+// or task state; it is used by the advanced console before allocation/release.
+func (s *Server) exclusiveGPUStatus(c *gin.Context) {
+	node := strings.TrimSpace(c.Query("node"))
+	status := gpuExclusiveStatus{ObservedAt: time.Now().UTC(), Node: s.Config.ExclusiveGPUNode, VMIDs: append([]int{}, s.Config.ExclusiveGPUVMIDs...), Members: []gpuExclusiveMember{}, Conflicts: []string{}, ActiveTasks: []model.Task{}, ReleasePolicy: "仅在无活动会话、无未完成任务且锁可用时可释放；不会自动关机或重放操作"}
+	if node != "" && (len(node) > 100 || strings.ContainsAny(node, "/\\?#")) {
+		s.fail(c, 400, "invalid_node", "节点名称无效")
+		return
+	}
+	if status.Node == "" || len(status.VMIDs) == 0 {
+		c.JSON(200, status)
+		return
+	}
+	if node != "" && node != status.Node {
+		status.Conflicts = []string{"当前节点未配置独占显卡组"}
+		c.JSON(200, status)
+		return
+	}
+	status.Enabled = true
+	tx, err := s.DB.BeginTxx(c.Request.Context(), &sql.TxOptions{ReadOnly: true})
+	if err != nil {
+		s.dbError(c, err)
+		return
+	}
+	defer tx.Rollback()
+	locked, err := tryGPULock(c.Request.Context(), tx)
+	if err != nil {
+		s.dbError(c, err)
+		return
+	}
+	status.LockAvailable = locked
+	if !locked {
+		status.Conflicts = append(status.Conflicts, "独占显卡锁当前被占用")
+	}
+	var members []model.Desktop
+	members, err = s.gpuMembers(c.Request.Context(), tx)
+	if err != nil {
+		status.Conflicts = append(status.Conflicts, "独占显卡组成员不完整")
+	} else {
+		status.GroupComplete = true
+	}
+	var tasks []model.Task
+	if err = tx.SelectContext(c.Request.Context(), &tasks, `SELECT * FROM tasks WHERE kind='gpu-prepare' AND status IN ('queued','running','timeout') ORDER BY created_at DESC`); err == nil {
+		status.ActiveTasks = tasks
+		if len(tasks) > 0 {
+			status.Conflicts = append(status.Conflicts, "存在未完成或结果未核对的显卡切换任务")
+		}
+	}
+	pve := s.currentPVE()
+	for _, d := range members {
+		m := gpuExclusiveMember{DesktopID: d.ID, Name: d.Name, VMID: d.VMID, PowerState: d.PowerState, Conflicts: []string{}}
+		var n int
+		if err = tx.GetContext(c.Request.Context(), &n, `SELECT count(*) FROM terminals WHERE session_desktop_id=$1 AND last_seen>now()-interval '90 seconds'`, d.ID); err == nil {
+			m.ActiveSessions = n
+		}
+		if err = tx.GetContext(c.Request.Context(), &n, `SELECT count(*) FROM tasks WHERE target_id=$1 AND status IN ('queued','running','timeout') AND kind!='terminal'`, d.ID); err == nil {
+			m.ActiveTasks = n
+		}
+		if m.ActiveSessions > 0 {
+			m.Conflicts = append(m.Conflicts, "存在活跃终端会话")
+		}
+		if m.ActiveTasks > 0 {
+			m.Conflicts = append(m.Conflicts, "存在未完成管理任务")
+		}
+		if pve != nil {
+			power, powerErr := pve.Power(c.Request.Context(), d.Node, d.VMID)
+			if powerErr == nil {
+				m.PowerState, m.PowerObserved = power, true
+			} else {
+				m.PowerState = "unknown"
+				m.Conflicts = append(m.Conflicts, "无法读取 PVE 实际电源状态")
+			}
+		} else {
+			m.PowerState = "unknown"
+			m.Conflicts = append(m.Conflicts, "尚未配置 PVE")
+		}
+		if m.PowerObserved && m.PowerState != "stopped" {
+			m.Conflicts = append(m.Conflicts, "虚拟机当前仍在运行")
+		}
+		m.Eligible = len(m.Conflicts) == 0 && m.PowerState == "stopped"
+		status.Members = append(status.Members, m)
+	}
+	if !status.GroupComplete {
+		status.Conflicts = append(status.Conflicts, "数据库中的独占显卡组成员数量与配置不一致")
+	}
+	allEligible := len(status.Members) == len(status.VMIDs)
+	for _, member := range status.Members {
+		if !member.Eligible {
+			allEligible = false
+			break
+		}
+	}
+	status.AllocationReady = status.Enabled && status.GroupComplete && status.LockAvailable && len(status.ActiveTasks) == 0 && len(status.Conflicts) == 0 && allEligible
+	status.ReleaseReady = status.Enabled && status.GroupComplete && status.LockAvailable && len(status.ActiveTasks) == 0 && allEligible
+	c.JSON(200, status)
+}
+
 func (s *Server) exclusiveGPU(d model.Desktop) bool {
 	if d.Provider != "proxmox" || d.Node != s.Config.ExclusiveGPUNode {
 		return false