From c8ef936a71deaf3b5838b2dec25f3d6b902218b0 Mon Sep 17 00:00:00 2001 From: dushixiang Date: Sat, 23 May 2026 15:32:46 +0800 Subject: [PATCH] feat: add agent expire alerts --- internal/app.go | 27 +++ internal/service/alert_service.go | 194 +++++++++++++++++++++ internal/service/notifier.go | 7 + web/src/admin/pages/AlertRecords/index.tsx | 7 +- 4 files changed, 232 insertions(+), 3 deletions(-) diff --git a/internal/app.go b/internal/app.go index 732986f..f6b865c 100644 --- a/internal/app.go +++ b/internal/app.go @@ -100,6 +100,8 @@ func setup(app *orz.App) error { // 启动流量重置检查任务(每小时检查一次) go startTrafficResetCheck(ctx, components, app.Logger()) + // 启动机器到期提醒检查任务(每小时检查一次) + go startAgentExpireCheck(ctx, components, app.Logger()) // 启动 DDNS 定时任务 go components.DDNSService.Run(ctx) @@ -449,6 +451,31 @@ func startTrafficResetCheck(ctx context.Context, components *AppComponents, logg } } +// startAgentExpireCheck 启动机器到期提醒定时任务 +func startAgentExpireCheck(ctx context.Context, components *AppComponents, logger *zap.Logger) { + logger.Info("启动机器到期提醒检查任务") + + ticker := time.NewTicker(1 * time.Hour) // 每小时检查一次 + defer ticker.Stop() + + check := func() { + if err := components.AlertService.CheckAgentExpireAlerts(ctx); err != nil { + logger.Error("机器到期提醒检查失败", zap.Error(err)) + } + } + + check() + for { + select { + case <-ctx.Done(): + logger.Info("机器到期提醒检查任务已停止") + return + case <-ticker.C: + check() + } + } +} + // JWTAuthMiddleware JWT 认证中间件（必须登录） func JWTAuthMiddleware(accountHandler *handler.AccountHandler) echo.MiddlewareFunc { return func(next echo.HandlerFunc) echo.HandlerFunc { diff --git a/internal/service/alert_service.go b/internal/service/alert_service.go index 00b5b5c..b1bc53f 100644 --- a/internal/service/alert_service.go +++ b/internal/service/alert_service.go @@ -13,6 +13,11 @@ import ( "gorm.io/gorm" ) +const ( + agentExpireAlertType = "agent_expire" + agentExpireReminderThreshold = 7 +) + // AlertService 告警服务 type AlertService struct { Service *orz.Service @@ -341,6 +346,195 @@ func (s *AlertService) sendAlertNotificationSync(ctx context.Context, record *mo return s.notifier.SendNotificationByConfigs(ctx, enabledChannels, record, agent, alertConfig.MaskIP) } +// CheckAgentExpireAlerts 检查机器到期提醒 +func (s *AlertService) CheckAgentExpireAlerts(ctx context.Context) error { + alertConfig, err := s.propertyService.GetAlertConfig(ctx) + if err != nil { + s.logger.Error("获取全局告警配置失败", zap.Error(err)) + return err + } + + if !alertConfig.Enabled { + return nil + } + + now := time.Now().UnixMilli() + agents, err := s.agentRepo.FindAll(ctx) + if err != nil { + return err + } + + for _, agent := range agents { + if err := s.checkAgentExpireAlert(ctx, &agent, now); err != nil { + s.logger.Error("检查机器过期提醒失败", + zap.String("agentId", agent.ID), + zap.Error(err), + ) + } + } + + return nil +} + +func (s *AlertService) checkAgentExpireAlert(ctx context.Context, agent *models.Agent, now int64) error { + stateKey := fmt.Sprintf("%s:global:%s:%s", agent.ID, agentExpireAlertType, agent.ID) + state, err := s.AlertStateRepo.GetAlertState(ctx, stateKey) + if err != nil { + state = &models.AlertState{ + ID: stateKey, + AgentID: agent.ID, + AlertType: agentExpireAlertType, + } + } + + daysLeft := agentExpireDaysLeft(agent.ExpireTime, now) + state.AgentID = agent.ID + state.AlertType = agentExpireAlertType + state.Threshold = float64(agentExpireReminderThreshold) + state.Duration = 0 + state.Value = daysLeft + state.LastCheckTime = now + + dayMs := int64(24 * time.Hour / time.Millisecond) + var shouldFire, shouldResolve bool + if agent.ExpireTime > 0 && agent.ExpireTime-now <= int64(agentExpireReminderThreshold)*dayMs { + if !state.IsFiring { + shouldFire = true + state.IsFiring = true + } + } else { + if state.IsFiring { + shouldResolve = true + } + } + + if err := s.AlertStateRepo.SaveAlertState(ctx, state); err != nil { + return fmt.Errorf("保存告警状态失败: %w", err) + } + + if shouldFire { + s.fireAgentExpireAlert(ctx, agent, state, now) + } + if shouldResolve { + s.resolveAgentExpireAlert(ctx, agent, state) + } + + return nil +} + +func (s *AlertService) fireAgentExpireAlert(ctx context.Context, agent *models.Agent, state *models.AlertState, now int64) { + s.logger.Info("触发机器过期提醒", + zap.String("agentId", agent.ID), + zap.String("agentName", agent.Name), + zap.Int64("expireTime", agent.ExpireTime), + zap.Float64("daysLeft", state.Value), + ) + + record := &models.AlertRecord{ + AgentID: agent.ID, + AgentName: agent.Name, + AlertType: agentExpireAlertType, + Message: s.buildAgentExpireMessage(agent, now), + Threshold: float64(agentExpireReminderThreshold), + ActualValue: state.Value, + Level: s.calculateAgentExpireLevel(state.Value), + Status: "firing", + FiredAt: now, + CreatedAt: now, + } + + if err := s.AlertRecordRepo.CreateAlertRecord(ctx, record); err != nil { + s.logger.Error("创建机器过期提醒记录失败", zap.Error(err)) + return + } + + state.LastRecordID = record.ID + if err := s.AlertStateRepo.SaveAlertState(ctx, state); err != nil { + s.logger.Error("保存告警状态失败", zap.Error(err)) + return + } + + go s.sendAlertNotification(record, agent) +} + +func (s *AlertService) resolveAgentExpireAlert(ctx context.Context, agent *models.Agent, state *models.AlertState) { + s.logger.Info("机器过期提醒恢复", + zap.String("agentId", agent.ID), + zap.String("agentName", agent.Name), + zap.Int64("expireTime", agent.ExpireTime), + zap.Float64("daysLeft", state.Value), + ) + + if state.LastRecordID > 0 { + existingRecord, err := s.AlertRecordRepo.GetAlertRecordByID(ctx, state.LastRecordID) + if err != nil { + s.logger.Error("获取机器过期提醒记录失败", zap.Error(err)) + } else if existingRecord != nil && existingRecord.Status == "firing" { + now := time.Now().UnixMilli() + existingRecord.Status = "resolved" + existingRecord.ResolvedValue = state.Value + existingRecord.ResolvedAt = now + existingRecord.UpdatedAt = now + + if err := s.AlertRecordRepo.UpdateAlertRecord(ctx, existingRecord); err != nil { + s.logger.Error("更新机器过期提醒记录失败", zap.Error(err)) + } else { + go s.sendAlertNotification(existingRecord, agent) + } + } + } + + state.IsFiring = false + state.LastRecordID = 0 + if err := s.AlertStateRepo.SaveAlertState(ctx, state); err != nil { + s.logger.Error("保存告警状态失败", zap.Error(err)) + } +} + +func (s *AlertService) buildAgentExpireMessage(agent *models.Agent, now int64) string { + agentName := agent.Name + if agentName == "" { + agentName = agent.Hostname + } + if agentName == "" { + agentName = agent.ID + } + + expireAt := time.UnixMilli(agent.ExpireTime).Format("2006-01-02 15:04:05") + if agent.ExpireTime <= now { + return fmt.Sprintf("机器 %s 已于 %s 过期", agentName, expireAt) + } + + remaining := agent.ExpireTime - now + dayMs := int64(24 * time.Hour / time.Millisecond) + if remaining < dayMs { + return fmt.Sprintf("机器 %s 将于 %s 过期，剩余不足1天", agentName, expireAt) + } + + days := remaining / dayMs + if remaining%dayMs != 0 { + days++ + } + return fmt.Sprintf("机器 %s 将于 %s 过期，剩余%d天", agentName, expireAt, days) +} + +func (s *AlertService) calculateAgentExpireLevel(daysLeft float64) string { + if daysLeft <= 0 { + return "critical" + } + if daysLeft <= 3 { + return "warning" + } + return "info" +} + +func agentExpireDaysLeft(expireTime int64, now int64) float64 { + if expireTime <= 0 { + return 0 + } + return float64(expireTime-now) / float64(24*time.Hour/time.Millisecond) +} + // CheckMonitorAlerts 检查监控相关告警（证书和服务下线） func (s *AlertService) CheckMonitorAlerts(ctx context.Context) error { // 获取全局告警配置 diff --git a/internal/service/notifier.go b/internal/service/notifier.go index 03785e3..cec8d26 100644 --- a/internal/service/notifier.go +++ b/internal/service/notifier.go @@ -106,6 +106,13 @@ var alertTypeMetadataMap = map[string]AlertTypeMetadata{ ShowThreshold: true, ShowActual: true, }, + "agent_expire": { + Name: "机器过期提醒", + ThresholdUnit: "天", + ValueUnit: "天", + ShowThreshold: true, + ShowActual: true, + }, "ssh_login": { Name: "SSH登录成功", ThresholdUnit: "", diff --git a/web/src/admin/pages/AlertRecords/index.tsx b/web/src/admin/pages/AlertRecords/index.tsx index f18cc9e..9336df4 100644 --- a/web/src/admin/pages/AlertRecords/index.tsx +++ b/web/src/admin/pages/AlertRecords/index.tsx @@ -39,6 +39,7 @@ const AlertRecordList = () => { cert: 'HTTPS证书', service: '服务下线', agent_offline: '探针离线', + agent_expire: '机器到期', }; // 告警级别映射 @@ -175,7 +176,7 @@ const AlertRecordList = () => { if (record.alertType === 'network') { return `${record.threshold.toFixed(2)} MB/s`; } - if (record.alertType === 'cert') { + if (record.alertType === 'cert' || record.alertType === 'agent_expire') { return `${record.threshold.toFixed(0)} 天`; } if (record.alertType === 'service' || record.alertType === 'agent_offline') { @@ -192,7 +193,7 @@ const AlertRecordList = () => { if (record.alertType === 'network') { return `${record.actualValue.toFixed(2)} MB/s`; } - if (record.alertType === 'cert') { + if (record.alertType === 'cert' || record.alertType === 'agent_expire') { return `${record.actualValue.toFixed(0)} 天`; } if (record.alertType === 'service' || record.alertType === 'agent_offline') { @@ -213,7 +214,7 @@ const AlertRecordList = () => { if (record.alertType === 'network') { return `${record.resolvedValue.toFixed(2)} MB/s`; } - if (record.alertType === 'cert') { + if (record.alertType === 'cert' || record.alertType === 'agent_expire') { return `${record.resolvedValue.toFixed(0)} 天`; } if (record.alertType === 'service' || record.alertType === 'agent_offline') {