实现忽略告警和取消忽略告警接口及功能

2025-11-09 22:34:05 +08:00
parent 84fe20396b
commit b94aa6137c
11 changed files with 292 additions and 16 deletions
--- a/design/exceeding-threshold-alarm/index.md
+++ b/design/exceeding-threshold-alarm/index.md
@@ -136,3 +136,4 @@
 4. 实现告警发送任务
 5. 实现告警通知发送计划/全量采集计划改名
 6. 实现设备阈值检查任务
+7. 实现忽略告警和取消忽略告警接口及功能
--- a/internal/app/api/api.go
+++ b/internal/app/api/api.go
@@ -19,6 +19,7 @@ import (
 	"time"

 	_ "git.huangwc.com/pig/pig-farm-controller/docs" // 引入 swag 生成的 docs
+	"git.huangwc.com/pig/pig-farm-controller/internal/app/controller/alarm"
 	"git.huangwc.com/pig/pig-farm-controller/internal/app/controller/device"
 	"git.huangwc.com/pig/pig-farm-controller/internal/app/controller/health"
 	"git.huangwc.com/pig/pig-farm-controller/internal/app/controller/management"
@@ -53,6 +54,7 @@ type API struct {
 	pigBatchController  *management.PigBatchController       // 猪群控制器实例
 	monitorController   *monitor.Controller                  // 数据监控控制器实例
 	healthController    *health.Controller                   // 健康检查控制器实例
+	alarmController     *alarm.ThresholdAlarmController      // 阈值告警控制器
 	listenHandler       webhook.ListenHandler                // 设备上行事件监听器
 	analysisTaskManager *domain_plan.AnalysisPlanTaskManager // 计划触发器管理器实例
 }
@@ -69,6 +71,7 @@ func NewAPI(cfg config.ServerConfig,
 	planService service.PlanService,
 	userService service.UserService,
 	auditService service.AuditService,
+	alarmService service.ThresholdAlarmService,
 	tokenGenerator token.Generator,
 	listenHandler webhook.ListenHandler,
 ) *API {
@@ -106,6 +109,8 @@ func NewAPI(cfg config.ServerConfig,
 		monitorController: monitor.NewController(logs.AddCompName(baseCtx, "MonitorController"), monitorService),
 		// 在 NewAPI 中初始化健康检查控制器
 		healthController: health.NewController(logs.AddCompName(baseCtx, "HealthController")),
+		// 在 NewAPI 中初始化阈
+		alarmController: alarm.NewThresholdAlarmController(logs.AddCompName(baseCtx, "ThresholdAlarmController"), alarmService),
 	}

 	api.setupRoutes() // 设置所有路由
--- a/internal/app/api/router.go
+++ b/internal/app/api/router.go
@@ -187,6 +187,17 @@ func (a *API) setupRoutes() {
 			monitorGroup.GET("/notifications", a.monitorController.ListNotifications)
 		}
 		logger.Debug("数据监控相关接口注册成功 (需要认证和审计)")
+
+		// 告警相关路由组
+		alarmGroup := authGroup.Group("/alarm")
+		{
+			thresholdGroup := alarmGroup.Group("/thresholds")
+			{
+				thresholdGroup.POST("/:id/snooze", a.alarmController.SnoozeThresholdAlarm)              // 忽略阈值告警
+				thresholdGroup.POST("/:id/cancel-snooze", a.alarmController.CancelSnoozeThresholdAlarm) // 取消忽略阈值告警
+			}
+		}
+		logger.Debug("告警相关接口注册成功 (需要认证和审计)")
 	}

 	logger.Debug("所有接口注册成功")
--- a/internal/app/controller/alarm/threshold_alarm_controller.go
+++ b/internal/app/controller/alarm/threshold_alarm_controller.go
@@ -0,0 +1,109 @@
+package alarm
+
+import (
+	"context"
+	"errors"
+	"strconv"
+
+	"git.huangwc.com/pig/pig-farm-controller/internal/app/controller"
+	"git.huangwc.com/pig/pig-farm-controller/internal/app/dto"
+	"git.huangwc.com/pig/pig-farm-controller/internal/app/service"
+	"git.huangwc.com/pig/pig-farm-controller/internal/infra/logs"
+
+	"github.com/labstack/echo/v4"
+	"gorm.io/gorm"
+)
+
+// ThresholdAlarmController 阈值告警控制器，封装了所有与阈值告警配置相关的业务逻辑
+type ThresholdAlarmController struct {
+	ctx                   context.Context
+	thresholdAlarmService service.ThresholdAlarmService
+}
+
+// NewThresholdAlarmController 创建一个新的阈值告警控制器实例
+func NewThresholdAlarmController(
+	ctx context.Context,
+	thresholdAlarmService service.ThresholdAlarmService,
+) *ThresholdAlarmController {
+	return &ThresholdAlarmController{
+		ctx:                   ctx,
+		thresholdAlarmService: thresholdAlarmService,
+	}
+}
+
+// SnoozeThresholdAlarm godoc
+// @Summary      忽略阈值告警
+// @Description  根据告警ID忽略一个活跃的阈值告警，或更新其忽略时间
+// @Tags         告警管理
+// @Security     BearerAuth
+// @Accept       json
+// @Produce      json
+// @Param        id path string true "告警ID"
+// @Param        request body dto.SnoozeAlarmRequest true "忽略告警请求体"
+// @Success      200 {object} controller.Response "成功忽略告警"
+// @Router       /api/v1/alarm/threshold/{id}/snooze [post]
+func (t *ThresholdAlarmController) SnoozeThresholdAlarm(ctx echo.Context) error {
+	reqCtx, logger := logs.Trace(ctx.Request().Context(), t.ctx, "SnoozeThresholdAlarm")
+
+	const actionType = "忽略阈值告警"
+	alarmIDStr := ctx.Param("id")
+
+	alarmID, err := strconv.ParseUint(alarmIDStr, 10, 64)
+	if err != nil {
+		logger.Errorf("%s: 无效的告警ID: %s", actionType, alarmIDStr)
+		return controller.SendErrorWithAudit(ctx, controller.CodeBadRequest, "无效的告警ID: "+alarmIDStr, actionType, "无效的告警ID", alarmIDStr)
+	}
+
+	var req dto.SnoozeAlarmRequest
+	if err := ctx.Bind(&req); err != nil {
+		logger.Errorf("%s: 参数绑定失败: %v", actionType, err)
+		return controller.SendErrorWithAudit(ctx, controller.CodeBadRequest, "无效的请求体: "+err.Error(), actionType, "请求体绑定失败", req)
+	}
+
+	if err := t.thresholdAlarmService.SnoozeThresholdAlarm(reqCtx, uint(alarmID), req.DurationMinutes); err != nil {
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			logger.Warnf("%s: 告警不存在, ID: %d", actionType, alarmID)
+			return controller.SendErrorWithAudit(ctx, controller.CodeNotFound, "告警未找到", actionType, "告警不存在", alarmID)
+		}
+		logger.Errorf("%s: 服务层忽略告警失败: %v, ID: %d", actionType, err, alarmID)
+		return controller.SendErrorWithAudit(ctx, controller.CodeInternalError, "忽略告警失败: "+err.Error(), actionType, "服务层忽略告警失败", alarmID)
+	}
+
+	logger.Infof("%s: 告警已成功忽略, ID: %d", actionType, alarmID)
+	return controller.SendSuccessWithAudit(ctx, controller.CodeSuccess, "告警已成功忽略", nil, actionType, "告警已成功忽略", alarmID)
+}
+
+// CancelSnoozeThresholdAlarm godoc
+// @Summary      取消忽略阈值告警
+// @Description  根据告警ID取消对一个阈值告警的忽略状态
+// @Tags         告警管理
+// @Security     BearerAuth
+// @Accept       json
+// @Produce      json
+// @Param        id path string true "告警ID"
+// @Success      200 {object} controller.Response "成功取消忽略告警"
+// @Router       /api/v1/alarm/threshold/{id}/cancel-snooze [post]
+func (t *ThresholdAlarmController) CancelSnoozeThresholdAlarm(ctx echo.Context) error {
+	reqCtx, logger := logs.Trace(ctx.Request().Context(), t.ctx, "CancelSnoozeThresholdAlarm")
+
+	const actionType = "取消忽略阈值告警"
+	alarmIDStr := ctx.Param("id")
+
+	alarmID, err := strconv.ParseUint(alarmIDStr, 10, 64)
+	if err != nil {
+		logger.Errorf("%s: 无效的告警ID: %s", actionType, alarmIDStr)
+		return controller.SendErrorWithAudit(ctx, controller.CodeBadRequest, "无效的告警ID: "+alarmIDStr, actionType, "无效的告警ID", alarmIDStr)
+	}
+
+	if err := t.thresholdAlarmService.CancelSnoozeThresholdAlarm(reqCtx, uint(alarmID)); err != nil {
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			logger.Warnf("%s: 告警不存在, ID: %d", actionType, alarmID)
+			return controller.SendErrorWithAudit(ctx, controller.CodeNotFound, "告警未找到", actionType, "告警不存在", alarmID)
+		}
+		logger.Errorf("%s: 服务层取消忽略告警失败: %v, ID: %d", actionType, err, alarmID)
+		return controller.SendErrorWithAudit(ctx, controller.CodeInternalError, "取消忽略告警失败: "+err.Error(), actionType, "服务层取消忽略告警失败", alarmID)
+	}
+
+	logger.Infof("%s: 告警忽略状态已成功取消, ID: %d", actionType, alarmID)
+	return controller.SendSuccessWithAudit(ctx, controller.CodeSuccess, "告警忽略状态已成功取消", nil, actionType, "告警忽略状态已成功取消", alarmID)
+}
--- a/internal/app/dto/alarm_dto.go
+++ b/internal/app/dto/alarm_dto.go
@@ -0,0 +1,6 @@
+package dto
+
+// SnoozeAlarmRequest 定义了忽略告警的请求体
+type SnoozeAlarmRequest struct {
+	DurationMinutes uint `json:"duration_minutes" validate:"required,min=1"` // 忽略时长，单位分钟
+}
--- a/internal/app/service/threshold_alarm_service.go
+++ b/internal/app/service/threshold_alarm_service.go
@@ -0,0 +1,44 @@
+package service
+
+import (
+	"context"
+	"time"
+
+	domainAlarm "git.huangwc.com/pig/pig-farm-controller/internal/domain/alarm" // 引入领域层的 AlarmService
+	"git.huangwc.com/pig/pig-farm-controller/internal/infra/logs"
+)
+
+// ThresholdAlarmService 定义了阈值告警配置服务的接口。
+// 该服务负责管理阈值告警任务的配置，并将其与计划进行联动。
+type ThresholdAlarmService interface {
+	// SnoozeThresholdAlarm 忽略一个阈值告警，或更新其忽略时间。
+	SnoozeThresholdAlarm(ctx context.Context, alarmID uint, durationMinutes uint) error
+	// CancelSnoozeThresholdAlarm 取消对一个阈值告警的忽略状态。
+	CancelSnoozeThresholdAlarm(ctx context.Context, alarmID uint) error
+}
+
+// thresholdAlarmService 是 ThresholdAlarmService 接口的具体实现。
+type thresholdAlarmService struct {
+	ctx          context.Context
+	alarmService domainAlarm.AlarmService // 注入领域层的 AlarmService
+}
+
+// NewThresholdAlarmService 创建一个新的 ThresholdAlarmService 实例。
+func NewThresholdAlarmService(ctx context.Context, alarmService domainAlarm.AlarmService) ThresholdAlarmService {
+	return &thresholdAlarmService{
+		ctx:          ctx,
+		alarmService: alarmService,
+	}
+}
+
+// SnoozeThresholdAlarm 实现了忽略阈值告警的逻辑。
+func (s *thresholdAlarmService) SnoozeThresholdAlarm(ctx context.Context, alarmID uint, durationMinutes uint) error {
+	serviceCtx := logs.AddFuncName(ctx, s.ctx, "SnoozeThresholdAlarm")
+	return s.alarmService.SnoozeAlarm(serviceCtx, alarmID, time.Duration(durationMinutes)*time.Minute)
+}
+
+// CancelSnoozeThresholdAlarm 实现了取消忽略阈值告警的逻辑。
+func (s *thresholdAlarmService) CancelSnoozeThresholdAlarm(ctx context.Context, alarmID uint) error {
+	serviceCtx := logs.AddFuncName(ctx, s.ctx, "CancelSnoozeThresholdAlarm")
+	return s.alarmService.CancelAlarmSnooze(serviceCtx, alarmID)
+}
--- a/internal/core/application.go
+++ b/internal/core/application.go
@@ -61,6 +61,7 @@ func NewApplication(configPath string) (*Application, error) {
 		appServices.planService,
 		appServices.userService,
 		appServices.auditService,
+		appServices.thresholdAlarmService,
 		infra.tokenGenerator,
 		infra.lora.listenHandler,
 	)
--- a/internal/core/component_initializers.go
+++ b/internal/core/component_initializers.go
@@ -7,6 +7,7 @@ import (

 	"git.huangwc.com/pig/pig-farm-controller/internal/app/service"
 	"git.huangwc.com/pig/pig-farm-controller/internal/app/webhook"
+	"git.huangwc.com/pig/pig-farm-controller/internal/domain/alarm"
 	"git.huangwc.com/pig/pig-farm-controller/internal/domain/device"
 	domain_notify "git.huangwc.com/pig/pig-farm-controller/internal/domain/notify"
 	"git.huangwc.com/pig/pig-farm-controller/internal/domain/pig"
@@ -126,6 +127,7 @@ type DomainServices struct {
 	analysisPlanTaskManager plan.AnalysisPlanTaskManager
 	planService             plan.Service
 	notifyService           domain_notify.Service
+	alarmService            alarm.AlarmService
 }

 // initDomainServices 初始化所有的领域服务。
@@ -196,6 +198,13 @@ func initDomainServices(ctx context.Context, cfg *config.Config, infra *Infrastr
 		taskFactory,
 	)

+	// 告警服务
+	alarmService := alarm.NewAlarmService(
+		logs.AddCompName(baseCtx, "AlarmService"),
+		infra.repos.alarmRepo,
+		infra.repos.unitOfWork,
+	)
+
 	return &DomainServices{
 		pigPenTransferManager:   pigPenTransferManager,
 		pigTradeManager:         pigTradeManager,
@@ -207,18 +216,20 @@ func initDomainServices(ctx context.Context, cfg *config.Config, infra *Infrastr
 		planExecutionManager:    planExecutionManager,
 		planService:             planService,
 		notifyService:           notifyService,
+		alarmService:            alarmService,
 	}, nil
 }

 // AppServices 聚合了所有的应用服务实例。
 type AppServices struct {
-	pigFarmService  service.PigFarmService
-	pigBatchService service.PigBatchService
-	monitorService  service.MonitorService
-	deviceService   service.DeviceService
-	planService     service.PlanService
-	userService     service.UserService
-	auditService    service.AuditService
+	pigFarmService        service.PigFarmService
+	pigBatchService       service.PigBatchService
+	monitorService        service.MonitorService
+	deviceService         service.DeviceService
+	planService           service.PlanService
+	userService           service.UserService
+	auditService          service.AuditService
+	thresholdAlarmService service.ThresholdAlarmService
 }

 // initAppServices 初始化所有的应用服务。
@@ -254,14 +265,21 @@ func initAppServices(ctx context.Context, infra *Infrastructure, domainServices
 	planService := service.NewPlanService(logs.AddCompName(baseCtx, "AppPlanService"), domainServices.planService)
 	userService := service.NewUserService(logs.AddCompName(baseCtx, "UserService"), infra.repos.userRepo, infra.tokenGenerator, domainServices.notifyService)

+	// 初始化阈值告警服务
+	thresholdAlarmService := service.NewThresholdAlarmService(
+		logs.AddCompName(baseCtx, "ThresholdAlarmService"),
+		domainServices.alarmService,
+	)
+
 	return &AppServices{
-		pigFarmService:  pigFarmService,
-		pigBatchService: pigBatchService,
-		monitorService:  monitorService,
-		deviceService:   deviceService,
-		auditService:    auditService,
-		planService:     planService,
-		userService:     userService,
+		pigFarmService:        pigFarmService,
+		pigBatchService:       pigBatchService,
+		monitorService:        monitorService,
+		deviceService:         deviceService,
+		auditService:          auditService,
+		planService:           planService,
+		userService:           userService,
+		thresholdAlarmService: thresholdAlarmService,
 	}
 }

--- a/internal/domain/alarm/alarm_service.go
+++ b/internal/domain/alarm/alarm_service.go
@@ -3,6 +3,7 @@ package alarm
 import (
 	"context"
 	"errors"
+	"fmt"
 	"time"

 	"git.huangwc.com/pig/pig-farm-controller/internal/infra/logs"
@@ -21,6 +22,14 @@ type AlarmService interface {
 	// CloseAlarm 关闭一个活跃告警，将其归档到历史记录。
 	// 如果指定的告警当前不活跃，则不执行任何操作并返回 nil。
 	CloseAlarm(ctx context.Context, sourceType models.AlarmSourceType, sourceID uint, alarmCode models.AlarmCode, resolveMethod string, resolvedBy *uint) error
+
+	// SnoozeAlarm 忽略一个活跃告警，或更新其忽略时间。
+	// 如果告警不存在，将返回错误。
+	SnoozeAlarm(ctx context.Context, alarmID uint, duration time.Duration) error
+
+	// CancelAlarmSnooze 取消对一个告警的忽略状态。
+	// 如果告警不存在，或本就未被忽略，不执行任何操作并返回 nil。
+	CancelAlarmSnooze(ctx context.Context, alarmID uint) error
 }

 // alarmService 是 AlarmService 接口的具体实现。
@@ -122,3 +131,45 @@ func (s *alarmService) CloseAlarm(ctx context.Context, sourceType models.AlarmSo
 		return nil
 	})
 }
+
+// SnoozeAlarm 忽略一个活跃告警，或更新其忽略时间。
+func (s *alarmService) SnoozeAlarm(ctx context.Context, alarmID uint, duration time.Duration) error {
+	serviceCtx, logger := logs.Trace(ctx, s.ctx, "SnoozeAlarm")
+
+	if duration <= 0 {
+		return errors.New("忽略时长必须为正数")
+	}
+
+	ignoredUntil := time.Now().Add(duration)
+	err := s.alarmRepo.UpdateIgnoreStatus(serviceCtx, alarmID, true, &ignoredUntil)
+	if err != nil {
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			logger.Warnf("尝试忽略一个不存在的告警: %d", alarmID)
+			return fmt.Errorf("告警 %d 不存在", alarmID)
+		}
+		logger.Errorf("更新告警 %d 的忽略状态失败: %v", alarmID, err)
+		return err
+	}
+
+	logger.Infof("告警 %d 已被成功忽略，持续时间: %v", alarmID, duration)
+	return nil
+}
+
+// CancelAlarmSnooze 取消对一个告警的忽略状态。
+func (s *alarmService) CancelAlarmSnooze(ctx context.Context, alarmID uint) error {
+	serviceCtx, logger := logs.Trace(ctx, s.ctx, "CancelAlarmSnooze")
+
+	err := s.alarmRepo.UpdateIgnoreStatus(serviceCtx, alarmID, false, nil)
+	if err != nil {
+		// 如果告警本就不存在，这不是一个需要上报的错误
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			logger.Infof("尝试取消忽略一个不存在的告警: %d，无需操作", alarmID)
+			return nil
+		}
+		logger.Errorf("取消告警 %d 的忽略状态失败: %v", alarmID, err)
+		return err
+	}
+
+	logger.Infof("告警 %d 的忽略状态已被成功取消。", alarmID)
+	return nil
+}
--- a/internal/infra/repository/alarm_repository.go
+++ b/internal/infra/repository/alarm_repository.go
@@ -40,6 +40,9 @@ type AlarmRepository interface {
 	// DeleteActiveAlarmTx 在指定事务中根据主键 ID 删除一个活跃告警
 	DeleteActiveAlarmTx(ctx context.Context, tx *gorm.DB, id uint) error

+	// UpdateIgnoreStatus 更新指定告警的忽略状态
+	UpdateIgnoreStatus(ctx context.Context, id uint, isIgnored bool, ignoredUntil *time.Time) error
+
 	// ListActiveAlarms 支持分页和过滤的活跃告警列表查询。
 	// 返回活跃告警列表、总记录数和错误。
 	ListActiveAlarms(ctx context.Context, opts ActiveAlarmListOptions, page, pageSize int) ([]models.ActiveAlarm, int64, error)
@@ -80,7 +83,7 @@ func (r *gormAlarmRepository) CreateActiveAlarm(ctx context.Context, alarm *mode
 	return r.db.WithContext(repoCtx).Create(alarm).Error
 }

-// IsAlarmActive 检查具有相同来源和告警代码的告警当前是否处于活跃状态
+// IsAlarmActiveInUse 检查具有相同来源和告警代码的告警当前是否处于活跃表中
 func (r *gormAlarmRepository) IsAlarmActiveInUse(ctx context.Context, sourceType models.AlarmSourceType, sourceID uint, alarmCode models.AlarmCode) (bool, error) {
 	repoCtx := logs.AddFuncName(ctx, r.ctx, "IsAlarmActiveInUse")
 	var count int64
@@ -116,6 +119,30 @@ func (r *gormAlarmRepository) DeleteActiveAlarmTx(ctx context.Context, tx *gorm.
 	return tx.WithContext(repoCtx).Unscoped().Delete(&models.ActiveAlarm{}, id).Error
 }

+// UpdateIgnoreStatus 更新指定告警的忽略状态
+func (r *gormAlarmRepository) UpdateIgnoreStatus(ctx context.Context, id uint, isIgnored bool, ignoredUntil *time.Time) error {
+	repoCtx := logs.AddFuncName(ctx, r.ctx, "UpdateIgnoreStatus")
+	updates := map[string]interface{}{
+		"is_ignored":    isIgnored,
+		"ignored_until": ignoredUntil,
+	}
+
+	result := r.db.WithContext(repoCtx).
+		Model(&models.ActiveAlarm{}).
+		Where("id = ?", id).
+		Updates(updates)
+
+	if result.Error != nil {
+		return result.Error
+	}
+
+	if result.RowsAffected == 0 {
+		return gorm.ErrRecordNotFound
+	}
+
+	return nil
+}
+
 // ListActiveAlarms 实现了分页和过滤查询活跃告警记录的功能
 func (r *gormAlarmRepository) ListActiveAlarms(ctx context.Context, opts ActiveAlarmListOptions, page, pageSize int) ([]models.ActiveAlarm, int64, error) {
 	repoCtx := logs.AddFuncName(ctx, r.ctx, "ListActiveAlarms")
--- a/project_structure.txt
+++ b/project_structure.txt
@@ -40,6 +40,7 @@ go.mod
 go.sum
 internal/app/api/api.go
 internal/app/api/router.go
+internal/app/controller/alarm/threshold_alarm_controller.go
 internal/app/controller/auth_utils.go
 internal/app/controller/device/device_controller.go
 internal/app/controller/health/health_controller.go
@@ -53,6 +54,7 @@ internal/app/controller/monitor/monitor_controller.go
 internal/app/controller/plan/plan_controller.go
 internal/app/controller/response.go
 internal/app/controller/user/user_controller.go
+internal/app/dto/alarm_dto.go
 internal/app/dto/device_converter.go
 internal/app/dto/device_dto.go
 internal/app/dto/monitor_converter.go
@@ -73,6 +75,7 @@ internal/app/service/pig_batch_service.go
 internal/app/service/pig_farm_service.go
 internal/app/service/pig_service.go
 internal/app/service/plan_service.go
+internal/app/service/threshold_alarm_service.go
 internal/app/service/user_service.go
 internal/app/webhook/chirp_stack.go
 internal/app/webhook/chirp_stack_types.go