observe-server/app/observe/service/event.go

package service

import (
	"fmt"
	aModel "go-admin/app/admin/models"
	"go-admin/app/observe/models"
	"go-admin/app/observe/service/dto"
	cDto "go-admin/common/dto"
	"go-admin/utils"
	"time"

	"github.com/pkg/errors"
	"gorm.io/gorm"
)

type Event struct {
	utils.OtService
}

func (e *Event) GetPage(c *dto.EventListReq, result *[]dto.EventListResp, count *int64) error {
	db := e.ChOrm.Table(models.TableNameEvent)
	if c.AppId > 0 {
		db.Where("AppId", c.AppId)
	}
	if c.AlertStatus == 2 || c.AlertStatus == 4 {
		db.Where("AlertStatus", c.AlertStatus)
	}
	// db.Scopes(cDto.MakeCondition(*c))
	if c.StartTime > 0 {
		db.Where("AppendTime>=?", c.StartTime)
	}
	if c.EndTime > 0 {
		db.Where("AppendTime<?", c.EndTime)
	}
	if c.ExceptionName != "" {
		db.Where("ExceptionName LIKE ?", fmt.Sprintf("%%%s%%", c.ExceptionName))
	}
	err := db.Debug().Scopes(cDto.Paginate(c.GetPageSize(), c.GetPageIndex())).Order("AppendTime DESC").Find(result).
		Limit(-1).Offset(-1).Count(count).Error
	if err != nil {
		if err == gorm.ErrRecordNotFound {
			return nil
		}
		return errors.Wrap(err, "查询事件信息列表失败")
	}
	return nil
}

func (e *Event) ExceptionNums(c *dto.EventExceptionNumsReq, result *[]dto.EventExceptionNumsResp, count *int64) error {
	db := e.Orm.Model(&aModel.OtFireEvents{}).Scopes(
		cDto.Paginate(c.GetPageSize(), c.GetPageIndex()),
	).Select("app_id", "app_alias", "app_name", "COUNT(1) AS exception_num").
		Where("is_know = 0 AND is_resolve = 0 AND is_ignore = 0")
	if c.StartTime > 0 {
		db.Where("created_at>=?", time.Unix(c.StartTime, 0).Format(time.DateTime))
	}
	if c.EndTime > 0 {
		db.Where("created_at<?", time.Unix(c.EndTime, 0).Format(time.DateTime))
	}
	appAliases := c.GetAppAliases()
	if len(appAliases) > 0 {
		db.Where("app_alias in ?", appAliases)
	} else { // 兼容旧逻辑， 后期去掉
		appIds := c.GetAppIds()
		if len(appIds) > 0 {
			db.Where("app_id IN ?", appIds)
		}
	}
	if err := db.Debug().Group("app_id, app_alias, app_name").Find(result).Limit(-1).Offset(-1).Count(count).Error; err != nil {
		return errors.Wrap(err, "执行查询应用异常数量失败")
	}
	return nil
}

func (e *Event) ExceptionNumByID(c *dto.EventExecptionNumGetByIDReq, count *int64) error {
	tbl := aModel.OtFireEvents{}.TableName()
	db := e.Orm.Table(fmt.Sprintf("%s t", tbl))
	db.Select([]string{
		"app_id",
		"COUNT(*) AS exception_num",
	})
	db.Where("app_id = ?", c.Id)
	db.Where("is_know = 0 AND is_resolve = 0 AND is_ignore = 0")
	if err := db.Count(count).Error; err != nil {
		return errors.Wrap(err, "执行查询业务异常数量失败")
	}
	return nil
}

func (e *Event) EventStatistic(c *dto.EventStatisticReq, result *[]dto.EventStatisticResp) error {
	db := e.ChOrm.Table(models.TableNameTracesError)
	db.Select([]string{
		"toTimeZone(toStartOfFiveMinutes(Timestamp), 'Asia/Shanghai') AS StartTime",
		"COUNT() AS Total",
	})
	db.Where("AppAlias = ?", c.AppAlias)
	db.Where("Timestamp >= ? AND Timestamp <= ?", c.StartTime, c.EndTime)
	db.Where("StatusCode='STATUS_CODE_ERROR' and StatusMessage!='' and StatusMessage != 'status code:0'")
	if c.Kind != "" && c.SubconditionValue != "" {
		switch c.Kind {
		case "app":
			db.Where("((ParentSpanId='' and SpanKind='SPAN_KIND_SERVER' and SpanAttributes['http.method'] != '' and SpanAttributes['rpc.system']='') OR SpanAttributes['db.system'] != '')") // 要求 只要http请求 和数据库 两种异常
		case "biz":
			// 改为错误分布， 与错误率*业务请求总数对上
			start := time.Unix(c.StartTime, 0).Truncate(time.Hour).Format(time.DateTime)
			end := time.Unix(c.EndTime, 0).Truncate(time.Hour).Add(time.Hour).Format(time.DateTime)
			return e.Orm.Model(&models.BizStatsHourly{}).
				Select("start_time, sum(error_num) as total").
				Where("biz_hash=?", c.BizHash).
				Where("start_time>=? and end_time<?", start, end).
				Group("start_time").Order("start_time").Find(result).Error
		case "interface":
		case "service":
			db.Where("ServiceName = ", c.SubconditionValue)
		default:

		}
	}
	// db.Where("StatusCode = 'STATUS_CODE_ERROR' AND StatusMessage != '' AND StatusMessage != 'status code:0'")
	db.Group("StartTime")
	if err := db.Order("StartTime Desc").Find(&result).Error; err != nil {
		return errors.Wrap(err, "计算范围时间内的异常失败")
	}
	return nil
}

func (e *Event) EventLists(c *dto.EventStatisticReq, result *[]dto.EventListsResp) error {
	res := make([]dto.EventListsResp, 0)
	db := e.ChOrm.Table(models.TableNameTracesError)
	db.Select(
		"ServiceName AS ServiceName",
		"any(if(SpanAttributes['db.system'] != '', '数据库异常', '业务异常')) AS ErrorType",
		"any(StatusMessage) as StatusMessage1",
		"COUNT() AS StatusMessageCount",
	)
	if c.AppAlias != "" {
		db.Where("AppAlias = ?", c.AppAlias)
	}
	if c.Kind != "" && c.SubconditionValue != "" {
		switch c.Kind {
		case "app":
		case "biz":
		case "interface":
			db.Where("")
		case "service":
			db.Where("ServiceName = ", c.SubconditionValue)
		default:

		}
	}
	db.Where("Timestamp >= ? AND Timestamp <= ?", c.StartTime, c.EndTime)
	// db.Where("StatusCode = 'STATUS_CODE_ERROR' AND StatusMessage != '' AND StatusMessage != 'status code:0'")
	db.Where("StatusCode='STATUS_CODE_ERROR' AND StatusMessage!='' and StatusMessage != 'status code:0'").
		Where("((ParentSpanId='' and  SpanKind='SPAN_KIND_SERVER' and SpanAttributes['http.method'] != '' and SpanAttributes['rpc.system']='') OR SpanAttributes['db.system'] != '')") // 要求 只要root span和数据库 两种异常
	db.Group("substring(StatusMessage, 1, 15)")
	db.Group("ServiceName")
	if err := db.Order("StatusMessageCount DESC;").Find(&res).Error; err != nil {
		return errors.Wrap(err, "统计范围时间内的具体异常失败")
	}
	svcNodes := []struct {
		ServiceName string
		Name        string
		Id          int64
	}{}
	if err := e.Orm.Model(&models.ServiceNode{}).Where("app_alias", c.AppAlias).Find(&svcNodes).Error; err != nil {
		return errors.Wrap(err, "获取服务列表失败")
	}
	svcMap := map[string]string{}
	svcId := map[string]int64{}
	for _, node := range svcNodes {
		svcMap[node.ServiceName] = node.Name
		svcId[node.ServiceName] = node.Id
	}
	for _, r := range res {
		r.ServiceNameCN = svcMap[r.ServiceName]
		r.ServiceId = svcId[r.ServiceName]
		*result = append(*result, r)
	}
	return nil
}