feat: alert rule api fill users and groups (#723)

cache/user_group.go

+package cache
+
+import (
+	"sync"
+
+	"github.com/didi/nightingale/v5/models"
+)
+
+type UserGroupMap struct {
+	sync.RWMutex
+	Data map[int64]*models.UserGroup
+}
+
+var UserGroupCache = &UserGroupMap{Data: make(map[int64]*models.UserGroup)}
+
+func (s *UserGroupMap) GetBy(id int64) *models.UserGroup {
+	s.RLock()
+	defer s.RUnlock()
+
+	return s.Data[id]
+}
+
+func (s *UserGroupMap) GetByIds(ids []int64) []*models.UserGroup {
+	s.RLock()
+	defer s.RUnlock()
+	var userGroups []*models.UserGroup
+	for _, id := range ids {
+		if s.Data[id] == nil {
+			continue
+		}
+		userGroups = append(userGroups, s.Data[id])
+	}
+
+	return userGroups
+}
+
+func (s *UserGroupMap) SetAll(userGroups map[int64]*models.UserGroup) {
+	s.Lock()
+	defer s.Unlock()
+	s.Data = userGroups
+}

http/router_alert_rule.go

@@ -3,16 +3,21 @@ package http
 import (
 	"encoding/json"
 	"net/http"
+	"strconv"
+	"strings"
 	"time"
 
 	"github.com/gin-gonic/gin"
 
+	"github.com/didi/nightingale/v5/cache"
 	"github.com/didi/nightingale/v5/config"
 	"github.com/didi/nightingale/v5/models"
 )
 
 func alertRuleGet(c *gin.Context) {
-	renderData(c, AlertRule(urlParamInt64(c, "id")), nil)
+	alertRule := AlertRule(urlParamInt64(c, "id"))
+	alertRuleFillUserAndGroups(alertRule)
+	renderData(c, alertRule, nil)
 }
 
 type alertRuleForm struct {
@@ -163,3 +168,22 @@ func alertRuleDel(c *gin.Context) {
 func notifyChannelsGet(c *gin.Context) {
 	renderData(c, config.Config.NotifyChannels, nil)
 }
+
+func alertRuleFillUserAndGroups(alertRule *models.AlertRule) {
+	uidStrs := strings.Fields(alertRule.NotifyUsers)
+	var uids []int64
+	for _, uidStr := range uidStrs {
+		uid, _ := strconv.ParseInt(uidStr, 10, 64)
+		uids = append(uids, uid)
+	}
+	alertRule.NotifyUsersDetail = cache.UserCache.GetByIds(uids)
+
+	gidStrs := strings.Fields(alertRule.NotifyGroups)
+	var gids []int64
+	for _, gidStr := range gidStrs {
+		gid, _ := strconv.ParseInt(gidStr, 10, 64)
+		gids = append(gids, gid)
+	}
+
+	alertRule.NotifyGroupsDetail = cache.UserGroupCache.GetByIds(gids)
+}

http/router_alert_rule_group.go

@@ -68,6 +68,10 @@ func alertRuleGroupGet(c *gin.Context) {
 
 func alertRuleOfGroupGet(c *gin.Context) {
 	ars, err := models.AlertRulesOfGroup(urlParamInt64(c, "id"))
+	for i := range ars {
+		alertRuleFillUserAndGroups(&ars[i])
+	}
+
 	renderData(c, ars, err)
 }
 

main.go

@@ -71,6 +71,7 @@ func main() {
 
 	timer.SyncResourceTags()
 	timer.SyncUsers()
+	timer.SyncUserGroups()
 	timer.SyncUserGroupMember()
 	timer.SyncClasspathReses()
 	timer.SyncCollectRules()

models/alert_rule.go

@@ -19,32 +19,34 @@ const ALERT_RULE_ACTIVE = 0
 const ALERT_RULE_DISABLED = 1
 
 type AlertRule struct {
-	Id               int64           `json:"id"`
-	GroupId          int64           `json:"group_id"`
-	Name             string          `json:"name"`
-	Type             int             `json:"type"` // 0: nightingale, 1: prometheus
-	Expression       json.RawMessage `json:"expression"`
-	Status           int             `json:"status"` // 0: active, 1: disabled
-	AppendTags       string          `json:"append_tags"`
-	EnableStime      string          `json:"enable_stime"`
-	EnableEtime      string          `json:"enable_etime"`
-	EnableDaysOfWeek string          `json:"enable_days_of_week"`
-	RecoveryNotify   int             `json:"recovery_notify"`
-	Priority         int             `json:"priority"`
-	NotifyChannels   string          `json:"notify_channels"`
-	NotifyGroups     string          `json:"notify_groups"`
-	NotifyUsers      string          `json:"notify_users"`
-	Callbacks        string          `json:"callbacks"`
-	RunbookUrl       string          `json:"runbook_url"`
-	Note             string          `json:"note"`
-	CreateAt         int64           `json:"create_at"`
-	CreateBy         string          `json:"create_by"`
-	UpdateAt         int64           `json:"update_at"`
-	UpdateBy         string          `json:"update_by"`
-	AlertDuration    int             `json:"alert_duration"` // 告警统计周期，PULL模型会当做P8S的for时间
-	PushExpr         PushExpression  `xorm:"-" json:"-"`
-	PullExpr         PullExpression  `xorm:"-" json:"-"`
-	FirstMetric      string          `xorm:"-" json:"-"` // Exps里可能有多个metric，只取第一个，给后续制作map使用
+	Id                 int64           `json:"id"`
+	GroupId            int64           `json:"group_id"`
+	Name               string          `json:"name"`
+	Type               int             `json:"type"` // 0: nightingale, 1: prometheus
+	Expression         json.RawMessage `json:"expression"`
+	Status             int             `json:"status"` // 0: active, 1: disabled
+	AppendTags         string          `json:"append_tags"`
+	EnableStime        string          `json:"enable_stime"`
+	EnableEtime        string          `json:"enable_etime"`
+	EnableDaysOfWeek   string          `json:"enable_days_of_week"`
+	RecoveryNotify     int             `json:"recovery_notify"`
+	Priority           int             `json:"priority"`
+	NotifyChannels     string          `json:"notify_channels"`
+	NotifyGroups       string          `json:"notify_groups"`
+	NotifyUsers        string          `json:"notify_users"`
+	Callbacks          string          `json:"callbacks"`
+	RunbookUrl         string          `json:"runbook_url"`
+	Note               string          `json:"note"`
+	CreateAt           int64           `json:"create_at"`
+	CreateBy           string          `json:"create_by"`
+	UpdateAt           int64           `json:"update_at"`
+	UpdateBy           string          `json:"update_by"`
+	AlertDuration      int             `json:"alert_duration"` // 告警统计周期，PULL模型会当做P8S的for时间
+	PushExpr           PushExpression  `xorm:"-" json:"-"`
+	PullExpr           PullExpression  `xorm:"-" json:"-"`
+	FirstMetric        string          `xorm:"-" json:"-"` // Exps里可能有多个metric，只取第一个，给后续制作map使用
+	NotifyUsersDetail  []*User         `xorm:"-" json:"notify_users_detail"`
+	NotifyGroupsDetail []*UserGroup    `xorm:"-" json:"notify_groups_detail"`
 }
 
 type PushExpression struct {

models/user_group.go

@@ -204,3 +204,19 @@ func (ug *UserGroup) Del() error {
 
 	return session.Commit()
 }
+
+func UserGroupGetAll() ([]UserGroup, error) {
+	var userGroups []UserGroup
+
+	err := DB.Find(&userGroups)
+	if err != nil {
+		logger.Errorf("mysql.error: select user_group fail: %v", err)
+		return userGroups, internalServerError
+	}
+
+	if len(userGroups) == 0 {
+		return []UserGroup{}, nil
+	}
+
+	return userGroups, nil
+}

timer/user_group.go

+package timer
+
+import (
+	"fmt"
+	"math/rand"
+	"time"
+
+	"github.com/didi/nightingale/v5/cache"
+	"github.com/didi/nightingale/v5/models"
+
+	"github.com/toolkits/pkg/logger"
+)
+
+// user_group_id->user_group 将数据库中的用户信息缓存在内存里，
+// 在生成告警事件的时候，根据用户ID快速找到用户的详情
+func SyncUserGroups() {
+	err := syncUserGroups()
+	if err != nil {
+		fmt.Println("timer: sync users fail:", err)
+		exit(1)
+	}
+
+	go loopSyncUserGroups()
+}
+
+func loopSyncUserGroups() {
+	randtime := rand.Intn(9000)
+	fmt.Printf("timer: sync users: random sleep %dms\n", randtime)
+	time.Sleep(time.Duration(randtime) * time.Millisecond)
+
+	for {
+		time.Sleep(time.Second * time.Duration(9))
+		err := syncUserGroups()
+		if err != nil {
+			logger.Warning("timer: sync users fail:", err)
+		}
+	}
+}
+
+func syncUserGroups() error {
+	start := time.Now()
+
+	userGroups, err := models.UserGroupGetAll()
+	if err != nil {
+		return err
+	}
+
+	userGroupsMap := make(map[int64]*models.UserGroup)
+	for i := range userGroups {
+		userGroupsMap[userGroups[i].Id] = &userGroups[i]
+	}
+
+	cache.UserGroupCache.SetAll(userGroupsMap)
+	logger.Debugf("timer: sync userGroups done, cost: %dms", time.Since(start).Milliseconds())
+
+	return nil
+}