fix judge prom

judge/handler.go

@@ -449,14 +449,14 @@ func sendEventIfNeed(status []bool, event *models.AlertEvent, stra *models.Alert
 	}
 
 	now := time.Now().Unix()
-	lastEvent, exists := LastEvents.Get(event.HashId)
+	lastEvent, exists := LastEvents.Get(event.RuleId, event.HashId)
 
 	switch event.IsPromePull {
 	case 0:
 		//	push型的 && 与条件型的
 		if exists && lastEvent.IsPromePull == 1 {
 			// 之前内存中的事件是pull型的，先清空内存中的事件
-			LastEvents.Del(event.HashId)
+			LastEvents.Del(event.RuleId, event.HashId)
 		}
 
 		if isTriggered {
@@ -476,7 +476,7 @@ func sendEventIfNeed(status []bool, event *models.AlertEvent, stra *models.Alert
 		// pull型的，产生的事件一定是触发了阈值的，即这个case里不存在recovery的场景，recovery的场景用resolve_timeout的cron来处理
 		if exists && lastEvent.IsPromePull == 0 {
 			// 之前内存中的事件是push型的，先清空内存中的事件
-			LastEvents.Del(event.HashId)
+			LastEvents.Del(event.RuleId, event.HashId)
 		}
 
 		// 1. 第一次来，并且AlertDuration=0，直接发送
@@ -490,7 +490,7 @@ func sendEventIfNeed(status []bool, event *models.AlertEvent, stra *models.Alert
 				SendEvent(event)
 			} else {
 				// 只有一条事件，显然无法满足for AlertDuration的时间，放到内存里等待
-				LastEvents.Set(event.HashId, event)
+				LastEvents.Set(event)
 			}
 			return
 		}
@@ -529,7 +529,7 @@ func sendEventIfNeed(status []bool, event *models.AlertEvent, stra *models.Alert
 
 func SendEvent(event *models.AlertEvent) {
 	// update last event
-	LastEvents.Set(event.HashId, event)
+	LastEvents.Set(event)
 	ok := EventQueue.PushFront(event)
 	if !ok {
 		logger.Errorf("push event:%v err", event)

judge/last_event.go

@@ -4,58 +4,85 @@ import (
 	"sync"
 	"time"
 
-	"github.com/toolkits/pkg/logger"
-
 	"github.com/didi/nightingale/v5/models"
+	"github.com/toolkits/pkg/logger"
 )
 
+// rule_id -> hash_id -> *models.AlertEvent
 type SafeEventMap struct {
 	sync.RWMutex
-	M map[string]*models.AlertEvent
+	M map[int64]map[string]*models.AlertEvent
 }
 
 var (
-	LastEvents = &SafeEventMap{M: make(map[string]*models.AlertEvent)}
+	LastEvents = &SafeEventMap{M: make(map[int64]map[string]*models.AlertEvent)}
 )
 
-func (s *SafeEventMap) Get(key string) (*models.AlertEvent, bool) {
+func (s *SafeEventMap) Get(ruleId int64, hashId string) (*models.AlertEvent, bool) {
 	s.RLock()
 	defer s.RUnlock()
-	event, exists := s.M[key]
-	return event, exists
+
+	m, has := s.M[ruleId]
+	if !has {
+		return nil, false
+	}
+
+	event, has := m[hashId]
+	return event, has
 }
 
-func (s *SafeEventMap) Set(key string, event *models.AlertEvent) {
+func (s *SafeEventMap) Set(event *models.AlertEvent) {
 	s.Lock()
 	defer s.Unlock()
-	s.M[key] = event
+
+	m, has := s.M[event.RuleId]
+	if !has {
+		m = make(map[string]*models.AlertEvent)
+		m[event.HashId] = event
+		s.M[event.RuleId] = m
+	} else {
+		s.M[event.RuleId][event.HashId] = event
+	}
 }
 
-func (s *SafeEventMap) Del(key string) {
+func (s *SafeEventMap) Del(ruleId int64, hashId string) {
 	s.Lock()
 	defer s.Unlock()
-	delete(s.M, key)
+
+	_, has := s.M[ruleId]
+	if !has {
+		return
+	}
+
+	delete(s.M[ruleId], hashId)
 }
 
-func (s *SafeEventMap) DeleteOrSendRecovery(promql string, toKeepKeys map[string]struct{}) {
+func (s *SafeEventMap) DeleteOrSendRecovery(ruleId int64, toKeepKeys map[string]struct{}) {
 	s.Lock()
 	defer s.Unlock()
-	for k, ev := range s.M {
+
+	m, has := s.M[ruleId]
+	if !has {
+		return
+	}
+
+	for k, ev := range m {
 		if _, loaded := toKeepKeys[k]; loaded {
 			continue
 		}
-		if ev.ReadableExpression == promql {
-			logger.Debugf("[to_del][ev.IsRecovery:%+v][ev.LastSend:%+v][promql:%v]", ev.IsRecovery, ev.LastSend, promql)
-			now := time.Now().Unix()
-			// promql 没查询到结果，需要将告警标记为已恢复并发送
-			// 同时需要满足 已经发送过触发信息，并且时间差满足 大于AlertDuration
-			// 为了避免 发送告警后 一个点 断点了就立即发送恢复信息的case
-			if ev.IsAlert() && ev.LastSend && now-ev.TriggerTime > ev.AlertDuration {
-				logger.Debugf("[prom.alert.MarkRecov][promql:%v][ev.RuleName:%v]", promql, ev.RuleName)
-				ev.MarkRecov()
-				EventQueue.PushFront(ev)
-				delete(s.M, k)
-			}
+
+		// 如果因为promql修改，导致本来是告警状态变成了恢复，也接受
+		logger.Debugf("[to_del][ev.IsRecovery:%+v][ev.LastSend:%+v]", ev.IsRecovery, ev.LastSend)
+
+		// promql 没查询到结果，需要将告警标记为已恢复并发送
+		// 同时需要满足 已经发送过触发信息，并且时间差满足 大于AlertDuration
+		// 为了避免 发送告警后 一个点 断点了就立即发送恢复信息的case
+		now := time.Now().Unix()
+		if ev.IsAlert() && ev.LastSend && now-ev.TriggerTime > ev.AlertDuration {
+			logger.Debugf("[prom.alert.MarkRecov][ev.RuleName:%v]", ev.RuleName)
+			ev.MarkRecov()
+			EventQueue.PushFront(ev)
+			delete(s.M[ruleId], k)
 		}
 	}
 }

judge/prome_pull.go

@@ -121,7 +121,7 @@ func handlePromqlVector(pv promql.Vector, r models.AlertRule) {
 	toKeepKeys := map[string]struct{}{}
 	if len(pv) == 0 {
 		// 说明没触发，或者没查询到，删掉rule-id开头的所有event
-		LastEvents.DeleteOrSendRecovery(r.PullExpr.PromQl, toKeepKeys)
+		LastEvents.DeleteOrSendRecovery(r.Id, toKeepKeys)
 
 		return
 	}
@@ -191,6 +191,6 @@ func handlePromqlVector(pv promql.Vector, r models.AlertRule) {
 		logger.Debugf("[handlePromqlVector_has_value][event:%+v]\n", event)
 		sendEventIfNeed([]bool{true}, event, &r)
 	}
-	LastEvents.DeleteOrSendRecovery(r.PullExpr.PromQl, toKeepKeys)
+	LastEvents.DeleteOrSendRecovery(r.Id, toKeepKeys)
 
 }