From 0fb4e4b723a124862d64af29ffb3a6db041d64ec Mon Sep 17 00:00:00 2001
From: ning <710leo@gmail.com>
Date: Tue, 27 May 2025 11:27:33 +0800
Subject: [PATCH] refactor: add eval duration

---
 alert/astats/stats.go | 10 ++++++++++
 alert/eval/eval.go    | 12 ++++++++++++
 2 files changed, 22 insertions(+)

diff --git a/alert/astats/stats.go b/alert/astats/stats.go
index 3d27e095..9d2e6de3 100644
--- a/alert/astats/stats.go
+++ b/alert/astats/stats.go
@@ -25,6 +25,7 @@ type Stats struct {
 	CounterHeartbeatErrorTotal  *prometheus.CounterVec
 	CounterSubEventTotal        *prometheus.CounterVec
 	GaugeQuerySeriesCount       *prometheus.GaugeVec
+	GaugeRuleEvalDuration       *prometheus.GaugeVec
 	GaugeNotifyRecordQueueSize  prometheus.Gauge
 }
 
@@ -136,6 +137,13 @@ func NewSyncStats() *Stats {
 		Help:      "The size of notify record queue.",
 	})
 
+	GaugeRuleEvalDuration := prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: namespace,
+		Subsystem: subsystem,
+		Name:      "rule_eval_duration_ms",
+		Help:      "Duration of rule eval in milliseconds.",
+	}, []string{"rule_id", "datasource_id"})
+
 	CounterVarFillingQuery := prometheus.NewCounterVec(prometheus.CounterOpts{
 		Namespace: namespace,
 		Subsystem: subsystem,
@@ -158,6 +166,7 @@ func NewSyncStats() *Stats {
 		CounterHeartbeatErrorTotal,
 		CounterSubEventTotal,
 		GaugeQuerySeriesCount,
+		GaugeRuleEvalDuration,
 		GaugeNotifyRecordQueueSize,
 		CounterVarFillingQuery,
 	)
@@ -177,6 +186,7 @@ func NewSyncStats() *Stats {
 		CounterHeartbeatErrorTotal:  CounterHeartbeatErrorTotal,
 		CounterSubEventTotal:        CounterSubEventTotal,
 		GaugeQuerySeriesCount:       GaugeQuerySeriesCount,
+		GaugeRuleEvalDuration:       GaugeRuleEvalDuration,
 		GaugeNotifyRecordQueueSize:  GaugeNotifyRecordQueueSize,
 		CounterVarFillingQuery:      CounterVarFillingQuery,
 	}
diff --git a/alert/eval/eval.go b/alert/eval/eval.go
index 08cc0ff1..ea513c6f 100644
--- a/alert/eval/eval.go
+++ b/alert/eval/eval.go
@@ -232,6 +232,10 @@ func (arw *AlertRuleWorker) Stop() {
 
 func (arw *AlertRuleWorker) GetPromAnomalyPoint(ruleConfig string) ([]models.AnomalyPoint, error) {
 	var lst []models.AnomalyPoint
+	start := time.Now()
+	defer func() {
+		arw.Processor.Stats.GaugeRuleEvalDuration.WithLabelValues(fmt.Sprintf("%v", arw.Rule.Id), fmt.Sprintf("%v", arw.Processor.DatasourceId())).Set(float64(time.Since(start).Milliseconds()))
+	}()
 
 	var rule *models.PromRuleConfig
 	if err := json.Unmarshal([]byte(ruleConfig), &rule); err != nil {
@@ -742,6 +746,10 @@ func combine(paramKeys []string, paraMap map[string][]string, index int, current
 
 func (arw *AlertRuleWorker) GetHostAnomalyPoint(ruleConfig string) ([]models.AnomalyPoint, error) {
 	var lst []models.AnomalyPoint
+	start := time.Now()
+	defer func() {
+		arw.Processor.Stats.GaugeRuleEvalDuration.WithLabelValues(fmt.Sprintf("%v", arw.Rule.Id), fmt.Sprintf("%v", arw.Processor.DatasourceId())).Set(float64(time.Since(start).Milliseconds()))
+	}()
 
 	var rule *models.HostRuleConfig
 	if err := json.Unmarshal([]byte(ruleConfig), &rule); err != nil {
@@ -1416,6 +1424,10 @@ func fillVar(curRealQuery string, paramKey string, val string) string {
 
 func (arw *AlertRuleWorker) GetAnomalyPoint(rule *models.AlertRule, dsId int64) ([]models.AnomalyPoint, []models.AnomalyPoint) {
 	// 获取查询和规则判断条件
+	start := time.Now()
+	defer func() {
+		arw.Processor.Stats.GaugeRuleEvalDuration.WithLabelValues(fmt.Sprintf("%v", arw.Rule.Id), fmt.Sprintf("%v", arw.Processor.DatasourceId())).Set(float64(time.Since(start).Milliseconds()))
+	}()
 	points := []models.AnomalyPoint{}
 	recoverPoints := []models.AnomalyPoint{}
 	ruleConfig := strings.TrimSpace(rule.RuleConfig)