Fix LatencyNative Indicator Grouping Support

slo/promql.go

@@ -2,6 +2,7 @@ package slo
 
 import (
 	"fmt"
+	"sort"
 	"time"
 
 	"github.com/prometheus/common/model"
@@ -633,14 +634,15 @@ func (o Objective) RequestRange(timerange time.Duration) string {
 
 		return expr.String()
 	case LatencyNative:
-		expr, err := parser.ParseExpr(`sum(histogram_count(rate(metric{}[1s])))`)
+		expr, err := parser.ParseExpr(`sum by (grouping) (histogram_count(rate(metric{}[1s])))`)
 		if err != nil {
 			return err.Error()
 		}
 
 		objectiveReplacer{
 			metric:   o.Indicator.LatencyNative.Total.Name,
 			matchers: o.Indicator.LatencyNative.Total.LabelMatchers,
+			grouping: o.Indicator.LatencyNative.Grouping,
 			window:   timerange,
 		}.replace(expr)
 
@@ -716,14 +718,26 @@ func (o Objective) ErrorsRange(timerange time.Duration) string {
 
 		return expr.String()
 	case LatencyNative:
-		expr, err := parser.ParseExpr(`1 - histogram_fraction(0,0.696969, sum(rate(metric{matchers="total"}[1s])))`)
+		expr, err := parser.ParseExpr(`1 - histogram_fraction(0,0.696969, sum by (grouping) (rate(metric{matchers="total"}[1s])))`)
 		if err != nil {
 			return err.Error()
 		}
 
+		groupingMap := map[string]struct{}{}
+		for _, s := range o.Indicator.LatencyNative.Grouping {
+			groupingMap[s] = struct{}{}
+		}
+
+		grouping := make([]string, 0, len(groupingMap))
+		for s := range groupingMap {
+			grouping = append(grouping, s)
+		}
+		sort.Strings(grouping)
+
 		objectiveReplacer{
 			metric:   o.Indicator.LatencyNative.Total.Name,
 			matchers: o.Indicator.LatencyNative.Total.LabelMatchers,
+			grouping: grouping,
 			window:   timerange,
 			target:   time.Duration(o.Indicator.LatencyNative.Latency).Seconds(),
 		}.replace(expr)
@@ -772,7 +786,7 @@ func (o Objective) DurationRange(timerange time.Duration, percentile float64) st
 
 		return expr.String()
 	case LatencyNative:
-		expr, err := parser.ParseExpr(`histogram_quantile(0.420, sum(rate(metric{matchers="total"}[1s])))`)
+		expr, err := parser.ParseExpr(`histogram_quantile(0.420, sum by (grouping) (rate(metric{matchers="total"}[1s])))`)
 		if err != nil {
 			return err.Error()
 		}

slo/promql_test.go

@@ -2,6 +2,7 @@ package slo
 
 import (
 	"strconv"
+	"strings"
 	"testing"
 	"time"
 
@@ -146,6 +147,11 @@ var (
 		}
 		return o
 	}
+	objectiveHTTPNativeLatencyGrouping = func() Objective {
+		o := objectiveHTTPNativeLatency()
+		o.Indicator.LatencyNative.Grouping = []string{"job", "handler"}
+		return o
+	}
 	objectiveHTTPLatencyGrouping = func() Objective {
 		o := objectiveHTTPLatency()
 		o.Indicator.Latency.Grouping = []string{"job", "handler"}
@@ -641,6 +647,13 @@ func TestObjective_QueryBurnrate(t *testing.T) {
 		name:      "http-latency-native",
 		objective: objectiveHTTPNativeLatency(),
 		expected:  `http_request_duration_seconds:burnrate5m{code=~"2..",job="metrics-service-thanos-receive-default",slo="monitoring-http-latency"}`,
+	}, {
+		name:      "http-latency-native-grouping",
+		objective: objectiveHTTPNativeLatencyGrouping(),
+		grouping: []*labels.Matcher{
+			{Type: labels.MatchEqual, Name: "handler", Value: "/api/v1/query"},
+		},
+		expected: `http_request_duration_seconds:burnrate5m{code=~"2..",handler="/api/v1/query",job="metrics-service-thanos-receive-default",slo="monitoring-http-latency"}`,
 	}, {
 		name:      "http-latency-grouping",
 		objective: objectiveHTTPLatencyGrouping(),
@@ -1086,3 +1099,24 @@ func TestReplacer(t *testing.T) {
 		})
 	}
 }
+
+func TestLatencyNativeBurnrateGrouping(t *testing.T) {
+	objective := objectiveHTTPNativeLatencyGrouping()
+
+	burnrateQuery := objective.Burnrate(5 * time.Minute)
+
+	require.True(t,
+		strings.Contains(burnrateQuery, "sum by (handler, job)") || strings.Contains(burnrateQuery, "sum by (job, handler)"),
+		"LatencyNative burnrate query should include grouping, got: %s", burnrateQuery)
+	require.Contains(t, burnrateQuery, "histogram_fraction", "LatencyNative burnrate should use histogram_fraction")
+
+	requestRangeQuery := objective.RequestRange(2 * time.Hour)
+	require.True(t,
+		strings.Contains(requestRangeQuery, "sum by (handler, job)") || strings.Contains(requestRangeQuery, "sum by (job, handler)"),
+		"LatencyNative RequestRange should include grouping, got: %s", requestRangeQuery)
+
+	errorsRangeQuery := objective.ErrorsRange(2 * time.Hour)
+	require.True(t,
+		strings.Contains(errorsRangeQuery, "sum by (handler, job)") || strings.Contains(errorsRangeQuery, "sum by (job, handler)"),
+		"LatencyNative ErrorsRange should include grouping, got: %s", errorsRangeQuery)
+}

slo/rules.go

@@ -512,7 +512,7 @@ func (o Objective) Burnrate(timerange time.Duration) string {
 
 		return expr.String()
 	case LatencyNative:
-		expr, err := parser.ParseExpr(`1 - histogram_fraction(0,0.696969, sum(rate(metric{matchers="total"}[1s])))`)
+		expr, err := parser.ParseExpr(`1 - histogram_fraction(0,0.696969, sum by (grouping) (rate(metric{matchers="total"}[1s])))`)
 		if err != nil {
 			return err.Error()
 		}
@@ -903,7 +903,7 @@ func (o Objective) IncreaseRules() (monitoringv1.RuleGroup, error) {
 			}
 		}
 
-		expr, err := parser.ParseExpr(`histogram_count(sum(increase(metric{matchers="total"}[1s])))`)
+		expr, err := parser.ParseExpr(`histogram_count(sum by (grouping) (increase(metric{matchers="total"}[1s])))`)
 		if err != nil {
 			return monitoringv1.RuleGroup{}, err
 		}
@@ -921,7 +921,7 @@ func (o Objective) IncreaseRules() (monitoringv1.RuleGroup, error) {
 			Labels: ruleLabels,
 		})
 
-		expr, err = parser.ParseExpr(`histogram_fraction(0, 0.696969, sum(increase(metric{matchers="total"}[1s]))) * histogram_count(sum(increase(metric{matchers="total"}[1s])))`)
+		expr, err = parser.ParseExpr(`histogram_fraction(0, 0.696969, sum by (grouping) (increase(metric{matchers="total"}[1s]))) * histogram_count(sum by (grouping) (increase(metric{matchers="total"}[1s])))`)
 		if err != nil {
 			return monitoringv1.RuleGroup{}, err
 		}