Optimize SLO performance with subquery-based calculations

.github/workflows/container.yaml

@@ -5,6 +5,7 @@ on:
     branches:
       - 'release-*'
       - 'main'
+      - 'improve-slo-performance-with-subqueries'
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.number || github.ref }}

Makefile

@@ -77,7 +77,7 @@ docker-push:
 # download controller-gen if necessary
 controller-gen:
 ifeq (, $(shell which controller-gen))
-	go install sigs.k8s.io/controller-tools/cmd/controller-gen@v0.14.0
+	go install sigs.k8s.io/controller-tools/cmd/controller-gen@v0.19.0
 CONTROLLER_GEN=$(GOBIN)/controller-gen
 else
 CONTROLLER_GEN=$(shell which controller-gen)

jsonnet/controller-gen/pyrra.dev_servicelevelobjectives.json

@@ -3,7 +3,7 @@
   "kind": "CustomResourceDefinition",
   "metadata": {
     "annotations": {
-      "controller-gen.kubebuilder.io/version": "v0.18.0"
+      "controller-gen.kubebuilder.io/version": "v0.19.0"
     },
     "name": "servicelevelobjectives.pyrra.dev"
   },
@@ -242,6 +242,10 @@
                     ],
                     "type": "string"
                   },
+                  "performance_over_accuracy": {
+                    "default": false,
+                    "type": "boolean"
+                  },
                   "target": {
                     "description": "Target is a string that's casted to a float64 between 0 - 100.\nIt represents the desired availability of the service in the given window.\nfloat64 are not supported: https://github.com/kubernetes-sigs/controller-tools/issues/245",
                     "type": "string"

kubernetes/api/v1alpha1/servicelevelobjective_types.go

@@ -99,6 +99,10 @@ type ServiceLevelObjectiveSpec struct {
 	// be ignored by Prometheus instances.
 	// More info: https://github.com/thanos-io/thanos/blob/main/docs/components/rule.md#partial-response
 	PartialResponseStrategy string `json:"partial_response_strategy,omitempty"`
+
+	// +optional
+	// +kubebuilder:default:=false
+	PerformanceOverAccuracy bool `json:"performance_over_accuracy,omitempty"`
 }
 
 // ServiceLevelIndicator defines the underlying indicator that is a Prometheus metric.
@@ -378,6 +382,7 @@ func (in *ServiceLevelObjective) Internal() (slo.Objective, error) {
 	if err != nil {
 		return slo.Objective{}, fmt.Errorf("failed to parse objective window: %w", err)
 	}
+
 	var alerting slo.Alerting
 	alerting.Disabled = false
 	if in.Spec.Alerting.Disabled != nil {
@@ -575,13 +580,14 @@ func (in *ServiceLevelObjective) Internal() (slo.Objective, error) {
 	}
 
 	return slo.Objective{
-		Labels:      ls,
-		Annotations: in.Annotations,
-		Description: in.Spec.Description,
-		Target:      target / 100,
-		Window:      window,
-		Config:      string(config),
-		Alerting:    alerting,
+		Labels:                  ls,
+		Annotations:             in.Annotations,
+		Description:             in.Spec.Description,
+		Target:                  target / 100,
+		Window:                  window,
+		PerformanceOverAccuracy: in.Spec.PerformanceOverAccuracy,
+		Config:                  string(config),
+		Alerting:                alerting,
 		Indicator: slo.Indicator{
 			Ratio:         ratio,
 			Latency:       latency,

slo/promql.go

@@ -589,6 +589,10 @@ func (r objectiveReplacer) replace(node parser.Node) {
 			n.Range = r.window
 		}
 		r.replace(n.VectorSelector)
+	case *parser.SubqueryExpr:
+		n.Range = r.window
+		n.Step = 5 * time.Minute
+		r.replace(n.Expr)
 	case *parser.VectorSelector:
 		if n.Name == "errorMetric" {
 			n.Name = r.errorMetric

slo/promql_test.go

@@ -42,6 +42,11 @@ var (
 			},
 		}
 	}
+	objectiveHTTPRatioGroupingLessAccurate = func() Objective {
+		o := objectiveHTTPRatio()
+		o.PerformanceOverAccuracy = true
+		return o
+	}
 	objectiveHTTPRatioGrouping = func() Objective {
 		o := objectiveHTTPRatio()
 		o.Indicator.Ratio.Grouping = []string{"job", "handler"}
@@ -98,6 +103,11 @@ var (
 		o.Indicator.Ratio.Grouping = []string{"job", "handler"}
 		return o
 	}
+	objectiveGRPCRatioGroupingLessAccuracy = func() Objective {
+		o := objectiveGRPCRatioGrouping()
+		o.PerformanceOverAccuracy = true
+		return o
+	}
 	objectiveHTTPLatency = func() Objective {
 		return Objective{
 			Labels: labels.FromStrings(labels.MetricName, "monitoring-http-latency"),
@@ -168,6 +178,11 @@ var (
 		o.Indicator.Latency.Total.LabelMatchers = append(o.Indicator.Latency.Total.LabelMatchers, matcher)
 		return o
 	}
+	objectiveHTTPLatencyGroupingRegexLessAccuracy = func() Objective {
+		o := objectiveHTTPLatencyGroupingRegex()
+		o.PerformanceOverAccuracy = true
+		return o
+	}
 	objectiveGRPCLatency = func() Objective {
 		return Objective{
 			Labels: labels.FromStrings(labels.MetricName, "monitoring-grpc-latency"),