Refactor recording-rules and alerts code (#3068)

* Refactor recording-rules and alerts code

Signed-off-by: avlitman <alitman@redhat.com>

* Remove promv1 from schema

Signed-off-by: avlitman <alitman@redhat.com>

---------

Signed-off-by: avlitman <alitman@redhat.com>

Makefile

@@ -156,7 +156,7 @@ openshift-ci-image-push:
 	./hack/build/osci-image-builder.sh
 
 generate-doc: build-docgen
-	_out/pkg/monitoring/tools/metricsdocs/metricsdocs > doc/metrics.md
+	_out/tools/metricsdocs/metricsdocs > doc/metrics.md
 
 bootstrap-ginkgo:
 	${DO_BAZ} ./hack/build/bootstrap-ginkgo.sh

doc/metrics.md

@@ -1,5 +1,8 @@
 # Containerized Data Importer metrics
 
+### kubevirt_cdi_clone_pods_high_restart
+The number of CDI clone pods with high restart count. Type: Gauge.
+
 ### kubevirt_cdi_cr_ready
 CDI install ready. Type: Gauge.
 
@@ -9,18 +12,15 @@ DataImportCron has an outdated import. Type: Gauge.
 ### kubevirt_cdi_datavolume_pending
 Number of DataVolumes pending for default storage class to be configured. Type: Gauge.
 
-### kubevirt_cdi_storageprofile_info
-`StorageProfiles` info labels: `storageclass`, `provisioner`, `complete` indicates if all storage profiles recommended PVC settings are complete, `default` indicates if it's the Kubernetes default storage class, `virtdefault` indicates if it's the default virtualization storage class, `rwx` indicates if the storage class supports `ReadWriteMany`, `smartclone` indicates if it supports snapshot or CSI based clone. Type: Gauge.
-
-### kubevirt_cdi_clone_pods_high_restart
-The number of CDI clone pods with high restart count. Type: Gauge.
-
 ### kubevirt_cdi_import_pods_high_restart
 The number of CDI import pods with high restart count. Type: Gauge.
 
 ### kubevirt_cdi_operator_up
 CDI operator status. Type: Gauge.
 
+### kubevirt_cdi_storageprofile_info
+`StorageProfiles` info labels: `storageclass`, `provisioner`, `complete` indicates if all storage profiles recommended PVC settings are complete, `default` indicates if it's the Kubernetes default storage class, `virtdefault` indicates if it's the default virtualization storage class, `rwx` indicates if the storage class supports `ReadWriteMany`, `smartclone` indicates if it supports snapshot or CSI based clone. Type: Gauge.
+
 ### kubevirt_cdi_upload_pods_high_restart
 The number of CDI upload server pods with high restart count. Type: Gauge.
 

go.mod

@@ -23,7 +23,7 @@ require (
 	github.com/kubernetes-csi/external-snapshotter/client/v6 v6.0.1
 	github.com/kubernetes-csi/lib-volume-populator v1.2.1-0.20230316163120-b62a0eee2c56
 	github.com/kubevirt/monitoring/pkg/metrics/parser v0.0.0-20230627123556-81a891d4462a
-	github.com/machadovilaca/operator-observability v0.0.9
+	github.com/machadovilaca/operator-observability v0.0.13
 	github.com/onsi/ginkgo/v2 v2.12.0
 	github.com/onsi/gomega v1.27.10
 	github.com/openshift/api v0.0.0-20240116035456-11ed2fbcb805

go.sum

@@ -1099,8 +1099,8 @@ github.com/kubevirt/monitoring/pkg/metrics/parser v0.0.0-20230627123556-81a891d4
 github.com/kubevirt/monitoring/pkg/metrics/parser v0.0.0-20230627123556-81a891d4462a/go.mod h1:qGj2agzgwQ27nYhP3xhLs+IBzE5+ALNUg8bDfMcwPqo=
 github.com/lyft/protoc-gen-star v0.6.0/go.mod h1:TGAoBVkt8w7MPG72TrKIu85MIdXwDuzJYeZuUPFPNwA=
 github.com/lyft/protoc-gen-star v0.6.1/go.mod h1:TGAoBVkt8w7MPG72TrKIu85MIdXwDuzJYeZuUPFPNwA=
-github.com/machadovilaca/operator-observability v0.0.9 h1:jL2jVh0YJNA3nSX216X74RDZiEPPvsgqXollYmMOQkg=
-github.com/machadovilaca/operator-observability v0.0.9/go.mod h1:NGkaR3HEYLScVQf6kQAyxWOSN1ltHcsEvHU/8iIJ8cE=
+github.com/machadovilaca/operator-observability v0.0.13 h1:9mhxEjkdE6pcl3ke8chbbAWxx25+K1m4Gq31yo+r2JU=
+github.com/machadovilaca/operator-observability v0.0.13/go.mod h1:e4Z3VhOXb9InkmSh00JjqBBijE+iD+YMzynBpKB3+gE=
 github.com/magiconair/properties v1.8.1/go.mod h1:PppfXfuXeibc/6YijjN8zIbojt8czPbwD3XqdrwzmxQ=
 github.com/mailru/easyjson v0.0.0-20180823135443-60711f1a8329/go.mod h1:C1wdFJiN94OJF2b5HbByQZoLdCWB1Yqtg26g4irojpc=
 github.com/mailru/easyjson v0.0.0-20190614124828-94de47d64c63/go.mod h1:C1wdFJiN94OJF2b5HbByQZoLdCWB1Yqtg26g4irojpc=

hack/build/bazel-build-metricsdocs.sh

@@ -29,10 +29,10 @@ mkdir -p ${CMD_OUT_DIR}/dump
 bazel build \
     --verbose_failures \
     --config=${ARCHITECTURE} \
-    //pkg/monitoring/tools/metricsdocs/...
+    //tools/metricsdocs/...
 
-rm -rf _out/pkg/monitoring/tools/metricsdocs
-mkdir -p _out/pkg/monitoring/tools/metricsdocs
-cp ./bazel-bin/pkg/monitoring/tools/metricsdocs/metricsdocs_/metricsdocs _out/pkg/monitoring/tools/metricsdocs/
+rm -rf _out/tools/metricsdocs
+mkdir -p _out/tools/metricsdocs
+cp ./bazel-bin/tools/metricsdocs/metricsdocs_/metricsdocs _out/tools/metricsdocs/
 
 bazel clean

hack/ci/prom_metric_linter.sh

@@ -23,7 +23,7 @@ set -e
 linter_image_tag="v0.0.1"
 
 PROJECT_ROOT="$(readlink -e "$(dirname "${BASH_SOURCE[0]}")"/../../)"
-export METRICS_COLLECTOR_PATH="${METRICS_COLLECTOR_PATH:-${PROJECT_ROOT}/pkg/monitoring/tools/prom-metrics-collector}"
+export METRICS_COLLECTOR_PATH="${METRICS_COLLECTOR_PATH:-${PROJECT_ROOT}/tools/prom-metrics-collector}"
 
 if [[ ! -d "$METRICS_COLLECTOR_PATH" ]]; then
     echo "Invalid METRICS_COLLECTOR_PATH: $METRICS_COLLECTOR_PATH is not a valid directory path"

pkg/controller/dataimportcron-conditions.go

@@ -24,7 +24,7 @@ import (
 	"k8s.io/apimachinery/pkg/types"
 
 	cdiv1 "kubevirt.io/containerized-data-importer-api/pkg/apis/core/v1beta1"
-	"kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-controller"
+	metrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-controller"
 )
 
 const (

pkg/monitoring/metrics/cdi-controller/dataimportcron.go

@@ -1,4 +1,4 @@
-package metrics
+package cdicontroller
 
 import (
 	"github.com/prometheus/client_golang/prometheus"

pkg/monitoring/metrics/cdi-controller/datavolume.go

@@ -1,4 +1,4 @@
-package metrics
+package cdicontroller
 
 import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"

pkg/monitoring/metrics/cdi-controller/metrics.go

@@ -1,4 +1,4 @@
-package metrics
+package cdicontroller
 
 import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"

pkg/monitoring/metrics/cdi-controller/storageprofile.go

@@ -1,4 +1,4 @@
-package metrics
+package cdicontroller
 
 import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"

pkg/monitoring/metrics/operator-controller/metrics.go

@@ -1,4 +1,4 @@
-package metrics
+package operatorcontroller
 
 import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"

pkg/monitoring/metrics/operator-controller/operator_metrics.go

@@ -1,4 +1,4 @@
-package metrics
+package operatorcontroller
 
 import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"

pkg/monitoring/rules/BUILD.bazel

@@ -0,0 +1,15 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_library")
+
+go_library(
+    name = "go_default_library",
+    srcs = ["rules.go"],
+    importpath = "kubevirt.io/containerized-data-importer/pkg/monitoring/rules",
+    visibility = ["//visibility:public"],
+    deps = [
+        "//pkg/common:go_default_library",
+        "//pkg/monitoring/rules/alerts:go_default_library",
+        "//pkg/monitoring/rules/recordingrules:go_default_library",
+        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules:go_default_library",
+        "//vendor/github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1:go_default_library",
+    ],
+)

pkg/monitoring/rules/alerts/BUILD.bazel

@@ -0,0 +1,18 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_library")
+
+go_library(
+    name = "go_default_library",
+    srcs = [
+        "operator.go",
+        "prometheus.go",
+    ],
+    importpath = "kubevirt.io/containerized-data-importer/pkg/monitoring/rules/alerts",
+    visibility = ["//visibility:public"],
+    deps = [
+        "//pkg/common:go_default_library",
+        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules:go_default_library",
+        "//vendor/github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1:go_default_library",
+        "//vendor/k8s.io/apimachinery/pkg/util/intstr:go_default_library",
+        "//vendor/k8s.io/utils/ptr:go_default_library",
+    ],
+)

pkg/monitoring/rules/alerts/operator.go

@@ -0,0 +1,109 @@
+package alerts
+
+import (
+	promv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
+	"k8s.io/utils/ptr"
+)
+
+var operatorAlerts = []promv1.Rule{
+	{
+		Alert: "CDIOperatorDown",
+		Expr:  intstr.FromString("kubevirt_cdi_operator_up == 0"),
+		For:   (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "CDI operator is down",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "critical",
+		},
+	},
+	{
+		Alert: "CDINotReady",
+		Expr:  intstr.FromString("kubevirt_cdi_cr_ready == 0"),
+		For:   (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "CDI is not available to use",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "critical",
+		},
+	},
+	{
+		Alert: "CDIDataVolumeUnusualRestartCount",
+		Expr:  intstr.FromString("kubevirt_cdi_import_pods_high_restart > 0 or kubevirt_cdi_upload_pods_high_restart > 0 or kubevirt_cdi_clone_pods_high_restart > 0"),
+		For:   (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "Some CDI population workloads have an unusual restart count, meaning they are probably failing and need to be investigated",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "warning",
+		},
+	},
+	{
+		Alert: "CDIStorageProfilesIncomplete",
+		Expr:  intstr.FromString(`sum by(storageclass,provisioner) ((kubevirt_cdi_storageprofile_info{complete="false"}>0))`),
+		For:   (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "Incomplete StorageProfile {{ $labels.storageclass }}, accessMode/volumeMode cannot be inferred by CDI for PVC population request",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "info",
+			operatorHealthImpactLabelKey: "warning",
+		},
+	},
+	{
+		Alert: "CDIDataImportCronOutdated",
+		Expr:  intstr.FromString("sum by(ns,cron_name) (kubevirt_cdi_dataimportcron_outdated) > 0"),
+		For:   (*promv1.Duration)(ptr.To("15m")),
+		Annotations: map[string]string{
+			"summary": "DataImportCron (recurring polling of VM templates disk image sources, also known as golden images) PVCs are not being updated on the defined schedule",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "info",
+			operatorHealthImpactLabelKey: "warning",
+		},
+	},
+	{
+		Alert: "CDINoDefaultStorageClass",
+		Expr: intstr.FromString(`sum(kubevirt_cdi_storageprofile_info{default="true"} or on() vector(0)) +
+								 sum(kubevirt_cdi_storageprofile_info{virtdefault="true"} or on() vector(0)) +
+								 (count(kubevirt_cdi_datavolume_pending == 0) or on() vector(0)) == 0`),
+		For: (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "No default StorageClass or virtualization StorageClass, and a DataVolume is pending for one",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "none",
+		},
+	},
+	{
+		Alert: "CDIMultipleDefaultVirtStorageClasses",
+		Expr:  intstr.FromString(`sum(kubevirt_cdi_storageprofile_info{virtdefault="true"} or on() vector(0)) > 1`),
+		For:   (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "More than one default virtualization StorageClass detected",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "none",
+		},
+	},
+	{
+		Alert: "CDIDefaultStorageClassDegraded",
+		Expr: intstr.FromString(`sum(kubevirt_cdi_storageprofile_info{default="true",rwx="true",smartclone="true"} or on() vector(0)) +
+								 sum(kubevirt_cdi_storageprofile_info{virtdefault="true",rwx="true",smartclone="true"} or on() vector(0)) == 0`),
+		For: (*promv1.Duration)(ptr.To("5m")),
+		Annotations: map[string]string{
+			"summary": "Default storage class has no smart clone or ReadWriteMany",
+		},
+		Labels: map[string]string{
+			severityAlertLabelKey:        "warning",
+			operatorHealthImpactLabelKey: "none",
+		},
+	},
+}

pkg/monitoring/rules/alerts/prometheus.go

@@ -0,0 +1,58 @@
+package alerts
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"strings"
+
+	"github.com/machadovilaca/operator-observability/pkg/operatorrules"
+	promv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
+	"kubevirt.io/containerized-data-importer/pkg/common"
+)
+
+const (
+	prometheusRunbookAnnotationKey = "runbook_url"
+	defaultRunbookURLTemplate      = "https://kubevirt.io/monitoring/runbooks/%s"
+	runbookURLTemplateEnv          = "RUNBOOK_URL_TEMPLATE"
+
+	severityAlertLabelKey        = "severity"
+	operatorHealthImpactLabelKey = "operator_health_impact"
+	partOfAlertLabelKey          = "kubernetes_operator_part_of"
+	componentAlertLabelKey       = "kubernetes_operator_component"
+	partOfAlertLabelValue        = "kubevirt"
+
+	componentAlertLabelValue = common.CDILabelValue
+)
+
+// Register sets up alert rules in the given namespace.
+func Register(namespace string) error {
+	alerts := [][]promv1.Rule{
+		operatorAlerts,
+	}
+
+	runbookURLTemplate := GetRunbookURLTemplate()
+	for _, alertGroup := range alerts {
+		for _, alert := range alertGroup {
+			alert.Labels[partOfAlertLabelKey] = partOfAlertLabelValue
+			alert.Labels[componentAlertLabelKey] = componentAlertLabelValue
+			alert.Annotations[prometheusRunbookAnnotationKey] = fmt.Sprintf(runbookURLTemplate, alert.Alert)
+		}
+	}
+
+	return operatorrules.RegisterAlerts(alerts...)
+}
+
+// GetRunbookURLTemplate fetches or defaults the runbook URL template.
+func GetRunbookURLTemplate() string {
+	runbookURLTemplate, exists := os.LookupEnv(runbookURLTemplateEnv)
+	if !exists {
+		runbookURLTemplate = defaultRunbookURLTemplate
+	}
+
+	if strings.Count(runbookURLTemplate, "%s") != 1 {
+		panic(errors.New("runbook URL template must have exactly 1 %s substring"))
+	}
+
+	return runbookURLTemplate
+}

pkg/monitoring/rules/recordingrules/BUILD.bazel

@@ -2,8 +2,17 @@ load("@io_bazel_rules_go//go:def.bzl", "go_library")
 
 go_library(
     name = "go_default_library",
-    srcs = ["recordingrules.go"],
+    srcs = [
+        "operator.go",
+        "pods.go",
+        "recordingrules.go",
+    ],
     importpath = "kubevirt.io/containerized-data-importer/pkg/monitoring/rules/recordingrules",
     visibility = ["//visibility:public"],
-    deps = ["//pkg/common:go_default_library"],
+    deps = [
+        "//pkg/common:go_default_library",
+        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatormetrics:go_default_library",
+        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules:go_default_library",
+        "//vendor/k8s.io/apimachinery/pkg/util/intstr:go_default_library",
+    ],
 )

pkg/monitoring/rules/recordingrules/operator.go

@@ -0,0 +1,24 @@
+package recordingrules
+
+import (
+	"fmt"
+
+	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"
+	"github.com/machadovilaca/operator-observability/pkg/operatorrules"
+	"k8s.io/apimachinery/pkg/util/intstr"
+)
+
+func operatorRecordingRules(namespace string) []operatorrules.RecordingRule {
+	return []operatorrules.RecordingRule{
+		{
+			MetricsOpts: operatormetrics.MetricOpts{
+				Name: "kubevirt_cdi_operator_up",
+				Help: "CDI operator status",
+			},
+			MetricType: operatormetrics.GaugeType,
+			Expr: intstr.FromString(
+				fmt.Sprintf("sum(up{namespace='%s', pod=~'cdi-operator-.*'} or vector(0))", namespace),
+			),
+		},
+	}
+}

pkg/monitoring/rules/recordingrules/pods.go

@@ -0,0 +1,45 @@
+package recordingrules
+
+import (
+	"fmt"
+	"strconv"
+
+	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"
+	"github.com/machadovilaca/operator-observability/pkg/operatorrules"
+	"k8s.io/apimachinery/pkg/util/intstr"
+
+	"kubevirt.io/containerized-data-importer/pkg/common"
+)
+
+var podsRecordingRules = []operatorrules.RecordingRule{
+	{
+		MetricsOpts: operatormetrics.MetricOpts{
+			Name: "kubevirt_cdi_import_pods_high_restart",
+			Help: "The number of CDI import pods with high restart count",
+		},
+		MetricType: operatormetrics.GaugeType,
+		Expr: intstr.FromString(
+			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'%s-.*', container='%s'} > %s) or on() vector(0)", common.ImporterPodName, common.ImporterPodName, strconv.Itoa(common.UnusualRestartCountThreshold)),
+		),
+	},
+	{
+		MetricsOpts: operatormetrics.MetricOpts{
+			Name: "kubevirt_cdi_upload_pods_high_restart",
+			Help: "The number of CDI upload server pods with high restart count",
+		},
+		MetricType: operatormetrics.GaugeType,
+		Expr: intstr.FromString(
+			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'%s-.*', container='%s'} > %s) or on() vector(0)", common.UploadPodName, common.UploadServerPodname, strconv.Itoa(common.UnusualRestartCountThreshold)),
+		),
+	},
+	{
+		MetricsOpts: operatormetrics.MetricOpts{
+			Name: "kubevirt_cdi_clone_pods_high_restart",
+			Help: "The number of CDI clone pods with high restart count",
+		},
+		MetricType: operatormetrics.GaugeType,
+		Expr: intstr.FromString(
+			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'.*%s', container='%s'} > %s) or on() vector(0)", common.ClonerSourcePodNameSuffix, common.ClonerSourcePodName, strconv.Itoa(common.UnusualRestartCountThreshold)),
+		),
+	},
+}

pkg/monitoring/rules/recordingrules/recordingrules.go

@@ -1,59 +1,11 @@
 package recordingrules
 
-import (
-	"fmt"
-	"strconv"
+import "github.com/machadovilaca/operator-observability/pkg/operatorrules"
 
-	"kubevirt.io/containerized-data-importer/pkg/common"
-)
-
-// MetricOpts represent CDI Prometheus Metrics
-type MetricOpts struct {
-	Name string
-	Help string
-	Type string
-}
-
-// RecordRulesDesc represent CDI Prometheus Record Rules
-type RecordRulesDesc struct {
-	Opts MetricOpts
-	Expr string
-}
-
-// GetRecordRulesDesc returns CDI Prometheus Record Rules
-func GetRecordRulesDesc(namespace string) []RecordRulesDesc {
-	return []RecordRulesDesc{
-		{
-			MetricOpts{
-				"kubevirt_cdi_operator_up",
-				"CDI operator status",
-				"Gauge",
-			},
-			fmt.Sprintf("sum(up{namespace='%s', pod=~'cdi-operator-.*'} or vector(0))", namespace),
-		},
-		{
-			MetricOpts{
-				"kubevirt_cdi_import_pods_high_restart",
-				"The number of CDI import pods with high restart count",
-				"Gauge",
-			},
-			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'%s-.*', container='%s'} > %s) or on() vector(0)", common.ImporterPodName, common.ImporterPodName, strconv.Itoa(common.UnusualRestartCountThreshold)),
-		},
-		{
-			MetricOpts{
-				"kubevirt_cdi_upload_pods_high_restart",
-				"The number of CDI upload server pods with high restart count",
-				"Gauge",
-			},
-			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'%s-.*', container='%s'} > %s) or on() vector(0)", common.UploadPodName, common.UploadServerPodname, strconv.Itoa(common.UnusualRestartCountThreshold)),
-		},
-		{
-			MetricOpts{
-				"kubevirt_cdi_clone_pods_high_restart",
-				"The number of CDI clone pods with high restart count",
-				"Gauge",
-			},
-			fmt.Sprintf("count(kube_pod_container_status_restarts_total{pod=~'.*%s', container='%s'} > %s) or on() vector(0)", common.ClonerSourcePodNameSuffix, common.ClonerSourcePodName, strconv.Itoa(common.UnusualRestartCountThreshold)),
-		},
-	}
+// Register sets up recording rules in the given namespace.
+func Register(namespace string) error {
+	return operatorrules.RegisterRecordingRules(
+		operatorRecordingRules(namespace),
+		podsRecordingRules,
+	)
 }

pkg/monitoring/rules/rules.go

@@ -0,0 +1,49 @@
+package rules
+
+import (
+	"github.com/machadovilaca/operator-observability/pkg/operatorrules"
+	promv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
+
+	"kubevirt.io/containerized-data-importer/pkg/common"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/alerts"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/recordingrules"
+)
+
+const (
+	ruleName = "prometheus-cdi-rules"
+)
+
+// SetupRules initializes recording and alert rules in a namespace.
+func SetupRules(namespace string) error {
+	if err := recordingrules.Register(namespace); err != nil {
+		return err
+	}
+
+	if err := alerts.Register(namespace); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+// BuildPrometheusRule creates a PrometheusRule in a namespace.
+func BuildPrometheusRule(namespace string) (*promv1.PrometheusRule, error) {
+	return operatorrules.BuildPrometheusRule(
+		ruleName,
+		namespace,
+		map[string]string{
+			common.CDIComponentLabel:  "",
+			common.PrometheusLabelKey: common.PrometheusLabelValue,
+		},
+	)
+}
+
+// ListRecordingRules returns all configured recording rules.
+func ListRecordingRules() []operatorrules.RecordingRule {
+	return operatorrules.ListRecordingRules()
+}
+
+// ListAlerts returns all configured alert rules.
+func ListAlerts() []promv1.Rule {
+	return operatorrules.ListAlerts()
+}

pkg/monitoring/tools/metricsdocs/metricsdocs.go

@@ -1,81 +0,0 @@
-package main
-
-import (
-	"fmt"
-
-	"github.com/machadovilaca/operator-observability/pkg/docs"
-	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"
-	"github.com/machadovilaca/operator-observability/pkg/operatorrules"
-	cdiMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-controller"
-	operatorMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/operator-controller"
-	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/recordingrules"
-)
-
-const tpl = `# Containerized Data Importer metrics
-{{- range . }}
-
-{{ $deprecatedVersion := "" -}}
-{{- with index .ExtraFields "DeprecatedVersion" -}}
-    {{- $deprecatedVersion = printf " in %s" . -}}
-{{- end -}}
-
-{{- $stabilityLevel := "" -}}
-{{- if and (.ExtraFields.StabilityLevel) (ne .ExtraFields.StabilityLevel "STABLE") -}}
-	{{- $stabilityLevel = printf "[%s%s] " .ExtraFields.StabilityLevel $deprecatedVersion -}}
-{{- end -}}
-
-### {{ .Name }}
-{{ print $stabilityLevel }}{{ .Help }}. Type: {{ .Type -}}.
-
-{{- end }}
-
-## Developing new metrics
-
-All metrics documented here are auto-generated and reflect exactly what is being
-exposed. After developing new metrics or changing old ones please regenerate
-this document.
-`
-
-func main() {
-	err := operatorMetrics.SetupMetrics()
-	if err != nil {
-		panic(err)
-	}
-
-	err = cdiMetrics.SetupMetrics()
-	if err != nil {
-		panic(err)
-	}
-
-	metricsList := operatorMetrics.ListMetrics()
-	recordingRulesList := convertToRecordingRules(recordingrules.GetRecordRulesDesc(""))
-
-	docsString := docs.BuildMetricsDocsWithCustomTemplate(metricsList, recordingRulesList, tpl)
-	fmt.Print(docsString)
-}
-
-func convertToRecordingRules(recordRulesDesc []recordingrules.RecordRulesDesc) []operatorrules.RecordingRule {
-	var recordingRules []operatorrules.RecordingRule
-
-	for _, ruleDesc := range recordRulesDesc {
-		recordingRule := operatorrules.RecordingRule{
-			MetricsOpts: operatormetrics.MetricOpts{
-				Name: ruleDesc.Opts.Name,
-				Help: ruleDesc.Opts.Help,
-				// Assuming the rest of the fields are correctly mapped
-			},
-			MetricType: convertRulesType(ruleDesc.Opts.Type),
-		}
-		recordingRules = append(recordingRules, recordingRule)
-	}
-	return recordingRules
-}
-
-// when adding new recording rule please note that
-func convertRulesType(metricType string) operatormetrics.MetricType {
-	if metricType == "Gauge" {
-		return operatormetrics.GaugeType
-		// ... other cases ...
-	}
-	return ""
-}

pkg/operator/controller/BUILD.bazel

@@ -24,7 +24,7 @@ go_library(
         "//pkg/controller/common:go_default_library",
         "//pkg/feature-gates:go_default_library",
         "//pkg/monitoring/metrics/operator-controller:go_default_library",
-        "//pkg/monitoring/rules/recordingrules:go_default_library",
+        "//pkg/monitoring/rules:go_default_library",
         "//pkg/operator:go_default_library",
         "//pkg/operator/resources/cert:go_default_library",
         "//pkg/operator/resources/cluster:go_default_library",
@@ -59,7 +59,6 @@ go_library(
         "//vendor/k8s.io/apimachinery/pkg/runtime:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/runtime/schema:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/types:go_default_library",
-        "//vendor/k8s.io/apimachinery/pkg/util/intstr:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/util/strategicpatch:go_default_library",
         "//vendor/k8s.io/apiserver/pkg/authentication/user:go_default_library",
         "//vendor/k8s.io/client-go/kubernetes:go_default_library",
@@ -94,6 +93,8 @@ go_test(
     embed = [":go_default_library"],
     deps = [
         "//pkg/common:go_default_library",
+        "//pkg/monitoring/rules:go_default_library",
+        "//pkg/monitoring/rules/alerts:go_default_library",
         "//pkg/operator/resources/cert:go_default_library",
         "//pkg/operator/resources/cluster:go_default_library",
         "//pkg/operator/resources/namespaced:go_default_library",

pkg/operator/controller/controller.go

@@ -40,6 +40,7 @@ import (
 
 	cdiv1 "kubevirt.io/containerized-data-importer-api/pkg/apis/core/v1beta1"
 	metrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/operator-controller"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules"
 	"kubevirt.io/containerized-data-importer/pkg/operator"
 	cdicerts "kubevirt.io/containerized-data-importer/pkg/operator/resources/cert"
 	cdicluster "kubevirt.io/containerized-data-importer/pkg/operator/resources/cluster"
@@ -129,6 +130,11 @@ func newReconciler(mgr manager.Manager) (*ReconcileCDI, error) {
 		return nil, err
 	}
 
+	err = rules.SetupRules(namespace)
+	if err != nil {
+		return nil, err
+	}
+
 	recorder := mgr.GetEventRecorderFor("operator-controller")
 
 	r := &ReconcileCDI{

pkg/operator/controller/controller_test.go

@@ -25,9 +25,8 @@ import (
 	"strconv"
 	"strings"
 
-	"kubevirt.io/controller-lifecycle-operator-sdk/pkg/sdk/callbacks"
-
 	sdkapi "kubevirt.io/controller-lifecycle-operator-sdk/api"
+	"kubevirt.io/controller-lifecycle-operator-sdk/pkg/sdk/callbacks"
 	sdkr "kubevirt.io/controller-lifecycle-operator-sdk/pkg/sdk/reconciler"
 
 	. "github.com/onsi/ginkgo/v2"
@@ -57,6 +56,8 @@ import (
 
 	cdiv1 "kubevirt.io/containerized-data-importer-api/pkg/apis/core/v1beta1"
 	"kubevirt.io/containerized-data-importer/pkg/common"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/alerts"
 	clusterResources "kubevirt.io/containerized-data-importer/pkg/operator/resources/cluster"
 	namespaceResources "kubevirt.io/containerized-data-importer/pkg/operator/resources/namespaced"
 	utils "kubevirt.io/containerized-data-importer/pkg/operator/resources/utils"
@@ -285,7 +286,7 @@ var _ = Describe("Controller", func() {
 				doReconcile(args)
 				Expect(setDeploymentsReady(args)).To(BeTrue())
 
-				runbookURLTemplate := getRunbookURLTemplate()
+				runbookURLTemplate := alerts.GetRunbookURLTemplate()
 
 				rule := &promv1.PrometheusRule{
 					ObjectMeta: metav1.ObjectMeta{
@@ -313,7 +314,7 @@ var _ = Describe("Controller", func() {
 					},
 				}
 
-				Expect(rule.Spec.Groups[0].Rules).To(ContainElement(cdiDownAlert))
+				Expect(rule.Spec.Groups[1].Rules).To(ContainElement(cdiDownAlert))
 				Expect(rule.Labels[common.AppKubernetesPartOfLabel]).To(Equal("testing"))
 				validateEvents(args.reconciler, createReadyEventValidationMap())
 			})
@@ -1737,6 +1738,11 @@ func createReconciler(client client.Client) *ReconcileCDI {
 		Namespace:              namespace,
 	}
 
+	err := rules.SetupRules(namespace)
+	if err != nil {
+		panic(err)
+	}
+
 	recorder := record.NewFakeRecorder(250)
 	r := &ReconcileCDI{
 		client:         client,

pkg/operator/controller/prometheus.go

@@ -18,11 +18,9 @@ package controller
 
 import (
 	"context"
-	"errors"
 	"fmt"
 	"os"
 	"reflect"
-	"strings"
 
 	"github.com/go-logr/logr"
 	promv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
@@ -31,14 +29,13 @@ import (
 	"k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
-	"k8s.io/apimachinery/pkg/util/intstr"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
 	"sigs.k8s.io/controller-runtime/pkg/source"
 
 	"kubevirt.io/containerized-data-importer/pkg/common"
 	cc "kubevirt.io/containerized-data-importer/pkg/controller/common"
-	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/recordingrules"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules"
 	cdinamespaced "kubevirt.io/containerized-data-importer/pkg/operator/resources/namespaced"
 	"kubevirt.io/containerized-data-importer/pkg/util"
 
@@ -52,12 +49,6 @@ const (
 	defaultMonitoringNs       = "monitoring"
 	defaultRunbookURLTemplate = "https://kubevirt.io/monitoring/runbooks/%s"
 	runbookURLTemplateEnv     = "RUNBOOK_URL_TEMPLATE"
-	severityAlertLabelKey     = "severity"
-	healthImpactAlertLabelKey = "operator_health_impact"
-	partOfAlertLabelKey       = "kubernetes_operator_part_of"
-	partOfAlertLabelValue     = "kubevirt"
-	componentAlertLabelKey    = "kubernetes_operator_component"
-	componentAlertLabelValue  = common.CDILabelValue
 )
 
 func ensurePrometheusResourcesExist(ctx context.Context, c client.Client, scheme *runtime.Scheme, owner metav1.Object) error {
@@ -134,166 +125,15 @@ func isPrometheusDeployed(logger logr.Logger, c client.Client, namespace string)
 	return true, nil
 }
 
-func getRecordRules(namespace string) []promv1.Rule {
-	var recordRules []promv1.Rule
-
-	for _, rrd := range recordingrules.GetRecordRulesDesc(namespace) {
-		recordRules = append(recordRules, generateRecordRule(rrd.Opts.Name, rrd.Expr))
-	}
-
-	return recordRules
-}
-
-func getAlertRules(runbookURLTemplate string) []promv1.Rule {
-	return []promv1.Rule{
-		generateAlertRule(
-			"CDIOperatorDown",
-			"kubevirt_cdi_operator_up == 0",
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "CDI operator is down",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIOperatorDown"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "critical",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDINotReady",
-			"kubevirt_cdi_cr_ready == 0",
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "CDI is not available to use",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDINotReady"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "critical",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDIDataVolumeUnusualRestartCount",
-			`kubevirt_cdi_import_pods_high_restart > 0 or
-			kubevirt_cdi_upload_pods_high_restart > 0 or
-			kubevirt_cdi_clone_pods_high_restart > 0`,
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "Some CDI population workloads have an unusual restart count, meaning they are probably failing and need to be investigated",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIDataVolumeUnusualRestartCount"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "warning",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDIStorageProfilesIncomplete",
-			`sum by(storageclass,provisioner) ((kubevirt_cdi_storageprofile_info{complete="false"}>0))`,
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "Incomplete StorageProfile {{ $labels.storageclass }}, accessMode/volumeMode cannot be inferred by CDI for PVC population request",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIStorageProfilesIncomplete"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "info",
-				healthImpactAlertLabelKey: "warning",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDIDataImportCronOutdated",
-			`sum by(ns,cron_name) (kubevirt_cdi_dataimportcron_outdated) > 0`,
-			promv1.Duration("15m"),
-			map[string]string{
-				"summary":     "DataImportCron (recurring polling of VM templates disk image sources, also known as golden images) PVCs are not being updated on the defined schedule",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIDataImportCronOutdated"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "info",
-				healthImpactAlertLabelKey: "warning",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDINoDefaultStorageClass",
-			`sum(kubevirt_cdi_storageprofile_info{default="true"} or on() vector(0)) +
-			sum(kubevirt_cdi_storageprofile_info{virtdefault="true"} or on() vector(0)) +
-			(count(kubevirt_cdi_datavolume_pending == 0) or on() vector(0)) == 0`,
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "No default StorageClass or virtualization StorageClass, and a DataVolume is pending for one",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDINoDefaultStorageClass"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "none",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDIMultipleDefaultVirtStorageClasses",
-			`sum(kubevirt_cdi_storageprofile_info{virtdefault="true"} or on() vector(0)) > 1`,
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "More than one default virtualization StorageClass detected",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIMultipleDefaultVirtStorageClasses"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "none",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-		generateAlertRule(
-			"CDIDefaultStorageClassDegraded",
-			`sum(kubevirt_cdi_storageprofile_info{default="true",rwx="true",smartclone="true"} or on() vector(0)) +
-			sum(kubevirt_cdi_storageprofile_info{virtdefault="true",rwx="true",smartclone="true"} or on() vector(0)) == 0`,
-			promv1.Duration("5m"),
-			map[string]string{
-				"summary":     "Default storage class has no smart clone or ReadWriteMany",
-				"runbook_url": fmt.Sprintf(runbookURLTemplate, "CDIDefaultStorageClassDegraded"),
-			},
-			map[string]string{
-				severityAlertLabelKey:     "warning",
-				healthImpactAlertLabelKey: "none",
-				partOfAlertLabelKey:       partOfAlertLabelValue,
-				componentAlertLabelKey:    componentAlertLabelValue,
-			},
-		),
-	}
-}
-
 func newPrometheusRule(namespace string) *promv1.PrometheusRule {
-	runbookURLTemplate := getRunbookURLTemplate()
+	promRule, err := rules.BuildPrometheusRule(namespace)
+	if err != nil {
+		panic(err)
+	}
 
 	return &promv1.PrometheusRule{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      ruleName,
-			Namespace: namespace,
-			Labels: map[string]string{
-				common.CDIComponentLabel:  "",
-				common.PrometheusLabelKey: common.PrometheusLabelValue,
-			},
-		},
-		Spec: promv1.PrometheusRuleSpec{
-			Groups: []promv1.RuleGroup{
-				{
-					Name:  "cdi.rules",
-					Rules: append(getRecordRules(namespace), getAlertRules(runbookURLTemplate)...),
-				},
-			},
-		},
+		ObjectMeta: promRule.ObjectMeta,
+		Spec:       promRule.Spec,
 	}
 }
 
@@ -381,23 +221,6 @@ func newPrometheusServiceMonitor(namespace string) *promv1.ServiceMonitor {
 	}
 }
 
-func generateAlertRule(alert, expr string, duration promv1.Duration, annotations, labels map[string]string) promv1.Rule {
-	return promv1.Rule{
-		Alert:       alert,
-		Expr:        intstr.FromString(expr),
-		For:         &duration,
-		Annotations: annotations,
-		Labels:      labels,
-	}
-}
-
-func generateRecordRule(record, expr string) promv1.Rule {
-	return promv1.Rule{
-		Record: record,
-		Expr:   intstr.FromString(expr),
-	}
-}
-
 func (r *ReconcileCDI) watchPrometheusResources() error {
 	listObjs := []client.ObjectList{
 		&promv1.PrometheusRuleList{},
@@ -439,16 +262,3 @@ func (r *ReconcileCDI) watchPrometheusResources() error {
 
 	return nil
 }
-
-func getRunbookURLTemplate() string {
-	runbookURLTemplate, exists := os.LookupEnv(runbookURLTemplateEnv)
-	if !exists {
-		runbookURLTemplate = defaultRunbookURLTemplate
-	}
-
-	if strings.Count(runbookURLTemplate, "%s") != 1 {
-		panic(errors.New("runbook URL template must have exactly 1 %s substring"))
-	}
-
-	return runbookURLTemplate
-}

tools/metricsdocs/BUILD.bazel

@@ -3,15 +3,13 @@ load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_library")
 go_library(
     name = "go_default_library",
     srcs = ["metricsdocs.go"],
-    importpath = "kubevirt.io/containerized-data-importer/pkg/monitoring/tools/metricsdocs",
+    importpath = "kubevirt.io/containerized-data-importer/tools/metricsdocs",
     visibility = ["//visibility:private"],
     deps = [
         "//pkg/monitoring/metrics/cdi-controller:go_default_library",
         "//pkg/monitoring/metrics/operator-controller:go_default_library",
-        "//pkg/monitoring/rules/recordingrules:go_default_library",
+        "//pkg/monitoring/rules:go_default_library",
         "//vendor/github.com/machadovilaca/operator-observability/pkg/docs:go_default_library",
-        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatormetrics:go_default_library",
-        "//vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules:go_default_library",
     ],
 )
 

tools/metricsdocs/metricsdocs.go

@@ -0,0 +1,56 @@
+package main
+
+import (
+	"fmt"
+
+	"github.com/machadovilaca/operator-observability/pkg/docs"
+
+	cdiMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-controller"
+	operatorMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/operator-controller"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules"
+)
+
+const tpl = `# Containerized Data Importer metrics
+{{- range . }}
+
+{{ $deprecatedVersion := "" -}}
+{{- with index .ExtraFields "DeprecatedVersion" -}}
+    {{- $deprecatedVersion = printf " in %s" . -}}
+{{- end -}}
+
+{{- $stabilityLevel := "" -}}
+{{- if and (.ExtraFields.StabilityLevel) (ne .ExtraFields.StabilityLevel "STABLE") -}}
+	{{- $stabilityLevel = printf "[%s%s] " .ExtraFields.StabilityLevel $deprecatedVersion -}}
+{{- end -}}
+
+### {{ .Name }}
+{{ print $stabilityLevel }}{{ .Help }}. Type: {{ .Type -}}.
+
+{{- end }}
+
+## Developing new metrics
+
+All metrics documented here are auto-generated and reflect exactly what is being
+exposed. After developing new metrics or changing old ones please regenerate
+this document.
+`
+
+func main() {
+	err := operatorMetrics.SetupMetrics()
+	if err != nil {
+		panic(err)
+	}
+
+	err = cdiMetrics.SetupMetrics()
+	if err != nil {
+		panic(err)
+	}
+
+	if err := rules.SetupRules("test"); err != nil {
+		panic(err)
+	}
+
+	docsString := docs.BuildMetricsDocsWithCustomTemplate(operatorMetrics.ListMetrics(), rules.ListRecordingRules(), tpl)
+
+	fmt.Print(docsString)
+}

tools/prom-metrics-collector/BUILD.bazel

@@ -3,12 +3,12 @@ load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_library")
 go_library(
     name = "go_default_library",
     srcs = ["metrics_json_generator.go"],
-    importpath = "kubevirt.io/containerized-data-importer/pkg/monitoring/tools/prom-metrics-collector",
+    importpath = "kubevirt.io/containerized-data-importer/tools/prom-metrics-collector",
     visibility = ["//visibility:private"],
     deps = [
         "//pkg/monitoring/metrics/cdi-controller:go_default_library",
         "//pkg/monitoring/metrics/operator-controller:go_default_library",
-        "//pkg/monitoring/rules/recordingrules:go_default_library",
+        "//pkg/monitoring/rules:go_default_library",
         "//vendor/github.com/kubevirt/monitoring/pkg/metrics/parser:go_default_library",
     ],
 )

tools/prom-metrics-collector/metrics_json_generator.go

@@ -8,7 +8,7 @@ import (
 	"github.com/kubevirt/monitoring/pkg/metrics/parser"
 	cdiMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-controller"
 	operatorMetrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/operator-controller"
-	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules/recordingrules"
+	"kubevirt.io/containerized-data-importer/pkg/monitoring/rules"
 )
 
 // This should be used only for very rare cases where the naming conventions that are explained in the best practices:
@@ -27,6 +27,10 @@ func main() {
 		panic(err)
 	}
 
+	if err := rules.SetupRules("test"); err != nil {
+		panic(err)
+	}
+
 	var metricFamilies []parser.Metric
 
 	metricsList := operatorMetrics.ListMetrics()
@@ -40,13 +44,15 @@ func main() {
 		}
 	}
 
-	recordingRules := recordingrules.GetRecordRulesDesc("")
-	for _, r := range recordingRules {
-		metricFamilies = append(metricFamilies, parser.Metric{
-			Name: r.Opts.Name,
-			Help: r.Opts.Help,
-			Type: strings.ToUpper(r.Opts.Type),
-		})
+	rulesList := rules.ListRecordingRules()
+	for _, r := range rulesList {
+		if _, isExcludedMetric := excludedMetrics[r.GetOpts().Name]; !isExcludedMetric {
+			metricFamilies = append(metricFamilies, parser.Metric{
+				Name: r.GetOpts().Name,
+				Help: r.GetOpts().Help,
+				Type: strings.ToUpper(string(r.GetType())),
+			})
+		}
 	}
 
 	jsonBytes, err := json.Marshal(metricFamilies)

vendor/github.com/machadovilaca/operator-observability/pkg/docs/metrics.go

@@ -63,6 +63,8 @@ func BuildMetricsDocsWithCustomTemplate(
 		allDocs = append(allDocs, buildMetricsDocs(recordingRules)...)
 	}
 
+	sortMetricsDocs(allDocs)
+
 	buf := bytes.NewBufferString("")
 	err = tpl.Execute(buf, allDocs)
 	if err != nil {
@@ -89,7 +91,6 @@ func buildMetricsDocs[T docOptions](items []T) []metricDocs {
 			ExtraFields: metricOpts.ExtraFields,
 		}
 	}
-	sortMetricsDocs(metricsDocs)
 
 	return metricsDocs
 }

vendor/github.com/machadovilaca/operator-observability/pkg/operatormetrics/collector.go

@@ -2,6 +2,7 @@ package operatormetrics
 
 import (
 	"fmt"
+	"strings"
 
 	"github.com/prometheus/client_golang/prometheus"
 )
@@ -24,6 +25,16 @@ type CollectorResult struct {
 	Value  float64
 }
 
+func (c Collector) hash() string {
+	var sb strings.Builder
+
+	for _, cm := range c.Metrics {
+		sb.WriteString(cm.GetOpts().Name)
+	}
+
+	return sb.String()
+}
+
 func (c Collector) Describe(ch chan<- *prometheus.Desc) {
 	for _, cm := range c.Metrics {
 		cm.getCollector().Describe(ch)

vendor/github.com/machadovilaca/operator-observability/pkg/operatormetrics/registry.go

@@ -5,6 +5,10 @@ import (
 )
 
 type RegistryFunc func(c prometheus.Collector) error
+type UnregisterFunc func(c prometheus.Collector) bool
 
 // Register is the function used to register metrics and collectors by this package.
 var Register RegistryFunc = prometheus.Register
+
+// Unregister is the function used to unregister metrics and collectors by this package.
+var Unregister UnregisterFunc = prometheus.Unregister

vendor/github.com/machadovilaca/operator-observability/pkg/operatormetrics/wrapper_registry.go

@@ -1,15 +1,24 @@
 package operatormetrics
 
+import (
+	"cmp"
+	"fmt"
+	"slices"
+)
+
 var operatorRegistry = newRegistry()
 
 type operatorRegisterer struct {
-	registeredMetrics          map[string]Metric
+	registeredMetrics map[string]Metric
+
+	registeredCollectors       map[string]Collector
 	registeredCollectorMetrics map[string]Metric
 }
 
 func newRegistry() operatorRegisterer {
 	return operatorRegisterer{
 		registeredMetrics:          map[string]Metric{},
+		registeredCollectors:       map[string]Collector{},
 		registeredCollectorMetrics: map[string]Metric{},
 	}
 }
@@ -18,11 +27,17 @@ func newRegistry() operatorRegisterer {
 func RegisterMetrics(allMetrics ...[]Metric) error {
 	for _, metricList := range allMetrics {
 		for _, metric := range metricList {
-			err := Register(metric.getCollector())
+			if metricExists(metric) {
+				err := unregisterMetric(metric)
+				if err != nil {
+					return err
+				}
+			}
+
+			err := registerMetric(metric)
 			if err != nil {
 				return err
 			}
-			operatorRegistry.registeredMetrics[metric.GetOpts().Name] = metric
 		}
 	}
 
@@ -32,13 +47,16 @@ func RegisterMetrics(allMetrics ...[]Metric) error {
 // RegisterCollector registers the collector with the Prometheus registry.
 func RegisterCollector(collectors ...Collector) error {
 	for _, collector := range collectors {
-		err := Register(collector)
-		if err != nil {
-			return err
+		if collectorExists(collector) {
+			err := unregisterCollector(collector)
+			if err != nil {
+				return err
+			}
 		}
 
-		for _, cm := range collector.Metrics {
-			operatorRegistry.registeredCollectorMetrics[cm.GetOpts().Name] = cm
+		err := registerCollector(collector)
+		if err != nil {
+			return err
 		}
 	}
 
@@ -57,5 +75,92 @@ func ListMetrics() []Metric {
 		result = append(result, rc)
 	}
 
+	slices.SortFunc(result, func(a, b Metric) int {
+		return cmp.Compare(a.GetOpts().Name, b.GetOpts().Name)
+	})
+
 	return result
 }
+
+// CleanRegistry removes all registered metrics.
+func CleanRegistry() error {
+	for _, metric := range operatorRegistry.registeredMetrics {
+		err := unregisterMetric(metric)
+		if err != nil {
+			return err
+		}
+	}
+
+	for _, collector := range operatorRegistry.registeredCollectors {
+		err := unregisterCollector(collector)
+		if err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func metricExists(metric Metric) bool {
+	_, ok := operatorRegistry.registeredMetrics[metric.GetOpts().Name]
+	return ok
+}
+
+func unregisterMetric(metric Metric) error {
+	if succeeded := Unregister(metric.getCollector()); succeeded {
+		delete(operatorRegistry.registeredMetrics, metric.GetOpts().Name)
+		return nil
+	}
+
+	return fmt.Errorf("failed to unregister from Prometheus client metric %s", metric.GetOpts().Name)
+}
+
+func registerMetric(metric Metric) error {
+	err := Register(metric.getCollector())
+	if err != nil {
+		return err
+	}
+	operatorRegistry.registeredMetrics[metric.GetOpts().Name] = metric
+
+	return nil
+}
+
+func collectorExists(collector Collector) bool {
+	_, ok := operatorRegistry.registeredCollectors[collector.hash()]
+	return ok
+}
+
+func unregisterCollector(collector Collector) error {
+	if succeeded := Unregister(collector); succeeded {
+		delete(operatorRegistry.registeredCollectors, collector.hash())
+		for _, metric := range collector.Metrics {
+			delete(operatorRegistry.registeredCollectorMetrics, metric.GetOpts().Name)
+		}
+		return nil
+	}
+
+	return fmt.Errorf("failed to unregister from Prometheus client collector with metrics: %s", buildCollectorMetricListString(collector))
+}
+
+func registerCollector(collector Collector) error {
+	err := Register(collector)
+	if err != nil {
+		return err
+	}
+
+	operatorRegistry.registeredCollectors[collector.hash()] = collector
+	for _, cm := range collector.Metrics {
+		operatorRegistry.registeredCollectorMetrics[cm.GetOpts().Name] = cm
+	}
+
+	return nil
+}
+
+func buildCollectorMetricListString(collector Collector) string {
+	metricsList := ""
+	for _, metric := range collector.Metrics {
+		metricsList += metric.GetOpts().Name + ", "
+	}
+	metricsList = metricsList[:len(metricsList)-2]
+	return metricsList
+}

vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules/prometheusrules.go

@@ -1,7 +1,9 @@
 package operatorrules
 
 import (
+	"cmp"
 	"fmt"
+	"slices"
 
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
@@ -42,7 +44,7 @@ func buildPrometheusRuleSpec() (*promv1.PrometheusRuleSpec, error) {
 	if len(operatorRegistry.registeredAlerts) != 0 {
 		groups = append(groups, promv1.RuleGroup{
 			Name:  "alerts.rules",
-			Rules: buildAlertsRules(),
+			Rules: ListAlerts(),
 		})
 	}
 
@@ -63,11 +65,9 @@ func buildRecordingRulesRules() []promv1.Rule {
 		})
 	}
 
-	return rules
-}
+	slices.SortFunc(rules, func(a, b promv1.Rule) int {
+		return cmp.Compare(a.Record, b.Record)
+	})
 
-func buildAlertsRules() []promv1.Rule {
-	var rules []promv1.Rule
-	rules = append(rules, operatorRegistry.registeredAlerts...)
 	return rules
 }

vendor/github.com/machadovilaca/operator-observability/pkg/operatorrules/registry.go

@@ -1,26 +1,32 @@
 package operatorrules
 
 import (
+	"cmp"
+	"slices"
+
 	promv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
 )
 
 var operatorRegistry = newRegistry()
 
 type operatorRegisterer struct {
-	registeredRecordingRules []RecordingRule
-	registeredAlerts         []promv1.Rule
+	registeredRecordingRules map[string]RecordingRule
+	registeredAlerts         map[string]promv1.Rule
 }
 
 func newRegistry() operatorRegisterer {
 	return operatorRegisterer{
-		registeredRecordingRules: []RecordingRule{},
+		registeredRecordingRules: map[string]RecordingRule{},
+		registeredAlerts:         map[string]promv1.Rule{},
 	}
 }
 
 // RegisterRecordingRules registers the given recording rules.
 func RegisterRecordingRules(recordingRules ...[]RecordingRule) error {
 	for _, recordingRuleList := range recordingRules {
-		operatorRegistry.registeredRecordingRules = append(operatorRegistry.registeredRecordingRules, recordingRuleList...)
+		for _, recordingRule := range recordingRuleList {
+			operatorRegistry.registeredRecordingRules[recordingRule.MetricsOpts.Name] = recordingRule
+		}
 	}
 
 	return nil
@@ -29,7 +35,9 @@ func RegisterRecordingRules(recordingRules ...[]RecordingRule) error {
 // RegisterAlerts registers the given alerts.
 func RegisterAlerts(alerts ...[]promv1.Rule) error {
 	for _, alertList := range alerts {
-		operatorRegistry.registeredAlerts = append(operatorRegistry.registeredAlerts, alertList...)
+		for _, alert := range alertList {
+			operatorRegistry.registeredAlerts[alert.Alert] = alert
+		}
 	}
 
 	return nil
@@ -37,10 +45,34 @@ func RegisterAlerts(alerts ...[]promv1.Rule) error {
 
 // ListRecordingRules returns the registered recording rules.
 func ListRecordingRules() []RecordingRule {
-	return operatorRegistry.registeredRecordingRules
+	var rules []RecordingRule
+	for _, rule := range operatorRegistry.registeredRecordingRules {
+		rules = append(rules, rule)
+	}
+
+	slices.SortFunc(rules, func(a, b RecordingRule) int {
+		return cmp.Compare(a.GetOpts().Name, b.GetOpts().Name)
+	})
+
+	return rules
 }
 
 // ListAlerts returns the registered alerts.
 func ListAlerts() []promv1.Rule {
-	return operatorRegistry.registeredAlerts
+	var alerts []promv1.Rule
+	for _, alert := range operatorRegistry.registeredAlerts {
+		alerts = append(alerts, alert)
+	}
+
+	slices.SortFunc(alerts, func(a, b promv1.Rule) int {
+		return cmp.Compare(a.Alert, b.Alert)
+	})
+
+	return alerts
+}
+
+// CleanRegistry removes all registered rules and alerts.
+func CleanRegistry() error {
+	operatorRegistry = newRegistry()
+	return nil
 }

vendor/modules.txt

@@ -357,8 +357,8 @@ github.com/kubernetes-csi/lib-volume-populator/populator-machinery
 # github.com/kubevirt/monitoring/pkg/metrics/parser v0.0.0-20230627123556-81a891d4462a
 ## explicit; go 1.20
 github.com/kubevirt/monitoring/pkg/metrics/parser
-# github.com/machadovilaca/operator-observability v0.0.9
-## explicit; go 1.20
+# github.com/machadovilaca/operator-observability v0.0.13
+## explicit; go 1.21
 github.com/machadovilaca/operator-observability/pkg/docs
 github.com/machadovilaca/operator-observability/pkg/operatormetrics
 github.com/machadovilaca/operator-observability/pkg/operatorrules