Merge pull request #281 from projectsyn/add-cluster-tenant-info

Add Cluster and Tenant info metrics

metrics/cluster_info_collector.go

@@ -0,0 +1,136 @@
+package metrics
+
+import (
+	"context"
+	"fmt"
+	"regexp"
+	"slices"
+	"strings"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/log"
+
+	synv1alpha1 "github.com/projectsyn/lieutenant-operator/api/v1alpha1"
+)
+
+//+kubebuilder:rbac:groups=syn.tools,resources=clusters,verbs=get;list;watch
+//+kubebuilder:rbac:groups=syn.tools,resources=clusters/status,verbs=get
+
+var clusterInfoDesc = prometheus.NewDesc(
+	"syn_lieutenant_cluster_info",
+	"Cluster information metric.",
+	[]string{"cluster", "tenant", "display_name"},
+	nil,
+)
+
+var clusterFactsDesc = prometheus.NewDesc(
+	"syn_lieutenant_cluster_facts",
+	"Lieutenant cluster facts.",
+	[]string{"cluster", "tenant", "display_name"},
+	nil,
+)
+
+// commodore build info has dynamic labels
+func newClusterFactsDesc(lbls ...string) *prometheus.Desc {
+	return prometheus.NewDesc(
+		"syn_lieutenant_cluster_facts",
+		"Lieutenant cluster facts. Keys are normalized to be valid Prometheus labels.",
+		lbls,
+		nil,
+	)
+}
+
+// ClusterInfoCollector is a Prometheus collector that collects cluster info metrics.
+type ClusterInfoCollector struct {
+	Client client.Client
+
+	Namespace string
+}
+
+var _ prometheus.Collector = &ClusterInfoCollector{}
+
+// Describe implements prometheus.Collector.
+// Sends the descriptors of the metrics to the channel.
+func (*ClusterInfoCollector) Describe(chan<- *prometheus.Desc) {}
+
+// Collect implements prometheus.Collector.
+// Iterates over all clusters and sends cluster information for each cluster.
+func (m *ClusterInfoCollector) Collect(ch chan<- prometheus.Metric) {
+	ctx := context.Background()
+
+	cls := synv1alpha1.ClusterList{}
+	if err := m.Client.List(ctx, &cls, client.InNamespace(m.Namespace)); err != nil {
+		err := fmt.Errorf("failed to list clusters: %w", err)
+		ch <- prometheus.NewInvalidMetric(clusterInfoDesc, err)
+	}
+
+	for _, cl := range cls.Items {
+		ch <- prometheus.MustNewConstMetric(
+			clusterInfoDesc,
+			prometheus.GaugeValue,
+			1,
+			cl.Name, cl.Spec.TenantRef.Name, cl.Spec.DisplayName,
+		)
+
+		if err := clusterFacts(cl, ch); err != nil {
+			log.Log.Info("failed to collect cluster facts", "error", err)
+		}
+	}
+}
+
+// clusterFacts collects the facts of a cluster and sends them as Prometheus metrics.
+// The keys of the facts are normalized to be valid Prometheus labels.
+// If the first character of a key is an underscore or an invalid character it is replaced with "fact_".
+// If a key is empty it is replaced with "_empty".
+// If a key is in the protected list after normalizing it is prefixed with "orig_".
+// If a key is a duplicate after normalizing it is suffixed with "_<n>" where n is the number of duplicates.
+func clusterFacts(cl synv1alpha1.Cluster, ch chan<- prometheus.Metric) error {
+	rks, vs := pairs(cl.Spec.Facts)
+	ks := make([]string, len(rks))
+	for i, k := range rks {
+		ks[i] = normalizeLabelKey(k, []string{"cluster", "tenant"}, "fact_")
+	}
+	seen := make(map[string]int)
+	for i, k := range ks {
+		if _, ok := seen[k]; ok {
+			ks[i] = fmt.Sprintf("%s_%d", k, seen[k])
+		}
+		seen[k]++
+	}
+
+	m, err := prometheus.NewConstMetric(
+		newClusterFactsDesc(append([]string{"cluster", "tenant"}, ks...)...),
+		prometheus.GaugeValue,
+		1,
+		append([]string{cl.Name, cl.Spec.TenantRef.Name}, vs...)...,
+	)
+	if err != nil {
+		return fmt.Errorf("failed to create metric for cluster %q: %w", cl.Name, err)
+	}
+	ch <- m
+	return nil
+}
+
+// https://prometheus.io/docs/concepts/data_model/#metric-names-and-labels
+var validKeyCharacters = regexp.MustCompile(`(?:^[^a-zA-Z_]|[^a-zA-Z0-9_])`)
+
+// normalizeLabelKey normalizes a key to be a valid Prometheus metric name.
+// It replaces invalid characters with underscores and prefixes the key with the given prefix if it starts with an underscore character.
+// If the key is empty it returns "_empty".
+// If the key is in the protected list after normalizing it prefixes the key with "orig_".
+func normalizeLabelKey(key string, protected []string, prefixForUnderscore string) string {
+	if key == "" {
+		return "_empty"
+	}
+
+	key = validKeyCharacters.ReplaceAllLiteralString(key, "_")
+	if strings.HasPrefix(key, "_") {
+		key = prefixForUnderscore + key
+	}
+	if slices.Contains(protected, key) {
+		key = "orig_" + key
+	}
+
+	return key
+}

metrics/cluster_info_collector_test.go

@@ -0,0 +1,89 @@
+package metrics_test
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/prometheus/client_golang/prometheus/testutil"
+	"github.com/stretchr/testify/require"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	synv1alpha1 "github.com/projectsyn/lieutenant-operator/api/v1alpha1"
+	"github.com/projectsyn/lieutenant-operator/metrics"
+)
+
+func Test_ClusterInfoCollector(t *testing.T) {
+	namespace := "testns"
+	expectedMetricNames := []string{
+		"syn_lieutenant_cluster_info",
+		"syn_lieutenant_cluster_facts",
+	}
+
+	c := prepareClient(t,
+		&synv1alpha1.Cluster{
+			ObjectMeta: metav1.ObjectMeta{
+				Namespace: namespace,
+				Name:      "c-empty",
+			},
+		},
+		&synv1alpha1.Cluster{
+			ObjectMeta: metav1.ObjectMeta{
+				Namespace: namespace,
+				Name:      "c2",
+			},
+			Spec: synv1alpha1.ClusterSpec{
+				DisplayName: "Cluster 2",
+				TenantRef: corev1.LocalObjectReference{
+					Name: "t2",
+				},
+				Facts: map[string]string{
+					"key":                            "value",
+					"_key":                           "value",
+					"0key-duplicate-after-normalize": "value",
+					"1key-duplicate-after-normalize": "value",
+					"2key-duplicate-after-normalize": "value",
+					"key-with847_💩_â-invalid-chars":  "value",
+					"cluster":                        "value",
+					"tenant":                         "value",
+				},
+			},
+		},
+	)
+
+	subject := &metrics.ClusterInfoCollector{
+		Client: c,
+
+		Namespace: namespace,
+	}
+
+	metrics := `# HELP syn_lieutenant_cluster_facts Lieutenant cluster facts. Keys are normalized to be valid Prometheus labels.
+# TYPE syn_lieutenant_cluster_facts gauge
+syn_lieutenant_cluster_facts{cluster="c-empty",tenant=""} 1
+syn_lieutenant_cluster_facts{cluster="c2",fact__key="value",fact__key_duplicate_after_normalize="value",fact__key_duplicate_after_normalize_1="value",fact__key_duplicate_after_normalize_2="value",key="value",key_with847_____invalid_chars="value",orig_cluster="value",orig_tenant="value",tenant="t2"} 1
+# HELP syn_lieutenant_cluster_info Cluster information metric.
+# TYPE syn_lieutenant_cluster_info gauge
+syn_lieutenant_cluster_info{cluster="c-empty",display_name="",tenant=""} 1
+syn_lieutenant_cluster_info{cluster="c2",display_name="Cluster 2",tenant="t2"} 1
+`
+	require.NoError(t,
+		testutil.CollectAndCompare(subject, strings.NewReader(metrics), expectedMetricNames...),
+	)
+}
+
+func Test_ClusterInfoCollector_ListFail(t *testing.T) {
+	namespace := "testns"
+
+	listErr := errors.New("whoopsie daisy")
+
+	c := prepareFailingClient(t, listErr)
+
+	subject := &metrics.ClusterInfoCollector{
+		Client: c,
+
+		Namespace: namespace,
+	}
+
+	require.ErrorContains(t, testutil.CollectAndCompare(subject, strings.NewReader(``)), listErr.Error())
+}

metrics/tenant_info_collector.go

@@ -0,0 +1,57 @@
+package metrics
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+
+	synv1alpha1 "github.com/projectsyn/lieutenant-operator/api/v1alpha1"
+)
+
+//+kubebuilder:rbac:groups=syn.tools,resources=tenants,verbs=get;list;watch
+//+kubebuilder:rbac:groups=syn.tools,resources=tenants/status,verbs=get
+
+var tenantInfoDesc = prometheus.NewDesc(
+	"syn_lieutenant_tenant_info",
+	"Tenant information metric.",
+	[]string{"tenant", "display_name"},
+	nil,
+)
+
+// TenantInfoCollector is a Prometheus collector that collects tenant info metrics.
+type TenantInfoCollector struct {
+	Client client.Client
+
+	Namespace string
+}
+
+var _ prometheus.Collector = &TenantInfoCollector{}
+
+// Describe implements prometheus.Collector.
+// Sends the descriptors of the metrics to the channel.
+func (*TenantInfoCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- tenantInfoDesc
+}
+
+// Collect implements prometheus.Collector.
+// Iterates over all tenants and sends tenant information for each tenant.
+func (m *TenantInfoCollector) Collect(ch chan<- prometheus.Metric) {
+	ctx := context.Background()
+
+	cls := synv1alpha1.TenantList{}
+	if err := m.Client.List(ctx, &cls, client.InNamespace(m.Namespace)); err != nil {
+		err := fmt.Errorf("failed to list tenants: %w", err)
+		ch <- prometheus.NewInvalidMetric(tenantInfoDesc, err)
+	}
+
+	for _, cl := range cls.Items {
+		ch <- prometheus.MustNewConstMetric(
+			tenantInfoDesc,
+			prometheus.GaugeValue,
+			1,
+			cl.Name, cl.Spec.DisplayName,
+		)
+	}
+}

metrics/tenant_info_collector_test.go

@@ -0,0 +1,70 @@
+package metrics_test
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/prometheus/client_golang/prometheus/testutil"
+	"github.com/stretchr/testify/require"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	synv1alpha1 "github.com/projectsyn/lieutenant-operator/api/v1alpha1"
+	"github.com/projectsyn/lieutenant-operator/metrics"
+)
+
+func Test_TenantInfoCollector(t *testing.T) {
+	namespace := "testns"
+	expectedMetricNames := []string{
+		"syn_lieutenant_tenant_info",
+	}
+
+	c := prepareClient(t,
+		&synv1alpha1.Tenant{
+			ObjectMeta: metav1.ObjectMeta{
+				Namespace: namespace,
+				Name:      "t-empty",
+			},
+		},
+		&synv1alpha1.Tenant{
+			ObjectMeta: metav1.ObjectMeta{
+				Namespace: namespace,
+				Name:      "t2",
+			},
+			Spec: synv1alpha1.TenantSpec{
+				DisplayName: "Tenant 2",
+			},
+		},
+	)
+
+	subject := &metrics.TenantInfoCollector{
+		Client: c,
+
+		Namespace: namespace,
+	}
+
+	metrics := `# HELP syn_lieutenant_tenant_info Tenant information metric.
+# TYPE syn_lieutenant_tenant_info gauge
+syn_lieutenant_tenant_info{display_name="",tenant="t-empty"} 1
+syn_lieutenant_tenant_info{display_name="Tenant 2",tenant="t2"} 1
+`
+	require.NoError(t,
+		testutil.CollectAndCompare(subject, strings.NewReader(metrics), expectedMetricNames...),
+	)
+}
+
+func Test_TenantInfoCollector_ListFail(t *testing.T) {
+	namespace := "testns"
+
+	listErr := errors.New("whoopsie daisy")
+
+	c := prepareFailingClient(t, listErr)
+
+	subject := &metrics.TenantInfoCollector{
+		Client: c,
+
+		Namespace: namespace,
+	}
+
+	require.ErrorContains(t, testutil.CollectAndCompare(subject, strings.NewReader(``)), listErr.Error())
+}