Merge pull request #118986 from logicalhan/beta-metric

promote sli metrics to beta

staging/src/k8s.io/component-base/metrics/prometheus/slis/metrics.go

@@ -37,7 +37,7 @@ var (
 			Namespace:      "kubernetes",
 			Name:           "healthcheck",
 			Help:           "This metric records the result of a single healthcheck.",
-			StabilityLevel: k8smetrics.ALPHA,
+			StabilityLevel: k8smetrics.BETA,
 		},
 		[]string{"name", "type"},
 	)
@@ -48,7 +48,7 @@ var (
 			Namespace:      "kubernetes",
 			Name:           "healthchecks_total",
 			Help:           "This metric records the results of all healthcheck.",
-			StabilityLevel: k8smetrics.ALPHA,
+			StabilityLevel: k8smetrics.BETA,
 		},
 		[]string{"name", "type", "status"},
 	)

staging/src/k8s.io/component-base/metrics/prometheus/slis/metrics_test.go

@@ -37,10 +37,10 @@ func TestObserveHealthcheck(t *testing.T) {
 	initialState := Error
 	healthcheckName := "healthcheck-a"
 	initialOutput := `
-        # HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck.
+        # HELP kubernetes_healthcheck [BETA] This metric records the result of a single healthcheck.
         # TYPE kubernetes_healthcheck gauge
         kubernetes_healthcheck{name="healthcheck-a",type="healthz"} 0
-        # HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck.
+        # HELP kubernetes_healthchecks_total [BETA] This metric records the results of all healthcheck.
         # TYPE kubernetes_healthchecks_total counter
         kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1
 `
@@ -57,10 +57,10 @@ func TestObserveHealthcheck(t *testing.T) {
 			hcType:   "healthz",
 			hcStatus: Success,
 			want: `
-        # HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck.
+        # HELP kubernetes_healthcheck [BETA] This metric records the result of a single healthcheck.
         # TYPE kubernetes_healthcheck gauge
         kubernetes_healthcheck{name="healthcheck-a",type="healthz"} 1
-        # HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck.
+        # HELP kubernetes_healthchecks_total [BETA] This metric records the results of all healthcheck.
         # TYPE kubernetes_healthchecks_total counter
         kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1
         kubernetes_healthchecks_total{name="healthcheck-a",status="success",type="healthz"} 1

test/instrumentation/error.go

@@ -25,7 +25,7 @@ import (
 const (
 	errNotDirectCall                = "Opts for STABLE metric was not directly passed to new metric function"
 	errPositionalArguments          = "Positional arguments are not supported"
-	errStabilityLevel               = "StabilityLevel should be passed STABLE, ALPHA or removed"
+	errStabilityLevel               = "StabilityLevel should be passed STABLE, BETA, ALPHA or removed"
 	errInvalidNewMetricCall         = "Invalid new metric call, please ensure code compiles"
 	errNonStringAttribute           = "Non string attribute is not supported"
 	errBadVariableAttribute         = "Metric attribute was not correctly set. Please use only global consts in same file"

test/instrumentation/main.go

@@ -55,7 +55,7 @@ func main() {
 		fmt.Fprintf(os.Stderr, "USAGE: %s <DIR or FILE or '-'> [...]\n", os.Args[0])
 		os.Exit(64)
 	}
-
+	stableMetricNames := map[string]struct{}{}
 	stableMetrics := []metric{}
 	errors := []error{}
 
@@ -66,7 +66,12 @@ func main() {
 			continue
 		}
 		ms, es := searchPathForStableMetrics(arg)
-		stableMetrics = append(stableMetrics, ms...)
+		for _, m := range ms {
+			if _, ok := stableMetricNames[m.Name]; !ok {
+				stableMetrics = append(stableMetrics, m)
+			}
+			stableMetricNames[m.Name] = struct{}{}
+		}
 		errors = append(errors, es...)
 	}
 	if addStdin {

test/instrumentation/main_test.go

@@ -601,7 +601,7 @@ var _ = metrics.NewCounter(
 `},
 		{
 			testName: "Fail on metric with stability set to function return",
-			err:      fmt.Errorf("testdata/metric.go:9:20: StabilityLevel should be passed STABLE, ALPHA or removed"),
+			err:      fmt.Errorf("testdata/metric.go:9:20: %s", errStabilityLevel),
 			src: `
 package test
 import "k8s.io/component-base/metrics"
@@ -616,7 +616,7 @@ var _ = metrics.NewCounter(
 `},
 		{
 			testName: "error for passing stability as string",
-			err:      fmt.Errorf("testdata/metric.go:6:20: StabilityLevel should be passed STABLE, ALPHA or removed"),
+			err:      fmt.Errorf("testdata/metric.go:6:20: %s", errStabilityLevel),
 			src: `
 package test
 import "k8s.io/component-base/metrics"
@@ -628,7 +628,7 @@ var _ = metrics.NewCounter(
 `},
 		{
 			testName: "error for passing stability as variable",
-			err:      fmt.Errorf("testdata/metric.go:7:20: StabilityLevel should be passed STABLE, ALPHA or removed"),
+			err:      fmt.Errorf("testdata/metric.go:7:20: %s", errStabilityLevel),
 			src: `
 package test
 import "k8s.io/component-base/metrics"

test/instrumentation/testdata/pkg/kubelet/metrics/metrics.go

@@ -132,6 +132,27 @@ var (
 		},
 		testLabels,
 	)
+	// healthcheck is a Prometheus Gauge metrics used for recording the results of a k8s healthcheck.
+	healthcheck = metrics.NewGaugeVec(
+		&metrics.GaugeOpts{
+			Namespace:      "kubernetes",
+			Name:           "healthcheck",
+			Help:           "This metric records the result of a single healthcheck.",
+			StabilityLevel: metrics.BETA,
+		},
+		[]string{"name", "type"},
+	)
+
+	// healthchecksTotal is a Prometheus Counter metrics used for counting the results of a k8s healthcheck.
+	healthchecksTotal = metrics.NewCounterVec(
+		&metrics.CounterOpts{
+			Namespace:      "kubernetes",
+			Name:           "healthchecks_total",
+			Help:           "This metric records the results of all healthcheck.",
+			StabilityLevel: metrics.BETA,
+		},
+		[]string{"name", "type", "status"},
+	)
 	// PodWorkerDuration is a Histogram that tracks the duration (in seconds) in takes to sync a single pod.
 	// Broken down by the operation type.
 	SummaryMaxAge = metrics.NewSummary(
@@ -567,6 +588,8 @@ func Register(collectors ...metrics.StableCollector) {
 		legacyregistry.MustRegister(NodeName)
 		legacyregistry.MustRegister(PodWorkerDuration)
 		legacyregistry.MustRegister(PodStartDuration)
+		legacyregistry.MustRegister(healthcheck)
+		legacyregistry.MustRegister(healthchecksTotal)
 		legacyregistry.MustRegister(CgroupManagerDuration)
 		legacyregistry.MustRegister(PodWorkerStartDuration)
 		legacyregistry.MustRegister(ContainersPerPodCount)

test/instrumentation/testdata/stable-metrics-list.yaml

@@ -409,3 +409,20 @@
   stabilityLevel: STABLE
   labels:
   - resource
+- name: healthcheck
+  namespace: kubernetes
+  help: This metric records the result of a single healthcheck.
+  type: Gauge
+  stabilityLevel: BETA
+  labels:
+  - name
+  - type
+- name: healthchecks_total
+  namespace: kubernetes
+  help: This metric records the results of all healthcheck.
+  type: Counter
+  stabilityLevel: BETA
+  labels:
+  - name
+  - status
+  - type

test/instrumentation/testdata/test-stable-metrics-list.yaml

@@ -192,6 +192,23 @@
   labels:
   - namespace
   - persistentvolumeclaim
+- name: healthcheck
+  namespace: kubernetes
+  help: This metric records the result of a single healthcheck.
+  type: Gauge
+  stabilityLevel: BETA
+  labels:
+  - name
+  - type
+- name: healthchecks_total
+  namespace: kubernetes
+  help: This metric records the results of all healthcheck.
+  type: Counter
+  stabilityLevel: BETA
+  labels:
+  - name
+  - status
+  - type
 - name: priority_level_seat_utilization
   subsystem: subsystem
   namespace: namespace