Merge pull request #81573 from irajdeep/irajdeep/change_runningPod_runningContainer_metrics

Convert kubelet metrics(running_pod_count and running_container_count) from non-standard prometheus collectors to standard gauges

pkg/kubelet/metrics/metrics.go

@@ -461,6 +461,26 @@ var (
 		},
 		[]string{"runtime_handler"},
 	)
+
+	// RunningPodCount is a gauge that tracks the number of Pods currently running
+	RunningPodCount = metrics.NewGauge(
+		&metrics.GaugeOpts{
+			Subsystem:      KubeletSubsystem,
+			Name:           "running_pod_count",
+			Help:           "Number of pods currently running",
+			StabilityLevel: metrics.ALPHA,
+		},
+	)
+	// RunningContainerCount is a gauge that tracks the number of containers currently running
+	RunningContainerCount = metrics.NewGaugeVec(
+		&metrics.GaugeOpts{
+			Subsystem:      KubeletSubsystem,
+			Name:           "running_container_count",
+			Help:           "Number of containers currently running",
+			StabilityLevel: metrics.ALPHA,
+		},
+		[]string{"container_state"},
+	)
 )
 
 var registerMetrics sync.Once
@@ -475,7 +495,6 @@ func Register(containerCache kubecontainer.RuntimeCache, collectors ...prometheu
 		legacyregistry.MustRegister(CgroupManagerDuration)
 		legacyregistry.MustRegister(PodWorkerStartDuration)
 		legacyregistry.MustRegister(ContainersPerPodCount)
-		legacyregistry.RawMustRegister(newPodAndContainerCollector(containerCache))
 		legacyregistry.MustRegister(PLEGRelistDuration)
 		legacyregistry.MustRegister(PLEGDiscardEvents)
 		legacyregistry.MustRegister(PLEGRelistInterval)
@@ -498,6 +517,8 @@ func Register(containerCache kubecontainer.RuntimeCache, collectors ...prometheu
 		legacyregistry.MustRegister(DeprecatedEvictionStatsAge)
 		legacyregistry.MustRegister(DeprecatedDevicePluginRegistrationCount)
 		legacyregistry.MustRegister(DeprecatedDevicePluginAllocationLatency)
+		legacyregistry.MustRegister(RunningContainerCount)
+		legacyregistry.MustRegister(RunningPodCount)
 		if utilfeature.DefaultFeatureGate.Enabled(features.DynamicKubeletConfig) {
 			legacyregistry.MustRegister(AssignedConfig)
 			legacyregistry.MustRegister(ActiveConfig)
@@ -520,60 +541,6 @@ func SinceInSeconds(start time.Time) float64 {
 	return time.Since(start).Seconds()
 }
 
-func newPodAndContainerCollector(containerCache kubecontainer.RuntimeCache) *podAndContainerCollector {
-	return &podAndContainerCollector{
-		containerCache: containerCache,
-	}
-}
-
-// Custom collector for current pod and container counts.
-type podAndContainerCollector struct {
-	// Cache for accessing information about running containers.
-	containerCache kubecontainer.RuntimeCache
-}
-
-// TODO(vmarmol): Split by source?
-var (
-	runningPodCountDesc = prometheus.NewDesc(
-		prometheus.BuildFQName("", KubeletSubsystem, "running_pod_count"),
-		"Number of pods currently running",
-		nil, nil)
-	runningContainerCountDesc = prometheus.NewDesc(
-		prometheus.BuildFQName("", KubeletSubsystem, "running_container_count"),
-		"Number of containers currently running",
-		nil, nil)
-)
-
-// Describe implements Prometheus' Describe method from the Collector interface. It sends all
-// available descriptions to the provided channel and retunrs once the last description has been sent.
-func (pc *podAndContainerCollector) Describe(ch chan<- *prometheus.Desc) {
-	ch <- runningPodCountDesc
-	ch <- runningContainerCountDesc
-}
-
-// Collect implements Prometheus' Collect method from the Collector interface. It's called by the Prometheus
-// registry when collecting metrics.
-func (pc *podAndContainerCollector) Collect(ch chan<- prometheus.Metric) {
-	runningPods, err := pc.containerCache.GetPods()
-	if err != nil {
-		klog.Warningf("Failed to get running container information while collecting metrics: %v", err)
-		return
-	}
-
-	runningContainers := 0
-	for _, p := range runningPods {
-		runningContainers += len(p.Containers)
-	}
-	ch <- prometheus.MustNewConstMetric(
-		runningPodCountDesc,
-		prometheus.GaugeValue,
-		float64(len(runningPods)))
-	ch <- prometheus.MustNewConstMetric(
-		runningContainerCountDesc,
-		prometheus.GaugeValue,
-		float64(runningContainers))
-}
-
 const configMapAPIPathFmt = "/api/v1/namespaces/%s/configmaps/%s"
 
 func configLabels(source *corev1.NodeConfigSource) (map[string]string, error) {

pkg/kubelet/pleg/BUILD

@@ -33,9 +33,11 @@ go_test(
     deps = [
         "//pkg/kubelet/container:go_default_library",
         "//pkg/kubelet/container/testing:go_default_library",
+        "//pkg/kubelet/metrics:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/types:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/util/clock:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/util/diff:go_default_library",
+        "//vendor/github.com/prometheus/client_model/go:go_default_library",
         "//vendor/github.com/stretchr/testify/assert:go_default_library",
     ],
 )

pkg/kubelet/pleg/generic.go

@@ -209,6 +209,8 @@ func (g *GenericPLEG) relist() {
 	g.updateRelistTime(timestamp)
 
 	pods := kubecontainer.Pods(podList)
+	// update running pod and container count
+	updateRunningPodAndContainerMetrics(pods)
 	g.podRecords.setCurrent(pods)
 
 	// Compare the old and the current pods, and generate events.
@@ -431,6 +433,24 @@ func getContainerState(pod *kubecontainer.Pod, cid *kubecontainer.ContainerID) p
 	return state
 }
 
+func updateRunningPodAndContainerMetrics(pods []*kubecontainer.Pod) {
+	// Set the number of running pods in the parameter
+	metrics.RunningPodCount.Set(float64(len(pods)))
+	// intermediate map to store the count of each "container_state"
+	containerStateCount := make(map[string]int)
+
+	for _, pod := range pods {
+		containers := pod.Containers
+		for _, container := range containers {
+			// update the corresponding "container_state" in map to set value for the gaugeVec metrics
+			containerStateCount[string(container.State)]++
+		}
+	}
+	for key, value := range containerStateCount {
+		metrics.RunningContainerCount.WithLabelValues(key).Set(float64(value))
+	}
+}
+
 func (pr podRecords) getOld(id types.UID) *kubecontainer.Pod {
 	r, ok := pr[id]
 	if !ok {

pkg/kubelet/pleg/generic_test.go

@@ -24,12 +24,14 @@ import (
 	"testing"
 	"time"
 
+	dto "github.com/prometheus/client_model/go"
 	"github.com/stretchr/testify/assert"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/apimachinery/pkg/util/clock"
 	"k8s.io/apimachinery/pkg/util/diff"
 	kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
 	containertest "k8s.io/kubernetes/pkg/kubelet/container/testing"
+	"k8s.io/kubernetes/pkg/kubelet/metrics"
 )
 
 const (
@@ -643,3 +645,49 @@ func TestRelistIPChange(t *testing.T) {
 		assert.Exactly(t, []*PodLifecycleEvent{event}, actualEvents)
 	}
 }
+
+func TestRunningPodAndContainerCount(t *testing.T) {
+	fakeRuntime := &containertest.FakeRuntime{}
+	runtimeCache, _ := kubecontainer.NewRuntimeCache(fakeRuntime)
+	metrics.Register(runtimeCache)
+	testPleg := newTestGenericPLEG()
+	pleg, runtime := testPleg.pleg, testPleg.runtime
+
+	runtime.AllPodList = []*containertest.FakePod{
+		{Pod: &kubecontainer.Pod{
+			ID: "1234",
+			Containers: []*kubecontainer.Container{
+				createTestContainer("c1", kubecontainer.ContainerStateRunning),
+				createTestContainer("c2", kubecontainer.ContainerStateUnknown),
+				createTestContainer("c3", kubecontainer.ContainerStateUnknown),
+			},
+		}},
+		{Pod: &kubecontainer.Pod{
+			ID: "4567",
+			Containers: []*kubecontainer.Container{
+				createTestContainer("c1", kubecontainer.ContainerStateExited),
+			},
+		}},
+	}
+
+	pleg.relist()
+
+	// assert for container count with label "running"
+	actualMetricRunningContainerCount := &dto.Metric{}
+	expectedMetricRunningContainerCount := float64(1)
+	metrics.RunningContainerCount.WithLabelValues(string(kubecontainer.ContainerStateRunning)).Write(actualMetricRunningContainerCount)
+	assert.Equal(t, expectedMetricRunningContainerCount, actualMetricRunningContainerCount.GetGauge().GetValue())
+
+	// assert for container count with label "unknown"
+	actualMetricUnknownContainerCount := &dto.Metric{}
+	expectedMetricUnknownContainerCount := float64(2)
+	metrics.RunningContainerCount.WithLabelValues(string(kubecontainer.ContainerStateUnknown)).Write(actualMetricUnknownContainerCount)
+	assert.Equal(t, expectedMetricUnknownContainerCount, actualMetricUnknownContainerCount.GetGauge().GetValue())
+
+	// assert for running pod count
+	actualMetricRunningPodCount := &dto.Metric{}
+	metrics.RunningPodCount.Write(actualMetricRunningPodCount)
+	expectedMetricRunningPodCount := float64(2)
+	assert.Equal(t, expectedMetricRunningPodCount, actualMetricRunningPodCount.GetGauge().GetValue())
+
+}