Record pleg pod relist interval and latency

pkg/kubelet/metrics/metrics.go

@@ -35,6 +35,8 @@ const (
 	DockerOperationsKey           = "docker_operations_latency_microseconds"
 	DockerErrorsKey               = "docker_errors"
 	PodWorkerStartLatencyKey      = "pod_worker_start_latency_microseconds"
+	PLEGRelistLatencyKey          = "pleg_relist_latency_microseconds"
+	PLEGRelistIntervalKey         = "pleg_relist_interval_microseconds"
 )
 
 var (
@@ -105,6 +107,20 @@ var (
 		},
 		[]string{"operation_type"},
 	)
+	PLEGRelistLatency = prometheus.NewSummary(
+		prometheus.SummaryOpts{
+			Subsystem: KubeletSubsystem,
+			Name:      PLEGRelistLatencyKey,
+			Help:      "Latency in microseconds for relisting pods in PLEG.",
+		},
+	)
+	PLEGRelistInterval = prometheus.NewSummary(
+		prometheus.SummaryOpts{
+			Subsystem: KubeletSubsystem,
+			Name:      PLEGRelistIntervalKey,
+			Help:      "Interval in microseconds between relisting in PLEG.",
+		},
+	)
 )
 
 var registerMetrics sync.Once
@@ -123,6 +139,8 @@ func Register(containerCache kubecontainer.RuntimeCache) {
 		prometheus.MustRegister(ContainersPerPodCount)
 		prometheus.MustRegister(DockerErrors)
 		prometheus.MustRegister(newPodAndContainerCollector(containerCache))
+		prometheus.MustRegister(PLEGRelistLatency)
+		prometheus.MustRegister(PLEGRelistInterval)
 	})
 }
 

pkg/kubelet/pleg/generic.go

@@ -22,6 +22,7 @@ import (
 
 	"github.com/golang/glog"
 	kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
+	"k8s.io/kubernetes/pkg/kubelet/metrics"
 	"k8s.io/kubernetes/pkg/types"
 	"k8s.io/kubernetes/pkg/util"
 )
@@ -50,6 +51,8 @@ type GenericPLEG struct {
 	eventChannel chan *PodLifecycleEvent
 	// The internal cache for container information.
 	containers map[string]containerInfo
+	// Time of the last relisting.
+	lastRelistTime time.Time
 }
 
 type containerInfo struct {
@@ -101,6 +104,17 @@ func generateEvent(podID types.UID, cid string, oldState, newState kubecontainer
 // with the internal pods/containers, and generats events accordingly.
 func (g *GenericPLEG) relist() {
 	glog.V(5).Infof("GenericPLEG: Relisting")
+	timestamp := time.Now()
+
+	if !g.lastRelistTime.IsZero() {
+		metrics.PLEGRelistInterval.Observe(metrics.SinceInMicroseconds(g.lastRelistTime))
+	}
+	defer func() {
+		// Update the relist time.
+		g.lastRelistTime = timestamp
+		metrics.PLEGRelistLatency.Observe(metrics.SinceInMicroseconds(timestamp))
+	}()
+
 	// Get all the pods.
 	pods, err := g.runtime.GetPods(true)
 	if err != nil {

pkg/metrics/kubelet_metrics.go

@@ -74,6 +74,12 @@ var KnownKubeletMetrics = map[string][]string{
 	"kubelet_generate_pod_status_latency_microseconds":       {"quantile"},
 	"kubelet_generate_pod_status_latency_microseconds_count": {},
 	"kubelet_generate_pod_status_latency_microseconds_sum":   {},
+	"kubelet_pleg_relist_latency_microseconds":               {"quantile"},
+	"kubelet_pleg_relist_latency_microseconds_sum":           {},
+	"kubelet_pleg_relist_latency_microseconds_count":         {},
+	"kubelet_pleg_relist_interval_microseconds":              {"quantile"},
+	"kubelet_pleg_relist_interval_microseconds_sum":          {},
+	"kubelet_pleg_relist_interval_microseconds_count":        {},
 	"kubelet_pod_start_latency_microseconds":                 {"quantile"},
 	"kubelet_pod_start_latency_microseconds_count":           {},
 	"kubelet_pod_start_latency_microseconds_sum":             {},