feature: add cri metrics

Signed-off-by: Starnop <[email protected]>

apis/metrics/metrics.go

@@ -1,9 +1,9 @@
 package metrics
 
 import (
-	"fmt"
 	"sync"
-	"time"
+
+	util_metrics "github.com/alibaba/pouch/pkg/utils/metrics"
 
 	"github.com/prometheus/client_golang/prometheus"
 )
@@ -14,34 +14,33 @@ func init() {
 }
 
 const (
-	namespace = "engine"
-	subsystem = "daemon"
+	subsystemPouch = "daemon"
 )
 
 var (
 	// ImagePullSummary records the summary of pulling image latency.
-	ImagePullSummary = newLabelSummary("image_pull_latency_microseconds", "Latency in microseconds to pull a image.", "image")
+	ImagePullSummary = util_metrics.NewLabelSummary(subsystemPouch, "image_pull_latency_microseconds", "Latency in microseconds to pull a image.", "image")
 
 	// ContainerActionsCounter records the number of container operations.
-	ContainerActionsCounter = newLabelCounter("container_actions_counter", "The number of container operations", "action")
+	ContainerActionsCounter = util_metrics.NewLabelCounter(subsystemPouch, "container_actions_counter", "The number of container operations", "action")
 
 	// ContainerSuccessActionsCounter records the number of container success operations.
-	ContainerSuccessActionsCounter = newLabelCounter("container_success_actions_counter", "The number of container success operations", "action")
+	ContainerSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemPouch, "container_success_actions_counter", "The number of container success operations", "action")
 
 	// ImageActionsCounter records the number of image operations.
-	ImageActionsCounter = newLabelCounter("image_actions_counter", "The number of image operations", "action")
+	ImageActionsCounter = util_metrics.NewLabelCounter(subsystemPouch, "image_actions_counter", "The number of image operations", "action")
 
 	// ImageSuccessActionsCounter the number of image success operations.
-	ImageSuccessActionsCounter = newLabelCounter("image_success_actions_counter", "The number of image success operations", "action")
+	ImageSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemPouch, "image_success_actions_counter", "The number of image success operations", "action")
 
 	// ContainerActionsTimer records the time cost of each container action.
-	ContainerActionsTimer = newLabelTimer("container_actions", "The number of seconds it takes to process each container action", "action")
+	ContainerActionsTimer = util_metrics.NewLabelTimer(subsystemPouch, "container_actions", "The number of seconds it takes to process each container action", "action")
 
 	// ImageActionsTimer records the time cost of each image action.
-	ImageActionsTimer = newLabelTimer("image_actions", "The number of seconds it takes to process each image action", "action")
+	ImageActionsTimer = util_metrics.NewLabelTimer(subsystemPouch, "image_actions", "The number of seconds it takes to process each image action", "action")
 
 	// EngineVersion records the version and commit information of the engine process.
-	EngineVersion = newLabelGauge("engine", "The version and commit information of the engine process", "commit")
+	EngineVersion = util_metrics.NewLabelGauge(subsystemPouch, "engine", "The version and commit information of the engine process", "commit")
 )
 
 var registerMetrics sync.Once
@@ -60,52 +59,3 @@ func Register() {
 		prometheus.MustRegister(ImageActionsTimer)
 	})
 }
-
-// SinceInMicroseconds gets the time since the specified start in microseconds.
-func SinceInMicroseconds(start time.Time) float64 {
-	return float64(time.Since(start).Nanoseconds() / time.Microsecond.Nanoseconds())
-}
-
-func newLabelSummary(name, help string, labels ...string) *prometheus.SummaryVec {
-	return prometheus.NewSummaryVec(
-		prometheus.SummaryOpts{
-			Namespace:   namespace,
-			Subsystem:   subsystem,
-			Name:        name,
-			Help:        help,
-			ConstLabels: nil,
-		}, labels)
-}
-
-func newLabelCounter(name, help string, labels ...string) *prometheus.CounterVec {
-	return prometheus.NewCounterVec(
-		prometheus.CounterOpts{
-			Namespace:   namespace,
-			Subsystem:   subsystem,
-			Name:        fmt.Sprintf("%s_%s", name, total),
-			Help:        help,
-			ConstLabels: nil,
-		}, labels)
-}
-
-func newLabelGauge(name, help string, labels ...string) *prometheus.GaugeVec {
-	return prometheus.NewGaugeVec(
-		prometheus.GaugeOpts{
-			Namespace:   namespace,
-			Subsystem:   subsystem,
-			Name:        fmt.Sprintf("%s_%s", name, Unit("info")),
-			Help:        help,
-			ConstLabels: nil,
-		}, labels)
-}
-
-func newLabelTimer(name, help string, labels ...string) *prometheus.HistogramVec {
-	return prometheus.NewHistogramVec(
-		prometheus.HistogramOpts{
-			Namespace:   namespace,
-			Subsystem:   subsystem,
-			Name:        fmt.Sprintf("%s_%s", name, seconds),
-			Help:        help,
-			ConstLabels: nil,
-		}, labels)
-}

apis/server/image_bridge.go

@@ -14,6 +14,7 @@ import (
 	"github.com/alibaba/pouch/apis/types"
 	"github.com/alibaba/pouch/daemon/mgr"
 	"github.com/alibaba/pouch/pkg/httputils"
+	util_metrics "github.com/alibaba/pouch/pkg/utils/metrics"
 
 	"github.com/gorilla/mux"
 	"github.com/opencontainers/go-digest"
@@ -39,7 +40,7 @@ func (s *Server) pullImage(ctx context.Context, rw http.ResponseWriter, req *htt
 
 	// record the time spent during image pull procedure.
 	defer func(start time.Time) {
-		metrics.ImagePullSummary.WithLabelValues(image).Observe(metrics.SinceInMicroseconds(start))
+		metrics.ImagePullSummary.WithLabelValues(image).Observe(util_metrics.SinceInMicroseconds(start))
 		metrics.ImageActionsTimer.WithLabelValues(label).Observe(time.Since(start).Seconds())
 	}(time.Now())
 

cri/metrics/metrics.go

@@ -0,0 +1,102 @@
+package metrics
+
+import (
+	"sync"
+
+	util_metrics "github.com/alibaba/pouch/pkg/utils/metrics"
+
+	"github.com/grpc-ecosystem/go-grpc-prometheus"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+func init() {
+	// Register prometheus metrics.
+	Register()
+}
+
+const (
+	subsystemCRI = "cri"
+)
+
+var (
+	// GRPCMetrics create some standard server metrics.
+	GRPCMetrics = grpc_prometheus.NewServerMetrics()
+
+	// PodActionsCounter records the number of pod operations.
+	PodActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "pod_actions_counter", "The number of pod operations", "action")
+
+	// PodSuccessActionsCounter records the number of pod success operations.
+	PodSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "pod_success_actions_counter", "The number of pod success operations", "action")
+
+	// PodActionsTimer records the time cost of each pod action.
+	PodActionsTimer = util_metrics.NewLabelTimer(subsystemCRI, "pod_actions", "The number of seconds it takes to process each pod action", "action")
+
+	// ContainerActionsCounter records the number of container operations.
+	ContainerActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "container_actions_counter", "The number of container operations", "action")
+
+	// ContainerSuccessActionsCounter records the number of container success operations.
+	ContainerSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "container_success_actions_counter", "The number of container success operations", "action")
+
+	// ContainerActionsTimer records the time cost of each container action.
+	ContainerActionsTimer = util_metrics.NewLabelTimer(subsystemCRI, "container_actions", "The number of seconds it takes to process each container action", "action")
+
+	// ImagePullSummary records the summary of pulling image latency.
+	ImagePullSummary = util_metrics.NewLabelSummary(subsystemCRI, "image_pull_latency_microseconds", "Latency in microseconds to pull a image.", "image")
+
+	// ImageActionsCounter records the number of image operations.
+	ImageActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "image_actions_counter", "The number of image operations", "action")
+
+	// ImageSuccessActionsCounter the number of image success operations.
+	ImageSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "image_success_actions_counter", "The number of image success operations", "action")
+
+	// ImageActionsTimer records the time cost of each image action.
+	ImageActionsTimer = util_metrics.NewLabelTimer(subsystemCRI, "image_actions", "The number of seconds it takes to process each image action", "action")
+
+	// VolumeActionsCounter records the number of volume operations.
+	VolumeActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "volume_actions_counter", "The number of volume operations", "action")
+
+	// VolumeSuccessActionsCounter the number of volume success operations.
+	VolumeSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "volume_success_actions_counter", "The number of volume success operations", "action")
+
+	// VolumeActionsTimer records the time cost of each volume action.
+	VolumeActionsTimer = util_metrics.NewLabelTimer(subsystemCRI, "volume_actions", "The number of seconds it takes to process each volume action", "action")
+
+	// RuntimeActionsCounter records the number of runtime operations.
+	RuntimeActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "runtime_actions_counter", "The number of runtime operations", "action")
+
+	// RuntimeSuccessActionsCounter the number of runtime success operations.
+	RuntimeSuccessActionsCounter = util_metrics.NewLabelCounter(subsystemCRI, "runtime_success_actions_counter", "The number of runtime success operations", "action")
+
+	// RuntimeActionsTimer records the time cost of each runtime action.
+	RuntimeActionsTimer = util_metrics.NewLabelTimer(subsystemCRI, "runtime_actions", "The number of seconds it takes to process each runtime action", "action")
+)
+
+var registerMetrics sync.Once
+
+// Register all metrics.
+func Register() {
+	registerMetrics.Do(func() {
+		// EnableHandlingTimeHistogram turns on recording of handling time
+		// of RPCs. Histogram metrics can be very expensive for Prometheus
+		// to retain and query.
+		GRPCMetrics.EnableHandlingTimeHistogram()
+
+		prometheus.MustRegister(GRPCMetrics)
+		prometheus.MustRegister(PodActionsCounter)
+		prometheus.MustRegister(PodSuccessActionsCounter)
+		prometheus.MustRegister(PodActionsTimer)
+		prometheus.MustRegister(ContainerActionsCounter)
+		prometheus.MustRegister(ContainerSuccessActionsCounter)
+		prometheus.MustRegister(ContainerActionsTimer)
+		prometheus.MustRegister(ImagePullSummary)
+		prometheus.MustRegister(ImageActionsCounter)
+		prometheus.MustRegister(ImageSuccessActionsCounter)
+		prometheus.MustRegister(ImageActionsTimer)
+		prometheus.MustRegister(VolumeActionsCounter)
+		prometheus.MustRegister(VolumeSuccessActionsCounter)
+		prometheus.MustRegister(VolumeActionsTimer)
+		prometheus.MustRegister(RuntimeActionsCounter)
+		prometheus.MustRegister(RuntimeSuccessActionsCounter)
+		prometheus.MustRegister(RuntimeActionsTimer)
+	})
+}