Register imagePullThroughput and count with MiB

Shukui Yang · ialidzhikov · commit 711cebd484d4 · 2024-02-26T09:31:47.000+02:00
Signed-off-by: Shukui Yang <yangshukui@bytedance.com> (cherry picked from commit db22327)
diff --git a/pkg/cri/sbserver/image_pull.go b/pkg/cri/sbserver/image_pull.go
@@ -213,8 +213,9 @@ func (c *criService) PullImage(ctx context.Context, r *runtime.PullImageRequest)
 		}
 	}
 
+	const mbToByte = 1024 * 1024
 	size, _ := image.Size(ctx)
-	imagePullingSpeed := float64(size) / time.Since(startTime).Seconds()
+	imagePullingSpeed := float64(size) / mbToByte / time.Since(startTime).Seconds()
 	imagePullThroughput.Observe(imagePullingSpeed)
 
 	log.G(ctx).Infof("Pulled image %q with image id %q, repo tag %q, repo digest %q, size %q in %s", imageRef, imageID,
diff --git a/pkg/cri/sbserver/metrics.go b/pkg/cri/sbserver/metrics.go
@@ -42,13 +42,18 @@ var (
 
 	imagePulls           metrics.LabeledCounter
 	inProgressImagePulls metrics.Gauge
-	//  pull duration / (image size / 1MBi)
+	// image size in MB / image pull duration in seconds
 	imagePullThroughput prom.Histogram
 )
 
 func init() {
+	const (
+		namespace = "containerd"
+		subsystem = "cri_sandboxed"
+	)
+
 	// these CRI metrics record latencies for successful operations around a sandbox and container's lifecycle.
-	ns := metrics.NewNamespace("containerd", "cri_sandboxed", nil)
+	ns := metrics.NewNamespace(namespace, subsystem, nil)
 
 	sandboxListTimer = ns.NewTimer("sandbox_list", "time to list sandboxes")
 	sandboxCreateNetworkTimer = ns.NewTimer("sandbox_create_network", "time to create the network for a sandbox")
@@ -72,12 +77,14 @@ func init() {
 	inProgressImagePulls = ns.NewGauge("in_progress_image_pulls", "in progress pulls", metrics.Total)
 	imagePullThroughput = prom.NewHistogram(
 		prom.HistogramOpts{
-			Name:    "image_pulling_throughput",
-			Help:    "image pull throughput",
-			Buckets: prom.DefBuckets,
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "image_pulling_throughput",
+			Help:      "image pull throughput",
+			Buckets:   prom.DefBuckets,
 		},
 	)
-
+	ns.Add(imagePullThroughput)
 	metrics.Register(ns)
 }
 
diff --git a/pkg/cri/server/image_pull.go b/pkg/cri/server/image_pull.go
@@ -215,8 +215,9 @@ func (c *criService) PullImage(ctx context.Context, r *runtime.PullImageRequest)
 		}
 	}
 
+	const mbToByte = 1024 * 1024
 	size, _ := image.Size(ctx)
-	imagePullingSpeed := float64(size) / time.Since(startTime).Seconds()
+	imagePullingSpeed := float64(size) / mbToByte / time.Since(startTime).Seconds()
 	imagePullThroughput.Observe(imagePullingSpeed)
 
 	log.G(ctx).Infof("Pulled image %q with image id %q, repo tag %q, repo digest %q, size %q in %s", imageRef, imageID,
diff --git a/pkg/cri/server/metrics.go b/pkg/cri/server/metrics.go
@@ -42,13 +42,18 @@ var (
 
 	imagePulls           metrics.LabeledCounter
 	inProgressImagePulls metrics.Gauge
-	//  pull duration / (image size / 1MBi)
+	// image size in MB / image pull duration in seconds
 	imagePullThroughput prom.Histogram
 )
 
 func init() {
+	const (
+		namespace = "containerd"
+		subsystem = "cri"
+	)
+
 	// these CRI metrics record latencies for successful operations around a sandbox and container's lifecycle.
-	ns := metrics.NewNamespace("containerd", "cri", nil)
+	ns := metrics.NewNamespace(namespace, subsystem, nil)
 
 	sandboxListTimer = ns.NewTimer("sandbox_list", "time to list sandboxes")
 	sandboxCreateNetworkTimer = ns.NewTimer("sandbox_create_network", "time to create the network for a sandbox")
@@ -72,12 +77,15 @@ func init() {
 	inProgressImagePulls = ns.NewGauge("in_progress_image_pulls", "in progress pulls", metrics.Total)
 	imagePullThroughput = prom.NewHistogram(
 		prom.HistogramOpts{
-			Name:    "image_pulling_throughput",
-			Help:    "image pull throughput",
-			Buckets: prom.DefBuckets,
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "image_pulling_throughput",
+			Help:      "image pull throughput",
+			Buckets:   prom.DefBuckets,
 		},
 	)
 
+	ns.Add(imagePullThroughput)
 	metrics.Register(ns)
 }
 

Original file line number	Diff line number	Diff line change
`@@ -213,8 +213,9 @@ func (c criService) PullImage(ctx context.Context, r runtime.PullImageRequest)`
`213`	`213`	`}`
`214`	`214`	`}`
`215`	`215`
	`216`	`+ const mbToByte = 1024 * 1024`
`216`	`217`	`size, _ := image.Size(ctx)`
`217`		`- imagePullingSpeed := float64(size) / time.Since(startTime).Seconds()`
	`218`	`+ imagePullingSpeed := float64(size) / mbToByte / time.Since(startTime).Seconds()`
`218`	`219`	`imagePullThroughput.Observe(imagePullingSpeed)`
`219`	`220`
`220`	`221`	`log.G(ctx).Infof("Pulled image %q with image id %q, repo tag %q, repo digest %q, size %q in %s", imageRef, imageID,`
Original file line number	Diff line number	Diff line change
`@@ -215,8 +215,9 @@ func (c criService) PullImage(ctx context.Context, r runtime.PullImageRequest)`
`215`	`215`	`}`
`216`	`216`	`}`
`217`	`217`
	`218`	`+ const mbToByte = 1024 * 1024`
`218`	`219`	`size, _ := image.Size(ctx)`
`219`		`- imagePullingSpeed := float64(size) / time.Since(startTime).Seconds()`
	`220`	`+ imagePullingSpeed := float64(size) / mbToByte / time.Since(startTime).Seconds()`
`220`	`221`	`imagePullThroughput.Observe(imagePullingSpeed)`
`221`	`222`
`222`	`223`	`log.G(ctx).Infof("Pulled image %q with image id %q, repo tag %q, repo digest %q, size %q in %s", imageRef, imageID,`