CPU usage displayed by the PodMetrics and NodeMetrics tables does not seem intuitive

like this: 
```go
➜  ~ kubectl get PodMetrics
NAME                               CPU         MEMORY    WINDOW
test-deployment-6d56d679c5-hc8lr   67131864n   71252Ki   15.009s
test-deployment-6d56d679c5-w2pvq   75122454n   77316Ki   15.01s
➜  ~ kubectl get NodeMetrics
NAME       CPU          MEMORY      WINDOW
minikube   401910564n   1762588Ki   20.162s
```

Normally we would use mili core instead of nano core, this seems to be the relevant code for this part
https://github.com/kubernetes-sigs/metrics-server/blob/4084f0b6885a65ebe456abae8118a38fc7d93bf6/pkg/storage/types.go#L52-L84

But I am not sure if there will be any problem if we modify storage? (Because PodMetrics NodeMetrics) There are many dependent parties using it. A better way is to use a hard-coded method in the table to change the unit of resource cpu from nano to milli

https://github.com/kubernetes-sigs/metrics-server/blob/4084f0b6885a65ebe456abae8118a38fc7d93bf6/pkg/api/table.go#L62-L67


	func resourceUsage(last, prev MetricsPoint) (corev1.ResourceList, api.TimeInfo, error) {
	if last.StartTime.Before(prev.StartTime) {
	return corev1.ResourceList{}, api.TimeInfo{}, fmt.Errorf("unexpected decrease in startTime of node/container")
	}
	if last.CumulativeCpuUsed < prev.CumulativeCpuUsed {
	return corev1.ResourceList{}, api.TimeInfo{}, fmt.Errorf("unexpected decrease in cumulative CPU usage value")
	}
	window := last.Timestamp.Sub(prev.Timestamp)
	cpuUsage := float64(last.CumulativeCpuUsed-prev.CumulativeCpuUsed) / window.Seconds()
	return corev1.ResourceList{
	corev1.ResourceCPU: uint64Quantity(uint64(cpuUsage), resource.DecimalSI, -9),
	corev1.ResourceMemory: uint64Quantity(last.MemoryUsage, resource.BinarySI, 0),
	}, api.TimeInfo{
	Timestamp: last.Timestamp,
	Window: window,
	}, nil
	}

	// uint64Quantity converts a uint64 into a Quantity, which only has constructors
	// that work with int64 (except for parse, which requires costly round-trips to string).
	// We lose precision until we fit in an int64 if greater than the max int64 value.
	func uint64Quantity(val uint64, format resource.Format, scale resource.Scale) resource.Quantity {
	q := *resource.NewScaledQuantity(int64(val), scale)
	if val > math.MaxInt64 {
	// lose an decimal order-of-magnitude precision,
	// so we can fit into a scaled quantity
	klog.V(2).InfoS("Found unexpectedly large resource value, losing precision to fit in scaled resource.Quantity", "value", val)
	q = *resource.NewScaledQuantity(int64(val/10), resource.Scale(1)+scale)
	}
	q.Format = format
	return q
	}

	row := make([]interface{}, 0, len(names)+1)
	row = append(row, pod.Name)
	for _, name := range names {
	v := usage[v1.ResourceName(name)]
	row = append(row, v.String())
	}

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

CPU usage displayed by the PodMetrics and NodeMetrics tables does not seem intuitive #1669

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

CPU usage displayed by the PodMetrics and NodeMetrics tables does not seem intuitive #1669

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions