benvin-utils/podgap/collect.go

package main

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"sort"

	corev1 "k8s.io/api/core/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/client-go/kubernetes"
	metricsv "k8s.io/metrics/pkg/client/clientset/versioned"
)

// row is one pod's aggregated (summed across containers) request/limit/usage.
type row struct {
	Namespace string `json:"namespace"`
	Pod       string `json:"pod"`

	CPUReqM int64 `json:"cpuRequestMillicores"`
	CPULimM int64 `json:"cpuLimitMillicores"`
	CPUUseM int64 `json:"cpuUsageMillicores"`
	MemReqB int64 `json:"memRequestBytes"`
	MemLimB int64 `json:"memLimitBytes"`
	MemUseB int64 `json:"memUsageBytes"`
}

func (r row) cpuUtil() float64 {
	if r.CPUReqM == 0 {
		return 0
	}
	return float64(r.CPUUseM) / float64(r.CPUReqM)
}

func (r row) memUtil() float64 {
	if r.MemReqB == 0 {
		return 0
	}
	return float64(r.MemUseB) / float64(r.MemReqB)
}

// cpuWasteM / memWasteB are reserved-but-unused amounts (negative = over request).
func (r row) cpuWasteM() int64 { return r.CPUReqM - r.CPUUseM }
func (r row) memWasteB() int64 { return r.MemReqB - r.MemUseB }

// collect lists pod specs and pod metrics and joins them by namespace/name.
func collect(ctx context.Context, core kubernetes.Interface, mc metricsv.Interface, ns string) ([]row, error) {
	pods, err := core.CoreV1().Pods(ns).List(ctx, metav1.ListOptions{})
	if err != nil {
		return nil, fmt.Errorf("list pods: %w", err)
	}
	metrics, err := mc.MetricsV1beta1().PodMetricses(ns).List(ctx, metav1.ListOptions{})
	if err != nil {
		return nil, fmt.Errorf("list pod metrics: %w", err)
	}

	type usage struct{ cpuM, memB int64 }
	used := make(map[string]usage, len(metrics.Items))
	for _, pm := range metrics.Items {
		var u usage
		for _, c := range pm.Containers {
			u.cpuM += c.Usage.Cpu().MilliValue()
			u.memB += c.Usage.Memory().Value()
		}
		used[pm.Namespace+"/"+pm.Name] = u
	}

	rows := make([]row, 0, len(pods.Items))
	for _, p := range pods.Items {
		// Only running pods have meaningful live usage to compare against.
		if p.Status.Phase != corev1.PodRunning {
			continue
		}
		r := row{Namespace: p.Namespace, Pod: p.Name}
		for _, c := range p.Spec.Containers {
			r.CPUReqM += c.Resources.Requests.Cpu().MilliValue()
			r.CPULimM += c.Resources.Limits.Cpu().MilliValue()
			r.MemReqB += c.Resources.Requests.Memory().Value()
			r.MemLimB += c.Resources.Limits.Memory().Value()
		}
		if u, ok := used[p.Namespace+"/"+p.Name]; ok {
			r.CPUUseM = u.cpuM
			r.MemUseB = u.memB
		}
		rows = append(rows, r)
	}
	return rows, nil
}

func filterRows(rows []row, opts options) []row {
	if !opts.overOnly && !opts.underOnly {
		return rows
	}
	out := rows[:0]
	for _, r := range rows {
		if opts.overOnly {
			overCPU := r.CPUReqM > 0 && r.cpuUtil() < opts.threshold
			overMem := r.MemReqB > 0 && r.memUtil() < opts.threshold
			if overCPU || overMem {
				out = append(out, r)
				continue
			}
		}
		if opts.underOnly {
			underCPU := r.CPUReqM > 0 && r.CPUUseM > r.CPUReqM
			underMem := r.MemReqB > 0 && r.MemUseB > r.MemReqB
			overLimCPU := r.CPULimM > 0 && r.CPUUseM > r.CPULimM*9/10
			overLimMem := r.MemLimB > 0 && r.MemUseB > r.MemLimB*9/10
			if underCPU || underMem || overLimCPU || overLimMem {
				out = append(out, r)
			}
		}
	}
	return out
}

func sortRows(rows []row, key string) {
	less := map[string]func(a, b row) bool{
		"cpu-waste": func(a, b row) bool { return a.cpuWasteM() > b.cpuWasteM() },
		"mem-waste": func(a, b row) bool { return a.memWasteB() > b.memWasteB() },
		"cpu-util":  func(a, b row) bool { return a.cpuUtil() > b.cpuUtil() },
		"mem-util":  func(a, b row) bool { return a.memUtil() > b.memUtil() },
		"cpu-use":   func(a, b row) bool { return a.CPUUseM > b.CPUUseM },
		"mem-use":   func(a, b row) bool { return a.MemUseB > b.MemUseB },
		"name":      func(a, b row) bool { return a.Namespace+a.Pod < b.Namespace+b.Pod },
	}
	fn, ok := less[key]
	if !ok {
		fn = less["cpu-waste"]
	}
	sort.SliceStable(rows, func(i, j int) bool { return fn(rows[i], rows[j]) })
}

func render(w io.Writer, rows []row, format string) error {
	switch format {
	case "json":
		enc := json.NewEncoder(w)
		enc.SetIndent("", "  ")
		return enc.Encode(rows)
	case "csv":
		fmt.Fprintln(w, "namespace,pod,cpu_req_m,cpu_use_m,cpu_lim_m,cpu_util,mem_req_bytes,mem_use_bytes,mem_lim_bytes,mem_util")
		for _, r := range rows {
			fmt.Fprintf(w, "%s,%s,%d,%d,%d,%.3f,%d,%d,%d,%.3f\n",
				r.Namespace, r.Pod, r.CPUReqM, r.CPUUseM, r.CPULimM, r.cpuUtil(),
				r.MemReqB, r.MemUseB, r.MemLimB, r.memUtil())
		}
		return nil
	}
	return fmt.Errorf("unknown output format %q", format)
}

// --- unit formatting ---

func cpu(m int64) string {
	if m == 0 {
		return "-"
	}
	if m < 0 {
		return "-" + cpu(-m)
	}
	if m < 1000 {
		return fmt.Sprintf("%dm", m)
	}
	return fmt.Sprintf("%.2f", float64(m)/1000)
}

func mem(b int64) string {
	if b == 0 {
		return "-"
	}
	if b < 0 {
		return "-" + mem(-b)
	}
	const ki = 1024
	switch {
	case b >= ki*ki*ki:
		return fmt.Sprintf("%.1fGi", float64(b)/(ki*ki*ki))
	case b >= ki*ki:
		return fmt.Sprintf("%.0fMi", float64(b)/(ki*ki))
	case b >= ki:
		return fmt.Sprintf("%.0fKi", float64(b)/ki)
	default:
		return fmt.Sprintf("%dB", b)
	}
}