otel/cgroups/cpu.go

package cgroups

import (
	"context"
	"fmt"
	"sync"

	cgroups "github.com/containerd/cgroups/v3/cgroup2"
	"github.com/containerd/cgroups/v3/cgroup2/stats"
	"go.opentelemetry.io/otel"
	"go.opentelemetry.io/otel/metric"
)

func WithProvider(provider metric.MeterProvider) Option {
	return func(c *config) {
		c.provider = provider
	}
}

func WithPrefic(name string) Option {
	return func(c *config) {
		c.prefix = name
	}
}

func WithControllers(ctrls ...Controller) Option {
	return func(c *config) {
		c.controllers = ctrls
	}
}

// Option supports configuring optional settings for runtime metrics.
type Option func(*config)

func newConfig(opts ...Option) config {
	cfg := config{
		controllers: []Controller{ControllerCPU},
		provider:    otel.GetMeterProvider(),
		prefix:      "cgroups.",
	}

	for _, opt := range opts {
		opt(&cfg)
	}

	return cfg
}

type config struct {
	controllers []Controller
	provider    metric.MeterProvider
	prefix      string
}

type cgroup struct {
	meter metric.Meter
	stats func() (*stats.Metrics, error)
}

func (c *cgroup) cpu(prefix string) (func(context.Context, metric.Observer, *stats.Metrics), []metric.Observable, error) {
	var (
		err           error
		usageUsec     metric.Int64ObservableGauge
		userUsec      metric.Int64ObservableGauge
		systemUsec    metric.Int64ObservableGauge
		nrPeriods     metric.Int64ObservableGauge
		nrThrottled   metric.Int64ObservableGauge
		throttledUsec metric.Int64ObservableGauge
	)

	if usageUsec, err = c.meter.Int64ObservableGauge(
		prefix+"usage_usec",
		metric.WithDescription("Total cpu usage"),
		metric.WithUnit("microseconds"),
	); err != nil {
		return nil, nil, fmt.Errorf("usage_usec:%w", err)
	}

	if userUsec, err = c.meter.Int64ObservableGauge(
		prefix+"user_usec",
		metric.WithDescription("Current cpu usage in user space"),
		metric.WithUnit("microseconds"),
	); err != nil {
		return nil, nil, fmt.Errorf("user_usec:%w", err)
	}

	if systemUsec, err = c.meter.Int64ObservableGauge(
		prefix+"system_usec",
		metric.WithDescription("Current cpu usage in kernel space"),
		metric.WithUnit("microseconds"),
	); err != nil {
		return nil, nil, fmt.Errorf("system_usec:%w", err)
	}

	if nrPeriods, err = c.meter.Int64ObservableGauge(
		prefix+"nr_periods",
		metric.WithDescription("Current cpu number of periods (only if controller is enabled)"),
	); err != nil {
		return nil, nil, fmt.Errorf("nr_periods:%w", err)
	}

	if nrThrottled, err = c.meter.Int64ObservableGauge(
		prefix+"nr_throttled",
		metric.WithDescription("Total number of times tasks have been throttled (only if controller is enabled)"),
	); err != nil {
		return nil, nil, fmt.Errorf("nr_throttled:%w", err)
	}

	if throttledUsec, err = c.meter.Int64ObservableGauge(
		prefix+"throttled_usec",
		metric.WithDescription("Total time duration for which tasks have been throttled. (only if controller is enabled)"),
		metric.WithUnit("microseconds"),
	); err != nil {
		return nil, nil, fmt.Errorf("throttled_usec:%w", err)
	}

	return func(ctx context.Context, o metric.Observer, m *stats.Metrics) {
			o.ObserveInt64(usageUsec, int64(m.CPU.UsageUsec))
			o.ObserveInt64(userUsec, int64(m.CPU.UserUsec))
			o.ObserveInt64(systemUsec, int64(m.CPU.SystemUsec))
			o.ObserveInt64(nrPeriods, int64(m.CPU.NrPeriods))
			o.ObserveInt64(nrThrottled, int64(m.CPU.NrThrottled))
			o.ObserveInt64(throttledUsec, int64(m.CPU.ThrottledUsec))
		}, []metric.Observable{
			usageUsec, userUsec, systemUsec, nrPeriods, nrThrottled, throttledUsec,
		}, nil
}

func (c *cgroup) register(prefix string, ctrls ...Controller) (metric.Registration, error) {
	observable := make([]metric.Observable, 0, len(ctrls))
	callbacks := make([]func(context.Context, metric.Observer, *stats.Metrics), 0, len(ctrls))
	for _, ctrl := range ctrls {
		switch ctrl {
		case ControllerCPU:
			cpu, cpus, err := c.cpu(prefix + "cpu.")
			if err != nil {
				return nil, fmt.Errorf("controller cpu:%w", err)
			}

			observable = append(observable, cpus...)
			callbacks = append(callbacks, cpu)
		}
	}

	var lock sync.Mutex

	lock.Lock()
	defer lock.Unlock()

	ureg, err := c.meter.RegisterCallback(
		func(ctx context.Context, obsrv metric.Observer) error {
			lock.Lock()
			defer lock.Unlock()

			stats, uErr := c.stats()
			if uErr != nil {
				return fmt.Errorf("cpu get stat:%w", uErr)
			}

			for _, callback := range callbacks {
				callback(ctx, obsrv, stats)
			}

			return nil
		},
		observable...,
	)

	if err != nil {
		return nil, fmt.Errorf("register callback:%w", err)
	}

	return ureg, nil
}

func Start(group string, opts ...Option) (metric.Registration, error) {
	manager, err := cgroups.Load(group)
	if err != nil {
		return nil, fmt.Errorf("load:%w", err)
	}

	cfg := newConfig(opts...)

	cgr := cgroup{
		meter: cfg.provider.Meter(
			"gitoa.ru/go-4devs/otel/cgroups",
			metric.WithInstrumentationVersion(Version()),
		),
		stats: manager.Stat,
	}

	return cgr.register(cfg.prefix+group+".", cfg.controllers...)
}