frostfs-http-gw/metrics/metrics.go

package metrics

import (
	"net/http"

	"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/pool"
	"github.com/prometheus/client_golang/prometheus"
	"github.com/prometheus/client_golang/prometheus/promhttp"
	"go.uber.org/zap"
)

const (
	namespace       = "frostfs_http_gw"
	stateSubsystem  = "state"
	poolSubsystem   = "pool"
	serverSubsystem = "server"
)

const (
	healthMetric      = "health"
	versionInfoMetric = "version_info"
)

const (
	overallErrorsMetric       = "overall_errors"
	overallNodeErrorsMetric   = "overall_node_errors"
	overallNodeRequestsMetric = "overall_node_requests"
	currentErrorMetric        = "current_errors"
	avgRequestDurationMetric  = "avg_request_duration"
)

const (
	methodGetBalance       = "get_balance"
	methodPutContainer     = "put_container"
	methodGetContainer     = "get_container"
	methodListContainer    = "list_container"
	methodDeleteContainer  = "delete_container"
	methodGetContainerEacl = "get_container_eacl"
	methodSetContainerEacl = "set_container_eacl"
	methodEndpointInfo     = "endpoint_info"
	methodNetworkInfo      = "network_info"
	methodPutObject        = "put_object"
	methodDeleteObject     = "delete_object"
	methodGetObject        = "get_object"
	methodHeadObject       = "head_object"
	methodRangeObject      = "range_object"
	methodCreateSession    = "create_session"
)

// HealthStatus of the gate application.
type HealthStatus int32

const (
	HealthStatusUndefined    HealthStatus = 0
	HealthStatusStarting     HealthStatus = 1
	HealthStatusReady        HealthStatus = 2
	HealthStatusShuttingDown HealthStatus = 3
)

type StatisticScraper interface {
	Statistic() pool.Statistic
}

type serverMetrics struct {
	endpointHealth *prometheus.GaugeVec
}

type GateMetrics struct {
	stateMetrics
	poolMetricsCollector
	serverMetrics
}

type stateMetrics struct {
	healthCheck prometheus.Gauge
	versionInfo *prometheus.GaugeVec
}

type poolMetricsCollector struct {
	scraper             StatisticScraper
	overallErrors       prometheus.Gauge
	overallNodeErrors   *prometheus.GaugeVec
	overallNodeRequests *prometheus.GaugeVec
	currentErrors       *prometheus.GaugeVec
	requestDuration     *prometheus.GaugeVec
}

// NewGateMetrics creates new metrics for http gate.
func NewGateMetrics(p StatisticScraper) *GateMetrics {
	stateMetric := newStateMetrics()
	stateMetric.register()

	poolMetric := newPoolMetricsCollector(p)
	poolMetric.register()

	serverMetric := newServerMetrics()
	serverMetric.register()

	return &GateMetrics{
		stateMetrics:         *stateMetric,
		poolMetricsCollector: *poolMetric,
		serverMetrics:        *serverMetric,
	}
}

func (g *GateMetrics) Unregister() {
	g.stateMetrics.unregister()
	prometheus.Unregister(&g.poolMetricsCollector)
	g.serverMetrics.unregister()
}

func newStateMetrics() *stateMetrics {
	return &stateMetrics{
		healthCheck: mustNewGauge(appMetricsDesc[stateSubsystem][healthMetric]),
		versionInfo: mustNewGaugeVec(appMetricsDesc[stateSubsystem][versionInfoMetric]),
	}
}

func (m stateMetrics) register() {
	prometheus.MustRegister(m.healthCheck)
	prometheus.MustRegister(m.versionInfo)
}

func (m stateMetrics) unregister() {
	prometheus.Unregister(m.healthCheck)
	prometheus.Unregister(m.versionInfo)
}

func (m stateMetrics) SetHealth(s HealthStatus) {
	m.healthCheck.Set(float64(s))
}

func (m stateMetrics) SetVersion(ver string) {
	m.versionInfo.WithLabelValues(ver).Set(1)
}

func newPoolMetricsCollector(p StatisticScraper) *poolMetricsCollector {
	return &poolMetricsCollector{
		scraper:             p,
		overallErrors:       mustNewGauge(appMetricsDesc[poolSubsystem][overallErrorsMetric]),
		overallNodeErrors:   mustNewGaugeVec(appMetricsDesc[poolSubsystem][overallNodeErrorsMetric]),
		overallNodeRequests: mustNewGaugeVec(appMetricsDesc[poolSubsystem][overallNodeRequestsMetric]),
		currentErrors:       mustNewGaugeVec(appMetricsDesc[poolSubsystem][currentErrorMetric]),
		requestDuration:     mustNewGaugeVec(appMetricsDesc[poolSubsystem][avgRequestDurationMetric]),
	}
}

func (m *poolMetricsCollector) Collect(ch chan<- prometheus.Metric) {
	m.updateStatistic()
	m.overallErrors.Collect(ch)
	m.overallNodeErrors.Collect(ch)
	m.overallNodeRequests.Collect(ch)
	m.currentErrors.Collect(ch)
	m.requestDuration.Collect(ch)
}

func (m *poolMetricsCollector) Describe(descs chan<- *prometheus.Desc) {
	m.overallErrors.Describe(descs)
	m.overallNodeErrors.Describe(descs)
	m.overallNodeRequests.Describe(descs)
	m.currentErrors.Describe(descs)
	m.requestDuration.Describe(descs)
}

func (m *poolMetricsCollector) register() {
	prometheus.MustRegister(m)
}

func (m *poolMetricsCollector) updateStatistic() {
	stat := m.scraper.Statistic()

	m.overallNodeErrors.Reset()
	m.overallNodeRequests.Reset()
	m.currentErrors.Reset()
	m.requestDuration.Reset()

	for _, node := range stat.Nodes() {
		m.overallNodeErrors.WithLabelValues(node.Address()).Set(float64(node.OverallErrors()))
		m.overallNodeRequests.WithLabelValues(node.Address()).Set(float64(node.Requests()))

		m.currentErrors.WithLabelValues(node.Address()).Set(float64(node.CurrentErrors()))
		m.updateRequestsDuration(node)
	}

	m.overallErrors.Set(float64(stat.OverallErrors()))
}

func (m *poolMetricsCollector) updateRequestsDuration(node pool.NodeStatistic) {
	m.requestDuration.WithLabelValues(node.Address(), methodGetBalance).Set(float64(node.AverageGetBalance().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodPutContainer).Set(float64(node.AveragePutContainer().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodGetContainer).Set(float64(node.AverageGetContainer().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodListContainer).Set(float64(node.AverageListContainer().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodDeleteContainer).Set(float64(node.AverageDeleteContainer().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodEndpointInfo).Set(float64(node.AverageEndpointInfo().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodNetworkInfo).Set(float64(node.AverageNetworkInfo().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodPutObject).Set(float64(node.AveragePutObject().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodDeleteObject).Set(float64(node.AverageDeleteObject().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodGetObject).Set(float64(node.AverageGetObject().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodHeadObject).Set(float64(node.AverageHeadObject().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodRangeObject).Set(float64(node.AverageRangeObject().Milliseconds()))
	m.requestDuration.WithLabelValues(node.Address(), methodCreateSession).Set(float64(node.AverageCreateSession().Milliseconds()))
}

func newServerMetrics() *serverMetrics {
	return &serverMetrics{
		endpointHealth: mustNewGaugeVec(appMetricsDesc[serverSubsystem][healthMetric]),
	}
}

func (m serverMetrics) register() {
	prometheus.MustRegister(m.endpointHealth)
}

func (m serverMetrics) unregister() {
	prometheus.Unregister(m.endpointHealth)
}

func (m serverMetrics) MarkHealthy(endpoint string) {
	m.endpointHealth.WithLabelValues(endpoint).Set(float64(1))
}

func (m serverMetrics) MarkUnhealthy(endpoint string) {
	m.endpointHealth.WithLabelValues(endpoint).Set(float64(0))
}

// NewPrometheusService creates a new service for gathering prometheus metrics.
func NewPrometheusService(log *zap.Logger, cfg Config) *Service {
	if log == nil {
		return nil
	}

	return &Service{
		Server: &http.Server{
			Addr:    cfg.Address,
			Handler: promhttp.Handler(),
		},
		enabled:     cfg.Enabled,
		serviceType: "Prometheus",
		log:         log.With(zap.String("service", "Prometheus")),
	}
}