frostfs-http-gw/metrics/metrics.go
Denis Kirillov 67c5818fc1 [#2] Rename internals
Signed-off-by: Denis Kirillov <d.kirillov@yadro.com>
2022-12-20 18:56:02 +03:00

236 lines
7.3 KiB
Go

package metrics
import (
"net/http"
"github.com/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"
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"
)
type GateMetrics struct {
stateMetrics
poolMetricsCollector
}
type stateMetrics struct {
healthCheck prometheus.Gauge
}
type poolMetricsCollector struct {
pool *pool.Pool
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 *pool.Pool) *GateMetrics {
stateMetric := newStateMetrics()
stateMetric.register()
poolMetric := newPoolMetricsCollector(p)
poolMetric.register()
return &GateMetrics{
stateMetrics: *stateMetric,
poolMetricsCollector: *poolMetric,
}
}
func (g *GateMetrics) Unregister() {
g.stateMetrics.unregister()
prometheus.Unregister(&g.poolMetricsCollector)
}
func newStateMetrics() *stateMetrics {
return &stateMetrics{
healthCheck: prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: stateSubsystem,
Name: "health",
Help: "Current HTTP gateway state",
}),
}
}
func (m stateMetrics) register() {
prometheus.MustRegister(m.healthCheck)
}
func (m stateMetrics) unregister() {
prometheus.Unregister(m.healthCheck)
}
func (m stateMetrics) SetHealth(s int32) {
m.healthCheck.Set(float64(s))
}
func newPoolMetricsCollector(p *pool.Pool) *poolMetricsCollector {
overallErrors := prometheus.NewGauge(
prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: poolSubsystem,
Name: "overall_errors",
Help: "Total number of errors in pool",
},
)
overallNodeErrors := prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: poolSubsystem,
Name: "overall_node_errors",
Help: "Total number of errors for connection in pool",
},
[]string{
"node",
},
)
overallNodeRequests := prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: poolSubsystem,
Name: "overall_node_requests",
Help: "Total number of requests to specific node in pool",
},
[]string{
"node",
},
)
currentErrors := prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: poolSubsystem,
Name: "current_errors",
Help: "Number of errors on current connections that will be reset after the threshold",
},
[]string{
"node",
},
)
requestsDuration := prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: poolSubsystem,
Name: "avg_request_duration",
Help: "Average request duration (in milliseconds) for specific method on node in pool",
},
[]string{
"node",
"method",
},
)
return &poolMetricsCollector{
pool: p,
overallErrors: overallErrors,
overallNodeErrors: overallNodeErrors,
overallNodeRequests: overallNodeRequests,
currentErrors: currentErrors,
requestDuration: requestsDuration,
}
}
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.pool.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(), methodGetContainerEacl).Set(float64(node.AverageGetContainerEACL().Milliseconds()))
m.requestDuration.WithLabelValues(node.Address(), methodSetContainerEacl).Set(float64(node.AverageSetContainerEACL().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()))
}
// 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")),
}
}