traefik/pkg/metrics/prometheus.go

513 lines
16 KiB
Go
Raw Normal View History

package metrics
import (
2018-11-14 09:18:03 +00:00
"context"
2017-11-20 08:40:03 +00:00
"net/http"
"sort"
"strings"
"sync"
2017-11-20 08:40:03 +00:00
"github.com/containous/mux"
"github.com/containous/traefik/pkg/config/dynamic"
2019-03-15 08:42:03 +00:00
"github.com/containous/traefik/pkg/log"
"github.com/containous/traefik/pkg/safe"
"github.com/containous/traefik/pkg/types"
"github.com/go-kit/kit/metrics"
stdprometheus "github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promhttp"
)
const (
2018-08-06 12:58:03 +00:00
// MetricNamePrefix prefix of all metric names
MetricNamePrefix = "traefik_"
// server meta information
2018-08-06 12:58:03 +00:00
metricConfigPrefix = MetricNamePrefix + "config_"
configReloadsTotalName = metricConfigPrefix + "reloads_total"
configReloadsFailuresTotalName = metricConfigPrefix + "reloads_failure_total"
configLastReloadSuccessName = metricConfigPrefix + "last_reload_success"
configLastReloadFailureName = metricConfigPrefix + "last_reload_failure"
// entrypoint
2018-08-06 12:58:03 +00:00
metricEntryPointPrefix = MetricNamePrefix + "entrypoint_"
entrypointReqsTotalName = metricEntryPointPrefix + "requests_total"
entrypointReqDurationName = metricEntryPointPrefix + "request_duration_seconds"
entrypointOpenConnsName = metricEntryPointPrefix + "open_connections"
// backend level.
// MetricBackendPrefix prefix of all backend metric names
MetricBackendPrefix = MetricNamePrefix + "backend_"
backendReqsTotalName = MetricBackendPrefix + "requests_total"
backendReqDurationName = MetricBackendPrefix + "request_duration_seconds"
backendOpenConnsName = MetricBackendPrefix + "open_connections"
backendRetriesTotalName = MetricBackendPrefix + "retries_total"
backendServerUpName = MetricBackendPrefix + "server_up"
)
// promState holds all metric state internally and acts as the only Collector we register for Prometheus.
//
// This enables control to remove metrics that belong to outdated configuration.
// As an example why this is required, consider Traefik learns about a new service.
// It populates the 'traefik_server_backend_up' metric for it with a value of 1 (alive).
// When the backend is undeployed now the metric is still there in the client library
2018-06-05 10:32:03 +00:00
// and will be returned on the metrics endpoint until Traefik would be restarted.
//
2018-06-05 10:32:03 +00:00
// To solve this problem promState keeps track of Traefik's dynamic configuration.
// Metrics that "belong" to a dynamic configuration part like backends or entrypoints
// are removed after they were scraped at least once when the corresponding object
// doesn't exist anymore.
var promState = newPrometheusState()
// PrometheusHandler exposes Prometheus routes.
type PrometheusHandler struct{}
2018-11-14 09:18:03 +00:00
// Append adds Prometheus routes on a router.
func (h PrometheusHandler) Append(router *mux.Router) {
2017-11-20 08:40:03 +00:00
router.Methods(http.MethodGet).Path("/metrics").Handler(promhttp.Handler())
}
// RegisterPrometheus registers all Prometheus metrics.
// It must be called only once and failing to register the metrics will lead to a panic.
2018-11-14 09:18:03 +00:00
func RegisterPrometheus(ctx context.Context, config *types.Prometheus) Registry {
2018-08-06 12:58:03 +00:00
standardRegistry := initStandardRegistry(config)
2018-11-14 09:18:03 +00:00
if !registerPromState(ctx) {
2018-08-06 12:58:03 +00:00
return nil
}
return standardRegistry
}
func initStandardRegistry(config *types.Prometheus) Registry {
buckets := []float64{0.1, 0.3, 1.2, 5.0}
if config.Buckets != nil {
buckets = config.Buckets
}
safe.Go(func() {
promState.ListenValueUpdates()
})
configReloads := newCounterFrom(promState.collectors, stdprometheus.CounterOpts{
Name: configReloadsTotalName,
Help: "Config reloads",
}, []string{})
configReloadsFailures := newCounterFrom(promState.collectors, stdprometheus.CounterOpts{
Name: configReloadsFailuresTotalName,
Help: "Config failure reloads",
}, []string{})
lastConfigReloadSuccess := newGaugeFrom(promState.collectors, stdprometheus.GaugeOpts{
Name: configLastReloadSuccessName,
Help: "Last config reload success",
}, []string{})
lastConfigReloadFailure := newGaugeFrom(promState.collectors, stdprometheus.GaugeOpts{
Name: configLastReloadFailureName,
Help: "Last config reload failure",
}, []string{})
entrypointReqs := newCounterFrom(promState.collectors, stdprometheus.CounterOpts{
Name: entrypointReqsTotalName,
Help: "How many HTTP requests processed on an entrypoint, partitioned by status code, protocol, and method.",
}, []string{"code", "method", "protocol", "entrypoint"})
entrypointReqDurations := newHistogramFrom(promState.collectors, stdprometheus.HistogramOpts{
Name: entrypointReqDurationName,
Help: "How long it took to process the request on an entrypoint, partitioned by status code, protocol, and method.",
Buckets: buckets,
}, []string{"code", "method", "protocol", "entrypoint"})
entrypointOpenConns := newGaugeFrom(promState.collectors, stdprometheus.GaugeOpts{
Name: entrypointOpenConnsName,
Help: "How many open connections exist on an entrypoint, partitioned by method and protocol.",
}, []string{"method", "protocol", "entrypoint"})
backendReqs := newCounterFrom(promState.collectors, stdprometheus.CounterOpts{
Name: backendReqsTotalName,
Help: "How many HTTP requests processed on a backend, partitioned by status code, protocol, and method.",
}, []string{"code", "method", "protocol", "backend"})
backendReqDurations := newHistogramFrom(promState.collectors, stdprometheus.HistogramOpts{
Name: backendReqDurationName,
Help: "How long it took to process the request on a backend, partitioned by status code, protocol, and method.",
Buckets: buckets,
}, []string{"code", "method", "protocol", "backend"})
backendOpenConns := newGaugeFrom(promState.collectors, stdprometheus.GaugeOpts{
Name: backendOpenConnsName,
Help: "How many open connections exist on a backend, partitioned by method and protocol.",
}, []string{"method", "protocol", "backend"})
backendRetries := newCounterFrom(promState.collectors, stdprometheus.CounterOpts{
Name: backendRetriesTotalName,
Help: "How many request retries happened on a backend.",
}, []string{"backend"})
backendServerUp := newGaugeFrom(promState.collectors, stdprometheus.GaugeOpts{
Name: backendServerUpName,
Help: "Backend server is up, described by gauge value of 0 or 1.",
}, []string{"backend", "url"})
promState.describers = []func(chan<- *stdprometheus.Desc){
configReloads.cv.Describe,
configReloadsFailures.cv.Describe,
lastConfigReloadSuccess.gv.Describe,
lastConfigReloadFailure.gv.Describe,
entrypointReqs.cv.Describe,
entrypointReqDurations.hv.Describe,
entrypointOpenConns.gv.Describe,
backendReqs.cv.Describe,
backendReqDurations.hv.Describe,
backendOpenConns.gv.Describe,
backendRetries.cv.Describe,
backendServerUp.gv.Describe,
}
return &standardRegistry{
enabled: true,
configReloadsCounter: configReloads,
configReloadsFailureCounter: configReloadsFailures,
lastConfigReloadSuccessGauge: lastConfigReloadSuccess,
lastConfigReloadFailureGauge: lastConfigReloadFailure,
entrypointReqsCounter: entrypointReqs,
entrypointReqDurationHistogram: entrypointReqDurations,
entrypointOpenConnsGauge: entrypointOpenConns,
backendReqsCounter: backendReqs,
backendReqDurationHistogram: backendReqDurations,
backendOpenConnsGauge: backendOpenConns,
backendRetriesCounter: backendRetries,
backendServerUpGauge: backendServerUp,
}
}
2018-11-14 09:18:03 +00:00
func registerPromState(ctx context.Context) bool {
2018-08-06 12:58:03 +00:00
if err := stdprometheus.Register(promState); err != nil {
2018-11-14 09:18:03 +00:00
logger := log.FromContext(ctx)
2018-08-06 12:58:03 +00:00
if _, ok := err.(stdprometheus.AlreadyRegisteredError); !ok {
2018-11-14 09:18:03 +00:00
logger.Errorf("Unable to register Traefik to Prometheus: %v", err)
2018-08-06 12:58:03 +00:00
return false
}
2018-11-14 09:18:03 +00:00
logger.Debug("Prometheus collector already registered.")
2018-08-06 12:58:03 +00:00
}
return true
}
2018-06-05 10:32:03 +00:00
// OnConfigurationUpdate receives the current configuration from Traefik.
// It then converts the configuration to the optimized package internal format
// and sets it to the promState.
func OnConfigurationUpdate(configurations dynamic.Configurations) {
2018-06-05 10:32:03 +00:00
dynamicConfig := newDynamicConfig()
2018-11-14 09:18:03 +00:00
// FIXME metrics
// for _, config := range configurations {
// for _, frontend := range config.Frontends {
// for _, entrypointName := range frontend.EntryPoints {
// dynamicConfig.entrypoints[entrypointName] = true
// }
// }
//
// for backendName, backend := range config.Backends {
// dynamicConfig.backends[backendName] = make(map[string]bool)
// for _, server := range backend.Servers {
// dynamicConfig.backends[backendName][server.URL] = true
// }
// }
// }
2018-06-05 10:32:03 +00:00
promState.SetDynamicConfig(dynamicConfig)
}
func newPrometheusState() *prometheusState {
return &prometheusState{
2018-06-05 10:32:03 +00:00
collectors: make(chan *collector),
dynamicConfig: newDynamicConfig(),
state: make(map[string]*collector),
}
}
type prometheusState struct {
2018-06-05 10:32:03 +00:00
collectors chan *collector
describers []func(ch chan<- *stdprometheus.Desc)
mtx sync.Mutex
dynamicConfig *dynamicConfig
state map[string]*collector
}
2018-06-05 10:32:03 +00:00
func (ps *prometheusState) SetDynamicConfig(dynamicConfig *dynamicConfig) {
ps.mtx.Lock()
defer ps.mtx.Unlock()
2018-06-05 10:32:03 +00:00
ps.dynamicConfig = dynamicConfig
}
func (ps *prometheusState) ListenValueUpdates() {
for collector := range ps.collectors {
ps.mtx.Lock()
ps.state[collector.id] = collector
ps.mtx.Unlock()
}
}
// Describe implements prometheus.Collector and simply calls
// the registered describer functions.
func (ps *prometheusState) Describe(ch chan<- *stdprometheus.Desc) {
for _, desc := range ps.describers {
desc(ch)
}
}
// Collect implements prometheus.Collector. It calls the Collect
// method of all metrics it received on the collectors channel.
2018-06-05 10:32:03 +00:00
// It's also responsible to remove metrics that belong to an outdated configuration.
// The removal happens only after their Collect method was called to ensure that
// also those metrics will be exported on the current scrape.
func (ps *prometheusState) Collect(ch chan<- stdprometheus.Metric) {
ps.mtx.Lock()
defer ps.mtx.Unlock()
2018-06-05 10:32:03 +00:00
var outdatedKeys []string
for key, cs := range ps.state {
cs.collector.Collect(ch)
2018-06-05 10:32:03 +00:00
if ps.isOutdated(cs) {
outdatedKeys = append(outdatedKeys, key)
}
}
for _, key := range outdatedKeys {
2018-06-05 10:32:03 +00:00
ps.state[key].delete()
delete(ps.state, key)
}
}
2018-06-05 10:32:03 +00:00
// isOutdated checks whether the passed collector has labels that mark
// it as belonging to an outdated configuration of Traefik.
func (ps *prometheusState) isOutdated(collector *collector) bool {
labels := collector.labels
if entrypointName, ok := labels["entrypoint"]; ok && !ps.dynamicConfig.hasEntrypoint(entrypointName) {
return true
}
if backendName, ok := labels["backend"]; ok {
if !ps.dynamicConfig.hasBackend(backendName) {
return true
}
if url, ok := labels["url"]; ok && !ps.dynamicConfig.hasServerURL(backendName, url) {
return true
}
}
return false
}
func newDynamicConfig() *dynamicConfig {
return &dynamicConfig{
entrypoints: make(map[string]bool),
backends: make(map[string]map[string]bool),
}
}
// dynamicConfig holds the current configuration for entrypoints, backends,
// and server URLs in an optimized way to check for existence. This provides
// a performant way to check whether the collected metrics belong to the
// current configuration or to an outdated one.
type dynamicConfig struct {
entrypoints map[string]bool
backends map[string]map[string]bool
}
func (d *dynamicConfig) hasEntrypoint(entrypointName string) bool {
_, ok := d.entrypoints[entrypointName]
return ok
}
func (d *dynamicConfig) hasBackend(backendName string) bool {
_, ok := d.backends[backendName]
return ok
}
2018-06-05 10:32:03 +00:00
func (d *dynamicConfig) hasServerURL(backendName, serverURL string) bool {
if backend, hasBackend := d.backends[backendName]; hasBackend {
_, ok := backend[serverURL]
return ok
}
2018-06-05 10:32:03 +00:00
return false
}
2018-06-05 10:32:03 +00:00
func newCollector(metricName string, labels stdprometheus.Labels, c stdprometheus.Collector, delete func()) *collector {
return &collector{
2018-06-05 10:32:03 +00:00
id: buildMetricID(metricName, labels),
labels: labels,
collector: c,
2018-06-05 10:32:03 +00:00
delete: delete,
}
}
// collector wraps a Collector object from the Prometheus client library.
// It adds information on how many generations this metric should be present
// in the /metrics output, relatived to the time it was last tracked.
type collector struct {
2018-06-05 10:32:03 +00:00
id string
labels stdprometheus.Labels
collector stdprometheus.Collector
delete func()
}
2018-06-05 10:32:03 +00:00
func buildMetricID(metricName string, labels stdprometheus.Labels) string {
var labelNamesValues []string
for name, value := range labels {
labelNamesValues = append(labelNamesValues, name, value)
}
sort.Strings(labelNamesValues)
return metricName + ":" + strings.Join(labelNamesValues, "|")
}
func newCounterFrom(collectors chan<- *collector, opts stdprometheus.CounterOpts, labelNames []string) *counter {
cv := stdprometheus.NewCounterVec(opts, labelNames)
c := &counter{
name: opts.Name,
cv: cv,
collectors: collectors,
}
if len(labelNames) == 0 {
c.Add(0)
}
return c
}
type counter struct {
name string
cv *stdprometheus.CounterVec
labelNamesValues labelNamesValues
collectors chan<- *collector
}
func (c *counter) With(labelValues ...string) metrics.Counter {
return &counter{
name: c.name,
cv: c.cv,
labelNamesValues: c.labelNamesValues.With(labelValues...),
collectors: c.collectors,
}
}
func (c *counter) Add(delta float64) {
2018-06-05 10:32:03 +00:00
labels := c.labelNamesValues.ToLabels()
collector := c.cv.With(labels)
collector.Add(delta)
2018-06-05 10:32:03 +00:00
c.collectors <- newCollector(c.name, labels, collector, func() {
c.cv.Delete(labels)
})
}
func (c *counter) Describe(ch chan<- *stdprometheus.Desc) {
c.cv.Describe(ch)
}
func newGaugeFrom(collectors chan<- *collector, opts stdprometheus.GaugeOpts, labelNames []string) *gauge {
gv := stdprometheus.NewGaugeVec(opts, labelNames)
g := &gauge{
name: opts.Name,
gv: gv,
collectors: collectors,
}
if len(labelNames) == 0 {
g.Set(0)
}
return g
}
type gauge struct {
name string
gv *stdprometheus.GaugeVec
labelNamesValues labelNamesValues
collectors chan<- *collector
}
func (g *gauge) With(labelValues ...string) metrics.Gauge {
return &gauge{
name: g.name,
gv: g.gv,
labelNamesValues: g.labelNamesValues.With(labelValues...),
collectors: g.collectors,
}
}
2018-04-16 08:28:04 +00:00
func (g *gauge) Add(delta float64) {
2018-06-05 10:32:03 +00:00
labels := g.labelNamesValues.ToLabels()
collector := g.gv.With(labels)
2018-04-16 08:28:04 +00:00
collector.Add(delta)
2018-06-05 10:32:03 +00:00
g.collectors <- newCollector(g.name, labels, collector, func() {
g.gv.Delete(labels)
})
2018-04-16 08:28:04 +00:00
}
func (g *gauge) Set(value float64) {
2018-06-05 10:32:03 +00:00
labels := g.labelNamesValues.ToLabels()
collector := g.gv.With(labels)
collector.Set(value)
2018-06-05 10:32:03 +00:00
g.collectors <- newCollector(g.name, labels, collector, func() {
g.gv.Delete(labels)
})
}
func (g *gauge) Describe(ch chan<- *stdprometheus.Desc) {
g.gv.Describe(ch)
}
func newHistogramFrom(collectors chan<- *collector, opts stdprometheus.HistogramOpts, labelNames []string) *histogram {
hv := stdprometheus.NewHistogramVec(opts, labelNames)
return &histogram{
name: opts.Name,
hv: hv,
collectors: collectors,
}
}
type histogram struct {
name string
hv *stdprometheus.HistogramVec
labelNamesValues labelNamesValues
collectors chan<- *collector
}
func (h *histogram) With(labelValues ...string) metrics.Histogram {
return &histogram{
name: h.name,
hv: h.hv,
labelNamesValues: h.labelNamesValues.With(labelValues...),
collectors: h.collectors,
}
}
func (h *histogram) Observe(value float64) {
2018-06-05 10:32:03 +00:00
labels := h.labelNamesValues.ToLabels()
collector := h.hv.With(labels)
collector.Observe(value)
2018-06-05 10:32:03 +00:00
h.collectors <- newCollector(h.name, labels, collector, func() {
h.hv.Delete(labels)
})
}
func (h *histogram) Describe(ch chan<- *stdprometheus.Desc) {
h.hv.Describe(ch)
}
// labelNamesValues is a type alias that provides validation on its With method.
// Metrics may include it as a member to help them satisfy With semantics and
// save some code duplication.
type labelNamesValues []string
// With validates the input, and returns a new aggregate labelNamesValues.
func (lvs labelNamesValues) With(labelValues ...string) labelNamesValues {
if len(labelValues)%2 != 0 {
labelValues = append(labelValues, "unknown")
}
return append(lvs, labelValues...)
}
// ToLabels is a convenience method to convert a labelNamesValues
// to the native prometheus.Labels.
func (lvs labelNamesValues) ToLabels() stdprometheus.Labels {
labels := stdprometheus.Labels{}
for i := 0; i < len(lvs); i += 2 {
labels[lvs[i]] = lvs[i+1]
}
return labels
}