daemon/metrics.go
3343d234
 package daemon
 
e4c03623
 import (
0e8e8f0f
 	"path/filepath"
e4c03623
 	"sync"
 
0e8e8f0f
 	"github.com/docker/docker/pkg/mount"
 	"github.com/docker/docker/pkg/plugingetter"
62c1f0ef
 	metrics "github.com/docker/go-metrics"
0e8e8f0f
 	"github.com/pkg/errors"
e4c03623
 	"github.com/prometheus/client_golang/prometheus"
1009e6a4
 	"github.com/sirupsen/logrus"
e4c03623
 )
3343d234
 
0e8e8f0f
 const metricsPluginType = "MetricsCollector"
 
3343d234
 var (
 	containerActions          metrics.LabeledTimer
 	imageActions              metrics.LabeledTimer
 	networkActions            metrics.LabeledTimer
a28b173a
 	engineInfo                metrics.LabeledGauge
3343d234
 	engineCpus                metrics.Gauge
 	engineMemory              metrics.Gauge
 	healthChecksCounter       metrics.Counter
 	healthChecksFailedCounter metrics.Counter
e4c03623
 
 	stateCtr *stateCounter
3343d234
 )
 
 func init() {
 	ns := metrics.NewNamespace("engine", "daemon", nil)
 	containerActions = ns.NewLabeledTimer("container_actions", "The number of seconds it takes to process each container action", "action")
 	for _, a := range []string{
 		"start",
 		"changes",
 		"commit",
 		"create",
 		"delete",
 	} {
 		containerActions.WithValues(a).Update(0)
 	}
e4c03623
 
3343d234
 	networkActions = ns.NewLabeledTimer("network_actions", "The number of seconds it takes to process each network action", "action")
a28b173a
 	engineInfo = ns.NewLabeledGauge("engine", "The information related to the engine and the OS it is running on", metrics.Unit("info"),
3343d234
 		"version",
 		"commit",
 		"architecture",
a28b173a
 		"graphdriver",
 		"kernel", "os",
 		"os_type",
 		"daemon_id", // ID is a randomly generated unique identifier (e.g. UUID4)
3343d234
 	)
 	engineCpus = ns.NewGauge("engine_cpus", "The number of cpus that the host system of the engine has", metrics.Unit("cpus"))
 	engineMemory = ns.NewGauge("engine_memory", "The number of bytes of memory that the host system of the engine has", metrics.Bytes)
 	healthChecksCounter = ns.NewCounter("health_checks", "The total number of health checks")
 	healthChecksFailedCounter = ns.NewCounter("health_checks_failed", "The total number of failed health checks")
 	imageActions = ns.NewLabeledTimer("image_actions", "The number of seconds it takes to process each image action", "action")
e4c03623
 
 	stateCtr = newStateCounter(ns.NewDesc("container_states", "The count of containers in various states", metrics.Unit("containers"), "state"))
 	ns.Add(stateCtr)
 
3343d234
 	metrics.Register(ns)
 }
e4c03623
 
 type stateCounter struct {
 	mu     sync.Mutex
 	states map[string]string
 	desc   *prometheus.Desc
 }
 
 func newStateCounter(desc *prometheus.Desc) *stateCounter {
 	return &stateCounter{
 		states: make(map[string]string),
 		desc:   desc,
 	}
 }
 
 func (ctr *stateCounter) get() (running int, paused int, stopped int) {
 	ctr.mu.Lock()
 	defer ctr.mu.Unlock()
 
 	states := map[string]int{
 		"running": 0,
 		"paused":  0,
 		"stopped": 0,
 	}
 	for _, state := range ctr.states {
 		states[state]++
 	}
 	return states["running"], states["paused"], states["stopped"]
 }
 
 func (ctr *stateCounter) set(id, label string) {
 	ctr.mu.Lock()
 	ctr.states[id] = label
 	ctr.mu.Unlock()
 }
 
 func (ctr *stateCounter) del(id string) {
 	ctr.mu.Lock()
 	delete(ctr.states, id)
 	ctr.mu.Unlock()
 }
 
 func (ctr *stateCounter) Describe(ch chan<- *prometheus.Desc) {
 	ch <- ctr.desc
 }
 
 func (ctr *stateCounter) Collect(ch chan<- prometheus.Metric) {
 	running, paused, stopped := ctr.get()
 	ch <- prometheus.MustNewConstMetric(ctr.desc, prometheus.GaugeValue, float64(running), "running")
 	ch <- prometheus.MustNewConstMetric(ctr.desc, prometheus.GaugeValue, float64(paused), "paused")
 	ch <- prometheus.MustNewConstMetric(ctr.desc, prometheus.GaugeValue, float64(stopped), "stopped")
 }
0e8e8f0f
 
 func (d *Daemon) cleanupMetricsPlugins() {
 	ls := d.PluginStore.GetAllManagedPluginsByCap(metricsPluginType)
 	var wg sync.WaitGroup
 	wg.Add(len(ls))
 
9db2c624
 	for _, plugin := range ls {
 		p := plugin
0e8e8f0f
 		go func() {
 			defer wg.Done()
 			pluginStopMetricsCollection(p)
 		}()
 	}
 	wg.Wait()
 
 	if d.metricsPluginListener != nil {
 		d.metricsPluginListener.Close()
 	}
 }
 
 type metricsPlugin struct {
 	plugingetter.CompatPlugin
 }
 
 func (p metricsPlugin) sock() string {
 	return "metrics.sock"
 }
 
 func (p metricsPlugin) sockBase() string {
 	return filepath.Join(p.BasePath(), "run", "docker")
 }
 
 func pluginStartMetricsCollection(p plugingetter.CompatPlugin) error {
 	type metricsPluginResponse struct {
 		Err string
 	}
 	var res metricsPluginResponse
 	if err := p.Client().Call(metricsPluginType+".StartMetrics", nil, &res); err != nil {
 		return errors.Wrap(err, "could not start metrics plugin")
 	}
 	if res.Err != "" {
 		return errors.New(res.Err)
 	}
 	return nil
 }
 
 func pluginStopMetricsCollection(p plugingetter.CompatPlugin) {
 	if err := p.Client().Call(metricsPluginType+".StopMetrics", nil, nil); err != nil {
 		logrus.WithError(err).WithField("name", p.Name()).Error("error stopping metrics collector")
 	}
 
 	mp := metricsPlugin{p}
 	sockPath := filepath.Join(mp.sockBase(), mp.sock())
 	if err := mount.Unmount(sockPath); err != nil {
 		if mounted, _ := mount.Mounted(sockPath); mounted {
 			logrus.WithError(err).WithField("name", p.Name()).WithField("socket", sockPath).Error("error unmounting metrics socket for plugin")
 		}
 	}
 }