Add a new metric named ntp_server_reachable, transform the high drift…

… from a constant to a parameter

Closes #85

README.md

@@ -18,6 +18,7 @@ These are the metrics supported.
 - `ntp_precision_seconds`
 - `ntp_leap`
 - `ntp_scrape_duration_seconds`
+- `ntp_server_reachable`
 
 As an alternative to [the node-exporter's `time` module](https://github.com/prometheus/node_exporter/blob/master/docs/TIME.md), this exporter does not require an NTP component on localhost that it can talk to. We only look at the system clock and talk to the configured NTP server(s).
 
@@ -60,12 +61,20 @@ and connection options is defined by command-line options:
 ```
 -ntp.measurement-duration duration
    Duration of measurements in case of high (>10ms) drift. (default 30s)
+-ntp.high-drift duration
+   High drift threshold. (default 10ms)
 -ntp.protocol-version int
    NTP protocol version to use. (default 4)
 -ntp.server string
    NTP server to use (required).
 ```
 
+Command-line usage example:
+
+```sh
+ntp_exporter -ntp.server ntp.example.com -web.telemetry-path "/probe" -ntp.measurement-duration "5s" -ntp.high-drift "50ms"
+```
+
 ### Mode 2: Variable NTP server
 
 When the option `-ntp.source http` is specified, the NTP server and connection
@@ -75,11 +84,12 @@ request:
 - `target`: NTP server to use
 - `protocol`: NTP protocol version (2, 3 or 4)
 - `duration`: duration of measurements in case of high drift
+- `high-drift`: High drift threshold to trigger multiple probing
 
 For example:
 
 ```sh
-$ curl 'http://localhost:9559/metrics?target=ntp.example.com&protocol=4&duration=10s'
+$ curl 'http://localhost:9559/metrics?target=ntp.example.com&protocol=4&duration=10s&high-drift=100ms'
 ```
 
 ## Frequently asked questions (FAQ)

collector.go

@@ -30,11 +30,12 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-func CollectorInitial(target string, protocol int, duration time.Duration) Collector {
+func CollectorInitial(target string, protocol int, duration, highDrift time.Duration) Collector {
 	return Collector{
 		NtpServer:              target,
 		NtpProtocolVersion:     protocol,
 		NtpMeasurementDuration: duration,
+		NtpHighDrift:           highDrift,
 		buildInfo: prometheus.NewGaugeFunc(prometheus.GaugeOpts{
 			Namespace:   "ntp",
 			Name:        "build_info",
@@ -91,6 +92,11 @@ func CollectorInitial(target string, protocol int, duration time.Duration) Colle
 			Name:      "scrape_duration_seconds",
 			Help:      "ntp_exporter: Duration of a scrape job.",
 		}),
+		serverReachable: prometheus.NewGaugeVec(prometheus.GaugeOpts{
+			Namespace: "ntp",
+			Name:      "server_reachable",
+			Help:      "True if the NTP server is reachable by the NTP exporter.",
+		}, []string{"server"}),
 	}
 }
 
@@ -99,6 +105,7 @@ type Collector struct {
 	NtpServer              string
 	NtpProtocolVersion     int
 	NtpMeasurementDuration time.Duration
+	NtpHighDrift           time.Duration
 	buildInfo              prometheus.GaugeFunc
 	stratum                *prometheus.GaugeVec
 	drift                  *prometheus.GaugeVec
@@ -110,6 +117,7 @@ type Collector struct {
 	precision              *prometheus.GaugeVec
 	leap                   *prometheus.GaugeVec
 	scrapeDuration         prometheus.Summary
+	serverReachable        *prometheus.GaugeVec
 }
 
 // A single measurement returned by ntp server
@@ -138,11 +146,15 @@ func (c Collector) Describe(ch chan<- *prometheus.Desc) {
 	c.precision.Describe(ch)
 	c.leap.Describe(ch)
 	c.scrapeDuration.Describe(ch)
+	c.serverReachable.Describe(ch)
 }
 
 // Collect implements the prometheus.Collector interface.
 func (c Collector) Collect(ch chan<- prometheus.Metric) {
 	err := c.measure()
+
+	c.serverReachable.Collect(ch)
+
 	// only report data when measurement was successful
 	if err == nil {
 		c.buildInfo.Collect(ch)
@@ -163,17 +175,16 @@ func (c Collector) Collect(ch chan<- prometheus.Metric) {
 }
 
 func (c Collector) measure() error {
-	const highDrift = 0.01
-
 	begin := time.Now()
 	measurement, err := c.getClockOffsetAndStratum()
 
 	if err != nil {
+		c.serverReachable.WithLabelValues(c.NtpServer).Set(0)
 		return fmt.Errorf("couldn't get NTP measurement: %w", err)
 	}
 
 	// if clock drift is unusually high (e.g. >10ms): repeat measurements for 30 seconds and submit median value
-	if measurement.clockOffset > highDrift {
+	if measurement.clockOffset > c.NtpHighDrift.Seconds() {
 		// arrays of measurements used to calculate median
 		var measurementsClockOffset []float64
 		var measurementsStratum []float64
@@ -185,10 +196,11 @@ func (c Collector) measure() error {
 		var measurementsPrecision []float64
 		var measurementsLeap []float64
 
-		log.Printf("WARN: clock drift is above %.2fs, taking multiple measurements for %.2f seconds", highDrift, c.NtpMeasurementDuration.Seconds())
+		log.Printf("WARN: clock drift is above %.3fs, taking multiple measurements for %.2f seconds", c.NtpHighDrift.Seconds(), c.NtpMeasurementDuration.Seconds())
 		for time.Since(begin) < c.NtpMeasurementDuration {
 			nextMeasurement, err := c.getClockOffsetAndStratum()
 			if err != nil {
+				c.serverReachable.WithLabelValues(c.NtpServer).Set(0)
 				return fmt.Errorf("couldn't get NTP measurement: %w", err)
 			}
 
@@ -223,6 +235,7 @@ func (c Collector) measure() error {
 	c.rootDistance.WithLabelValues(c.NtpServer).Set(measurement.rootDistance)
 	c.precision.WithLabelValues(c.NtpServer).Set(measurement.precision)
 	c.leap.WithLabelValues(c.NtpServer).Set(measurement.leap)
+	c.serverReachable.WithLabelValues(c.NtpServer).Set(1)
 
 	c.scrapeDuration.Observe(time.Since(begin).Seconds())
 	return nil

main.go

@@ -45,6 +45,7 @@ var (
 	ntpServer              string
 	ntpProtocolVersion     int
 	ntpMeasurementDuration time.Duration
+	ntpHighDrift           time.Duration
 	ntpSource              string
 )
 
@@ -86,6 +87,7 @@ func init() {
 	flag.StringVar(&ntpServer, "ntp.server", "", "NTP server to use (required).")
 	flag.IntVar(&ntpProtocolVersion, "ntp.protocol-version", 4, "NTP protocol version to use.")
 	flag.DurationVar(&ntpMeasurementDuration, "ntp.measurement-duration", 30*time.Second, "Duration of measurements in case of high (>10ms) drift.")
+	flag.DurationVar(&ntpHighDrift, "ntp.high-drift", 10*time.Millisecond, "High drift threshold.")
 	flag.StringVar(&ntpSource, "ntp.source", "cli", "source of information about ntp server (cli / http).")
 	flag.Parse()
 }
@@ -96,6 +98,7 @@ func handlerMetrics(w http.ResponseWriter, r *http.Request) {
 	s := ntpServer
 	p := ntpProtocolVersion
 	d := ntpMeasurementDuration
+	hd := ntpHighDrift
 
 	if ntpSource == "http" {
 		for _, i := range []string{"target", "protocol", "duration"} {
@@ -123,10 +126,17 @@ func handlerMetrics(w http.ResponseWriter, r *http.Request) {
 			http.Error(w, err.Error(), http.StatusBadRequest)
 			return
 		}
+
+		if u, err := time.ParseDuration(r.URL.Query().Get("high-drift")); err == nil {
+			hd = u
+		} else {
+			http.Error(w, err.Error(), http.StatusBadRequest)
+			return
+		}
 	}
 
 	registry := prometheus.NewRegistry()
-	registry.MustRegister(CollectorInitial(s, p, d))
+	registry.MustRegister(CollectorInitial(s, p, d, hd))
 	h := promhttp.HandlerFor(registry, promhttp.HandlerOpts{ErrorLog: logger})
 	h.ServeHTTP(w, r)
 }