Added metric for DNS failures

Signed-off-by: Chris Green <34572557+cgreen12@users.noreply.github.com>

pkg/goldpinger/client.go

@@ -61,6 +61,7 @@ func checkDNS() *models.DNSResults {
 		_, err := net.LookupIP(host)
 		if err != nil {
 			dnsResult.Error = err.Error()
+			CountDnsError(host)
 		}
 		dnsResult.ResponseTimeMs = time.Since(start).Nanoseconds() / int64(time.Millisecond)
 		results[host] = dnsResult

pkg/goldpinger/stats.go

@@ -82,6 +82,16 @@ var (
 			"type",
 		},
 	)
+	goldpingerDnsErrorsCounter = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "goldpinger_dns_errors_total",
+			Help: "Statistics of DNS errors per instance",
+		},
+		[]string{
+			"goldpinger_instance",
+			"host",
+		},
+	)
 
 	bootTime = time.Now()
 )
@@ -92,6 +102,7 @@ func init() {
 	prometheus.MustRegister(goldpingerResponseTimePeersHistogram)
 	prometheus.MustRegister(goldpingerResponseTimeKubernetesHistogram)
 	prometheus.MustRegister(goldpingerErrorsCounter)
+	prometheus.MustRegister(goldpingerDnsErrorsCounter)
 	log.Println("Metrics setup - see /metrics")
 }
 
@@ -131,6 +142,14 @@ func CountError(errorType string) {
 	).Inc()
 }
 
+// counts instances of dns errors
+func CountDnsError(host string) {
+	goldpingerDnsErrorsCounter.WithLabelValues(
+		GoldpingerConfig.Hostname,
+		host,
+	).Inc()
+}
+
 // returns a timer for easy observing of the durations of calls to kubernetes API
 func GetLabeledKubernetesCallsTimer() *prometheus.Timer {
 	return prometheus.NewTimer(