Add stime metric and improve commenting

dns.go

@@ -13,27 +13,36 @@ import (
 	log "github.com/sirupsen/logrus"
 	"math"
 	"strconv"
+	"time"
 )
 
 //////////////////////////////////////////////////////////////////////////
+// data structures
+
+// for holding the metrics
 
 type DNSMetrics struct {
 	soa       *prometheus.GaugeVec
 	rtt       *prometheus.GaugeVec
 	valid     *prometheus.GaugeVec
+	stime     *prometheus.GaugeVec
 	label_map []prometheus.Labels
 }
 
+// for specifying DNS servers
+
 type DNSServer struct {
 	role string
 	name string
 	ip   uint8
 	addr string
-	soa  uint
+	soa  uint64
 }
 
+// hardcoded :( list of DNS servers to query
+
 var dns_servers = []*DNSServer{
-	&DNSServer{"master", "master.delegation-servers.dn42", 6, "[fd42:180:3de0:10:5054:ff:fe87:ea39]:53", 0},
+	&DNSServer{"master", "j.master.delegation-servers.dn42", 6, "[fd42:180:3de0:10:5054:ff:fe87:ea39]:53", 0},
 	&DNSServer{"delegation", "b.delegation-servers.dn42", 4, "172.20.129.1:53", 0},
 	&DNSServer{"delegation", "b.delegation-servers.dn42", 6, "[fd42:4242:2601:ac53::1]:53", 0},
 	&DNSServer{"delegation", "j.delegation-servers.dn42", 4, "172.20.1.18:53", 0},
@@ -56,6 +65,7 @@ var dns_servers = []*DNSServer{
 }
 
 //////////////////////////////////////////////////////////////////////////
+// initialisation function to register metrics
 
 func (m *DNSMetrics) Register() {
 
@@ -77,6 +87,12 @@ func (m *DNSMetrics) Register() {
 	}, []string{"role", "name", "ip", "addr"})
 	prometheus.MustRegister(m.valid)
 
+	m.stime = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "dn42_dns_stime",
+		Help: "Returns the time between now and the SOA serial number",
+	}, []string{"role", "name", "ip", "addr"})
+	prometheus.MustRegister(m.stime)
+
 	// pre-populate the labels
 	m.label_map = make([]prometheus.Labels, len(dns_servers))
 
@@ -92,28 +108,53 @@ func (m *DNSMetrics) Register() {
 }
 
 //////////////////////////////////////////////////////////////////////////
+// collect metrics for all DNS servers
 
 func (m *DNSMetrics) Collect() {
 
-	for ix, server := range dns_servers {
+	now := uint64(time.Now().Unix())
-		soa, rtt := server.Query()
-		server.soa = uint(soa)
 
+	// go through each server in turn
+	for ix, server := range dns_servers {
+
+		// query it
+		soa, rtt := server.Query()
+		server.soa = uint64(soa)
+
+		// SOA and RTT are direct metrics returned from the query
 		m.soa.With(m.label_map[ix]).Set(soa)
 		m.rtt.With(m.label_map[ix]).Set(rtt)
 
+		// check if the returned SOA matches j.master.delegation-servers.dn42
 		var valid uint = 0
 
 		if server.soa == 0 {
+			// didn't get a result, server issue
 			valid = 2
 		} else {
 			if server.soa != dns_servers[0].soa {
+				// SOA didn't match
 				valid = 1
 			} else {
 				valid = 0
 			}
 		}
 
+		// before setting whether the server is valid, calculate the stime
+		// (difference in time between now and the SOA, to allow checking that
+		// it is not stale)
+		// it's possible that the SOA could be in the future if there is a
+		// clock mismatch between monitor and DNS server, in which case this
+		// is flagged as a server error
+
+		if server.soa > now {
+			// server error
+			valid = 2
+		} else {
+			m.stime.With(m.label_map[ix]).Set(float64(now - server.soa))
+		}
+
+		// finally set the valid status
 		m.valid.With(m.label_map[ix]).Set(float64(valid))
 	}