Add new metric for initial call check. (#446)

Author: dennwc

pkg/sip/inbound.go

@@ -201,6 +201,7 @@ func (s *Server) onInvite(log *slog.Logger, req *sip.Request, tx sip.ServerTrans
 }
 
 func (s *Server) processInvite(req *sip.Request, tx sip.ServerTransaction) (retErr error) {
+	start := time.Now()
 	var state *CallState
 	ctx := context.Background()
 	defer func() {
@@ -265,6 +266,14 @@ func (s *Server) processInvite(req *sip.Request, tx sip.ServerTransaction) (retE
 	cmon := s.mon.NewCall(stats.Inbound, from.Host, to.Host)
 	cmon.InviteReq()
 	defer cmon.SessionDur()()
+	var checkDurOnce sync.Once
+	checkDur := cmon.CheckDur()
+	checked := func() {
+		checkDurOnce.Do(func() {
+			checkDur.Observe(time.Since(start).Seconds())
+		})
+	}
+	defer checked()
 	joinDur := cmon.JoinDur()
 
 	if !s.conf.HideInboundPort {
@@ -297,6 +306,7 @@ func (s *Server) processInvite(req *sip.Request, tx sip.ServerTransaction) (retE
 	}
 
 	r, err := s.handler.GetAuthCredentials(ctx, callInfo)
+	checked()
 	if err != nil {
 		cmon.InviteErrorShort("auth-error")
 		log.Warnw("Rejecting inbound, auth check failed", err)

pkg/stats/monitor.go

@@ -70,6 +70,7 @@ type Monitor struct {
 	durSession      *prometheus.HistogramVec
 	durCall         *prometheus.HistogramVec
 	durJoin         *prometheus.HistogramVec
+	durCheck        *prometheus.HistogramVec
 	cpuLoad         prometheus.Gauge
 	sdpSize         *prometheus.HistogramVec
 	nodeAvailable   prometheus.GaugeFunc
@@ -191,6 +192,15 @@ func (m *Monitor) Start(conf *config.Config) error {
 		Buckets:     durBucketsLong,
 	}, []string{"dir"}))
 
+	m.durCheck = mustRegister(m, prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Namespace:   "livekit",
+		Subsystem:   "sip",
+		Name:        "dur_check_sec",
+		Help:        "SIP call check duration (from INVITE to an initial dispatch response)",
+		ConstLabels: prometheus.Labels{"node_id": conf.NodeID},
+		Buckets:     durBucketsOp,
+	}, []string{"dir"}))
+
 	m.durJoin = mustRegister(m, prometheus.NewHistogramVec(prometheus.HistogramOpts{
 		Namespace:   "livekit",
 		Subsystem:   "sip",
@@ -365,6 +375,10 @@ func (c *CallMonitor) CallDur() func() time.Duration {
 	return prometheus.NewTimer(c.m.durCall.With(c.labelsShort(nil))).ObserveDuration
 }
 
+func (c *CallMonitor) CheckDur() prometheus.Observer {
+	return c.m.durCheck.With(c.labelsShort(nil))
+}
+
 func (c *CallMonitor) JoinDur() func() time.Duration {
 	return prometheus.NewTimer(c.m.durJoin.With(c.labelsShort(nil))).ObserveDuration
 }