start telemetry participant worker to collect signal stats (#2538)

* start telemetry participant worker to collect signal stats

* format

* resolve room

* tidy

pkg/telemetry/events.go

@@ -80,16 +80,17 @@ func (t *telemetryService) ParticipantJoined(
 	shouldSendEvent bool,
 ) {
 	t.enqueue(func() {
-		prometheus.IncrementParticipantRtcConnected(1)
-		prometheus.AddParticipant()
-
-		t.createWorker(
+		_, found := t.getOrCreateWorker(
 			ctx,
 			livekit.RoomID(room.Sid),
 			livekit.RoomName(room.Name),
 			livekit.ParticipantID(participant.Sid),
 			livekit.ParticipantIdentity(participant.Identity),
 		)
+		if !found {
+			prometheus.IncrementParticipantRtcConnected(1)
+			prometheus.AddParticipant()
+		}
 
 		if shouldSendEvent {
 			ev := newParticipantEvent(livekit.AnalyticsEventType_PARTICIPANT_JOINED, room, participant)
@@ -117,18 +118,14 @@ func (t *telemetryService) ParticipantActive(
 			})
 		}
 
-		worker, ok := t.getWorker(livekit.ParticipantID(participant.Sid))
-		if !ok {
-			// in case of session migration, we may not have seen a Join event take place.
-			// we'd need to create the worker here before being able to process events
-			worker = t.createWorker(
-				ctx,
-				livekit.RoomID(room.Sid),
-				livekit.RoomName(room.Name),
-				livekit.ParticipantID(participant.Sid),
-				livekit.ParticipantIdentity(participant.Identity),
-			)
-
+		worker, found := t.getOrCreateWorker(
+			ctx,
+			livekit.RoomID(room.Sid),
+			livekit.RoomName(room.Name),
+			livekit.ParticipantID(participant.Sid),
+			livekit.ParticipantIdentity(participant.Identity),
+		)
+		if !found {
 			// need to also account for participant count
 			prometheus.AddParticipant()
 		}

pkg/telemetry/signalanddatastats.go

@@ -15,16 +15,17 @@
 package telemetry
 
 import (
+	"context"
 	"fmt"
+	"sync"
 	"time"
 
+	"github.com/frostbyte73/core"
 	"go.uber.org/atomic"
 
-	"github.com/frostbyte73/core"
+	"github.com/livekit/livekit-server/pkg/config"
 	"github.com/livekit/protocol/livekit"
 	"github.com/livekit/protocol/utils"
-
-	"github.com/livekit/livekit-server/pkg/config"
 )
 
 type BytesTrackType string
@@ -139,6 +140,61 @@ func (s *BytesTrackStats) reporter() {
 
 // -----------------------------------------------------------------------
 
+type BytesSignalStats struct {
+	BytesTrackStats
+	ctx context.Context
+
+	mu sync.Mutex
+	ri *livekit.Room
+	pi *livekit.ParticipantInfo
+}
+
+func NewBytesSignalStats(ctx context.Context, telemetry TelemetryService) *BytesSignalStats {
+	return &BytesSignalStats{
+		BytesTrackStats: BytesTrackStats{
+			telemetry: telemetry,
+		},
+		ctx: ctx,
+	}
+}
+
+func (s *BytesSignalStats) ResolveRoom(ri *livekit.Room) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.ri == nil && ri.GetSid() != "" {
+		s.ri = ri
+		s.maybeStart()
+	}
+}
+
+func (s *BytesSignalStats) ResolveParticipant(pi *livekit.ParticipantInfo) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.pi == nil {
+		s.pi = pi
+		s.maybeStart()
+	}
+}
+
+func (s *BytesSignalStats) maybeStart() {
+	if s.ri == nil || s.pi == nil {
+		return
+	}
+
+	s.pID = livekit.ParticipantID(s.pi.Sid)
+	s.trackID = BytesTrackIDForParticipantID(BytesTrackTypeSignal, s.pID)
+
+	s.telemetry.ParticipantJoined(s.ctx, s.ri, s.pi, nil, nil, false)
+	go s.reporter()
+}
+
+func (s *BytesSignalStats) reporter() {
+	s.BytesTrackStats.reporter()
+	s.telemetry.ParticipantLeft(s.ctx, s.ri, s.pi, false)
+}
+
+// -----------------------------------------------------------------------
+
 func BytesTrackIDForParticipantID(typ BytesTrackType, participantID livekit.ParticipantID) livekit.TrackID {
 	return livekit.TrackID(fmt.Sprintf("%s_%s%s", utils.TrackPrefix, string(typ), participantID))
 }

pkg/telemetry/telemetryservice.go

@@ -19,12 +19,13 @@ import (
 	"sync"
 	"time"
 
+	"golang.org/x/exp/maps"
+
 	"github.com/livekit/livekit-server/pkg/config"
 	"github.com/livekit/livekit-server/pkg/utils"
 	"github.com/livekit/protocol/livekit"
 	"github.com/livekit/protocol/logger"
 	"github.com/livekit/protocol/webhook"
-	"golang.org/x/exp/maps"
 )
 
 //go:generate go run github.com/maxbrunsfeld/counterfeiter/v6 . TelemetryService
@@ -153,12 +154,19 @@ func (t *telemetryService) getWorker(participantID livekit.ParticipantID) (worke
 	return
 }
 
-func (t *telemetryService) createWorker(ctx context.Context,
+func (t *telemetryService) getOrCreateWorker(ctx context.Context,
 	roomID livekit.RoomID,
 	roomName livekit.RoomName,
 	participantID livekit.ParticipantID,
 	participantIdentity livekit.ParticipantIdentity,
-) *StatsWorker {
+) (*StatsWorker, bool) {
+	t.lock.Lock()
+	defer t.lock.Unlock()
+
+	if worker, ok := t.workers[participantID]; ok {
+		return worker, true
+	}
+
 	worker := newStatsWorker(
 		ctx,
 		t,
@@ -168,11 +176,10 @@ func (t *telemetryService) createWorker(ctx context.Context,
 		participantIdentity,
 	)
 
-	t.lock.Lock()
 	t.workers[participantID] = worker
 	t.workersShadow = maps.Values(t.workers)
-	t.lock.Unlock()
-	return worker
+
+	return worker, false
 }
 
 func (t *telemetryService) cleanupWorkers() {