Merge pull request #235 from kerberos-io/fix/debugging-lost-keyframes

fix/debugging-lost-keyframes

.github/workflows/pr-description.yaml

@@ -2,6 +2,11 @@ name: Autofill PR description
 
 on: pull_request
 
+env:
+  ORGANIZATION: uugai
+  PROJECT: ${{ github.event.repository.name }}
+  PR_NUMBER: ${{ github.event.number }}
+  
 jobs:
   openai-pr-description:
     runs-on: ubuntu-22.04
@@ -16,4 +21,6 @@ jobs:
           azure_openai_api_key: ${{ secrets.AZURE_OPENAI_API_KEY }}
           azure_openai_endpoint: ${{ secrets.AZURE_OPENAI_ENDPOINT }}
           azure_openai_version: ${{ secrets.AZURE_OPENAI_VERSION }}
+          openai_model: ${{ secrets.OPENAI_MODEL }}
+          pull_request_url: https://pr${{ env.PR_NUMBER }}.api.kerberos.lol
           overwrite_description: true

.github/workflows/release-create.yml

@@ -44,11 +44,6 @@ jobs:
         run: |
           docker tag ${{matrix.architecture}} $REPO-arch:arch-${{matrix.architecture}}-${{github.event.inputs.tag || github.ref_name}}
           docker push $REPO-arch:arch-${{matrix.architecture}}-${{github.event.inputs.tag || github.ref_name}}
-      - name: Create new manifest
-        run: docker manifest create $REPO:${{ github.event.inputs.tag || github.ref_name }} $REPO-arch:arch-${{matrix.architecture}}-${{github.event.inputs.tag || github.ref_name}}
-      - name: Create latest manifest
-        run: docker manifest create $REPO:latest $REPO-arch:arch-${{matrix.architecture}}-${{github.event.inputs.tag || github.ref_name}}
-        if: github.event.inputs.tag == 'test'
       - name: Upload artifact
         uses: actions/upload-artifact@v4
         with:

machinery/main.go

@@ -101,31 +101,35 @@ func main() {
 	switch action {
 
 	case "version":
-		log.Log.Info("main.Main(): You are currrently running Kerberos Agent " + VERSION)
-
+		{
+			log.Log.Info("main.Main(): You are currrently running Kerberos Agent " + VERSION)
+		}
 	case "discover":
-		// Convert duration to int
-		timeout, err := time.ParseDuration(timeout + "ms")
-		if err != nil {
-			log.Log.Fatal("main.Main(): could not parse timeout: " + err.Error())
-			return
+		{
+			// Convert duration to int
+			timeout, err := time.ParseDuration(timeout + "ms")
+			if err != nil {
+				log.Log.Fatal("main.Main(): could not parse timeout: " + err.Error())
+				return
+			}
+			onvif.Discover(timeout)
 		}
-		onvif.Discover(timeout)
-
 	case "decrypt":
-		log.Log.Info("main.Main(): Decrypting: " + flag.Arg(0) + " with key: " + flag.Arg(1))
-		symmetricKey := []byte(flag.Arg(1))
+		{
+			log.Log.Info("main.Main(): Decrypting: " + flag.Arg(0) + " with key: " + flag.Arg(1))
+			symmetricKey := []byte(flag.Arg(1))
 
-		if symmetricKey == nil || len(symmetricKey) == 0 {
-			log.Log.Fatal("main.Main(): symmetric key should not be empty")
-			return
-		}
-		if len(symmetricKey) != 32 {
-			log.Log.Fatal("main.Main(): symmetric key should be 32 bytes")
-			return
-		}
+			if len(symmetricKey) == 0 {
+				log.Log.Fatal("main.Main(): symmetric key should not be empty")
+				return
+			}
+			if len(symmetricKey) != 32 {
+				log.Log.Fatal("main.Main(): symmetric key should be 32 bytes")
+				return
+			}
 
-		utils.Decrypt(flag.Arg(0), symmetricKey)
+			utils.Decrypt(flag.Arg(0), symmetricKey)
+		}
 
 	case "run":
 		{
@@ -213,6 +217,8 @@ func main() {
 			routers.StartWebserver(configDirectory, &configuration, &communication, &capture)
 		}
 	default:
-		log.Log.Error("main.Main(): Sorry I don't understand :(")
+		{
+			log.Log.Error("main.Main(): Sorry I don't understand :(")
+		}
 	}
 }

machinery/src/capture/gortsplib.go

@@ -85,12 +85,8 @@ type Golibrtsp struct {
 
 	Streams []packets.Stream
 
-	// FPS calculation fields
-	lastFrameTime    time.Time
-	frameTimeBuffer  []time.Duration
-	frameBufferSize  int
-	frameBufferIndex int
-	fpsMutex         sync.Mutex
+	// Per-stream FPS calculation (keyed by stream index)
+	fpsTrackers map[int8]*fpsTracker
 
 	// I-frame interval tracking fields
 	packetsSinceLastKeyframe int
@@ -101,6 +97,78 @@ type Golibrtsp struct {
 	keyframeMutex            sync.Mutex
 }
 
+// fpsTracker holds per-stream state for PTS-based FPS calculation.
+// Each video stream (H264 / H265) gets its own tracker so PTS
+// samples from different codecs never interleave.
+type fpsTracker struct {
+	mu              sync.Mutex
+	lastPTS         time.Duration
+	hasPTS          bool
+	frameTimeBuffer []time.Duration
+	bufferSize      int
+	bufferIndex     int
+	cachedFPS       float64 // latest computed FPS
+}
+
+func newFPSTracker(bufferSize int) *fpsTracker {
+	return &fpsTracker{
+		frameTimeBuffer: make([]time.Duration, bufferSize),
+		bufferSize:      bufferSize,
+	}
+}
+
+// update records a new PTS sample and returns the latest FPS estimate.
+// It must be called once per complete decoded frame (after Decode()
+// succeeds), not on every RTP packet fragment.
+func (ft *fpsTracker) update(pts time.Duration) float64 {
+	ft.mu.Lock()
+	defer ft.mu.Unlock()
+
+	if !ft.hasPTS {
+		ft.lastPTS = pts
+		ft.hasPTS = true
+		return 0
+	}
+
+	interval := pts - ft.lastPTS
+	ft.lastPTS = pts
+
+	// Skip invalid intervals (zero, negative, or very large which
+	// indicate a PTS discontinuity or wrap).
+	if interval <= 0 || interval > 5*time.Second {
+		return ft.cachedFPS
+	}
+
+	ft.frameTimeBuffer[ft.bufferIndex] = interval
+	ft.bufferIndex = (ft.bufferIndex + 1) % ft.bufferSize
+
+	var totalInterval time.Duration
+	validSamples := 0
+	for _, iv := range ft.frameTimeBuffer {
+		if iv > 0 {
+			totalInterval += iv
+			validSamples++
+		}
+	}
+	if validSamples == 0 {
+		return ft.cachedFPS
+	}
+	avgInterval := totalInterval / time.Duration(validSamples)
+	if avgInterval == 0 {
+		return ft.cachedFPS
+	}
+
+	ft.cachedFPS = float64(time.Second) / float64(avgInterval)
+	return ft.cachedFPS
+}
+
+// fps returns the most recent FPS estimate without recording a new sample.
+func (ft *fpsTracker) fps() float64 {
+	ft.mu.Lock()
+	defer ft.mu.Unlock()
+	return ft.cachedFPS
+}
+
 // Init function
 var H264FrameDecoder *Decoder
 var H265FrameDecoder *Decoder
@@ -548,18 +616,17 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 
 			if len(rtppkt.Payload) > 0 {
 
-				// decode timestamp
-				pts, ok := g.Client.PacketPTS(g.VideoH264Media, rtppkt)
-				pts2, ok := g.Client.PacketPTS2(g.VideoH264Media, rtppkt)
-				if !ok {
-					log.Log.Debug("capture.golibrtsp.Start(): " + "unable to get PTS")
+				// decode timestamps — validate each call separately
+				pts, okPTS := g.Client.PacketPTS(g.VideoH264Media, rtppkt)
+				pts2, okPTS2 := g.Client.PacketPTS2(g.VideoH264Media, rtppkt)
+				if !okPTS2 {
+					log.Log.Debug("capture.golibrtsp.Start(): unable to get PTS2 from PacketPTS2")
 					return
 				}
 
-				// Extract access units from RTP packets
-				// We need to do this, because the decoder expects a full
-				// access unit. Once we have a full access unit, we can
-				// decode it, and know if it's a keyframe or not.
+				// Extract access units from RTP packets.
+				// We need a complete access unit to determine whether
+				// this is a keyframe.
 				au, errDecode := g.VideoH264Decoder.Decode(rtppkt)
 				if errDecode != nil {
 					if errDecode != rtph264.ErrNonStartingPacketAndNoPrevious && errDecode != rtph264.ErrMorePacketsNeeded {
@@ -568,6 +635,18 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 					return
 				}
 
+				// Frame is complete — update per-stream FPS from PTS.
+				if okPTS {
+					ft := g.fpsTrackers[g.VideoH264Index]
+					if ft == nil {
+						ft = newFPSTracker(30)
+						g.fpsTrackers[g.VideoH264Index] = ft
+					}
+					if ptsFPS := ft.update(pts); ptsFPS > 0 && ptsFPS <= 120 {
+						g.Streams[g.VideoH264Index].FPS = ptsFPS
+					}
+				}
+
 				// We'll need to read out a few things.
 				// prepend an AUD. This is required by some players
 				filteredAU = [][]byte{
@@ -578,8 +657,10 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 				nonIDRPresent := false
 				idrPresent := false
 
+				var naluTypes []string
 				for _, nalu := range au {
 					typ := h264.NALUType(nalu[0] & 0x1F)
+					naluTypes = append(naluTypes, fmt.Sprintf("%s(%d,sz=%d)", typ.String(), int(typ), len(nalu)))
 					switch typ {
 					case h264.NALUTypeAccessUnitDelimiter:
 						continue
@@ -626,6 +707,11 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 					return
 				}
 
+				if idrPresent {
+					log.Log.Debug(fmt.Sprintf("capture.golibrtsp.Start(%s): IDR frame NALUs: [%s]",
+						streamType, fmt.Sprintf("%v", naluTypes)))
+				}
+
 				// Convert to packet.
 				enc, err := h264.AnnexBMarshal(filteredAU)
 				if err != nil {
@@ -651,7 +737,11 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 				keyframeInterval := g.trackKeyframeInterval(idrPresent)
 				if idrPresent && keyframeInterval > 0 {
 					avgInterval := g.getAverageKeyframeInterval()
-					gopDuration := float64(keyframeInterval) / g.Streams[g.VideoH265Index].FPS
+					fps := g.Streams[g.VideoH264Index].FPS
+					if fps <= 0 {
+						fps = 25.0 // Default fallback FPS
+					}
+					gopDuration := float64(keyframeInterval) / fps
 					gopSize := int(avgInterval) // Store GOP size in a separate variable
 					g.Streams[g.VideoH264Index].GopSize = gopSize
 					log.Log.Debug(fmt.Sprintf("capture.golibrtsp.Start(%s): Keyframe interval=%d packets, Avg=%.1f, GOP=%.1fs, GOPSize=%d",
@@ -716,18 +806,17 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 
 			if len(rtppkt.Payload) > 0 {
 
-				// decode timestamp
-				pts, ok := g.Client.PacketPTS(g.VideoH265Media, rtppkt)
-				pts2, ok := g.Client.PacketPTS2(g.VideoH265Media, rtppkt)
-				if !ok {
-					log.Log.Debug("capture.golibrtsp.Start(): " + "unable to get PTS")
+				// decode timestamps — validate each call separately
+				pts, okPTS := g.Client.PacketPTS(g.VideoH265Media, rtppkt)
+				pts2, okPTS2 := g.Client.PacketPTS2(g.VideoH265Media, rtppkt)
+				if !okPTS2 {
+					log.Log.Debug("capture.golibrtsp.Start(): unable to get PTS")
 					return
 				}
 
-				// Extract access units from RTP packets
-				// We need to do this, because the decoder expects a full
-				// access unit. Once we have a full access unit, we can
-				// decode it, and know if it's a keyframe or not.
+				// Extract access units from RTP packets.
+				// We need a complete access unit to determine whether
+				// this is a keyframe.
 				au, errDecode := g.VideoH265Decoder.Decode(rtppkt)
 				if errDecode != nil {
 					if errDecode != rtph265.ErrNonStartingPacketAndNoPrevious && errDecode != rtph265.ErrMorePacketsNeeded {
@@ -736,6 +825,18 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 					return
 				}
 
+				// Frame is complete — update per-stream FPS from PTS.
+				if okPTS {
+					ft := g.fpsTrackers[g.VideoH265Index]
+					if ft == nil {
+						ft = newFPSTracker(30)
+						g.fpsTrackers[g.VideoH265Index] = ft
+					}
+					if ptsFPS := ft.update(pts); ptsFPS > 0 && ptsFPS <= 120 {
+						g.Streams[g.VideoH265Index].FPS = ptsFPS
+					}
+				}
+
 				filteredAU = [][]byte{
 					{byte(h265.NALUType_AUD_NUT) << 1, 1, 0x50},
 				}
@@ -796,7 +897,11 @@ func (g *Golibrtsp) Start(ctx context.Context, streamType string, queue *packets
 				keyframeInterval := g.trackKeyframeInterval(isRandomAccess)
 				if isRandomAccess && keyframeInterval > 0 {
 					avgInterval := g.getAverageKeyframeInterval()
-					gopDuration := float64(keyframeInterval) / g.Streams[g.VideoH265Index].FPS
+					fps := g.Streams[g.VideoH265Index].FPS
+					if fps <= 0 {
+						fps = 25.0 // Default fallback FPS
+					}
+					gopDuration := float64(keyframeInterval) / fps
 					gopSize := int(avgInterval) // Store GOP size in a separate variable
 					g.Streams[g.VideoH265Index].GopSize = gopSize
 					log.Log.Debug(fmt.Sprintf("capture.golibrtsp.Start(%s): Keyframe interval=%d packets, Avg=%.1f, GOP=%.1fs, GOPSize=%d",
@@ -1179,10 +1284,11 @@ func WriteMPEG4Audio(forma *format.MPEG4Audio, aus [][]byte) ([]byte, error) {
 
 // Initialize FPS calculation buffers
 func (g *Golibrtsp) initFPSCalculation() {
-	g.frameBufferSize = 30 // Store last 30 frame intervals
-	g.frameTimeBuffer = make([]time.Duration, g.frameBufferSize)
-	g.frameBufferIndex = 0
-	g.lastFrameTime = time.Time{}
+	// Ensure the per-stream FPS trackers map exists. Individual trackers
+	// can be created lazily when a given stream index is first used.
+	if g.fpsTrackers == nil {
+		g.fpsTrackers = make(map[int8]*fpsTracker)
+	}
 
 	// Initialize I-frame interval tracking
 	g.keyframeBufferSize = 10 // Store last 10 keyframe intervals
@@ -1192,50 +1298,11 @@ func (g *Golibrtsp) initFPSCalculation() {
 	g.lastKeyframePacketCount = 0
 }
 
-// Calculate FPS from frame timestamps
-func (g *Golibrtsp) calculateFPSFromTimestamps() float64 {
-	g.fpsMutex.Lock()
-	defer g.fpsMutex.Unlock()
-
-	if g.lastFrameTime.IsZero() {
-		g.lastFrameTime = time.Now()
-		return 0
-	}
-
-	now := time.Now()
-	interval := now.Sub(g.lastFrameTime)
-	g.lastFrameTime = now
-
-	// Store the interval
-	g.frameTimeBuffer[g.frameBufferIndex] = interval
-	g.frameBufferIndex = (g.frameBufferIndex + 1) % g.frameBufferSize
-
-	// Calculate average FPS from stored intervals
-	var totalInterval time.Duration
-	validSamples := 0
-
-	for _, interval := range g.frameTimeBuffer {
-		if interval > 0 {
-			totalInterval += interval
-			validSamples++
-		}
-	}
-
-	if validSamples == 0 {
-		return 0
-	}
-
-	avgInterval := totalInterval / time.Duration(validSamples)
-	if avgInterval == 0 {
-		return 0
-	}
-
-	return float64(time.Second) / float64(avgInterval)
-}
-
-// Get enhanced FPS information from SPS with fallback
+// Get enhanced FPS information from SPS with fallback to PTS-based calculation.
+// The PTS-based FPS is computed per completed frame via fpsTracker.update(),
+// so by the time this is called we already have a good estimate.
 func (g *Golibrtsp) getEnhancedFPS(sps *h264.SPS, streamIndex int8) float64 {
-	// First try to get FPS from SPS
+	// First try to get FPS from SPS VUI parameters
 	spsFPS := sps.FPS()
 
 	// Check if SPS FPS is reasonable (between 1 and 120 fps)
@@ -1244,11 +1311,13 @@ func (g *Golibrtsp) getEnhancedFPS(sps *h264.SPS, streamIndex int8) float64 {
 		return spsFPS
 	}
 
-	// Fallback to timestamp-based calculation
-	timestampFPS := g.calculateFPSFromTimestamps()
-	if timestampFPS > 0 && timestampFPS <= 120 {
-		log.Log.Debug(fmt.Sprintf("capture.golibrtsp.getEnhancedFPS(): Timestamp FPS: %.2f", timestampFPS))
-		return timestampFPS
+	// Fallback to PTS-based FPS (already calculated per-frame)
+	if ft := g.fpsTrackers[streamIndex]; ft != nil {
+		ptsFPS := ft.fps()
+		if ptsFPS > 0 && ptsFPS <= 120 {
+			log.Log.Debug(fmt.Sprintf("capture.golibrtsp.getEnhancedFPS(): PTS FPS: %.2f", ptsFPS))
+			return ptsFPS
+		}
 	}
 
 	// Return SPS FPS even if it seems unreasonable, or default

machinery/src/capture/main.go

@@ -280,7 +280,7 @@ func HandleRecordStream(queue *packets.Queue, configDirectory string, configurat
 					vpsNALUS := configuration.Config.Capture.IPCamera.VPSNALUs
 
 					// Create a video file, and set the dimensions.
-					mp4Video = video.NewMP4(fullName, spsNALUS, ppsNALUS, vpsNALUS)
+					mp4Video = video.NewMP4(fullName, spsNALUS, ppsNALUS, vpsNALUS, configuration.Config.Capture.MaxLengthRecording)
 					mp4Video.SetWidth(width)
 					mp4Video.SetHeight(height)
 
@@ -500,7 +500,7 @@ func HandleRecordStream(queue *packets.Queue, configDirectory string, configurat
 				vpsNALUS := configuration.Config.Capture.IPCamera.VPSNALUs
 
 				// Create a video file, and set the dimensions.
-				mp4Video := video.NewMP4(fullName, spsNALUS, ppsNALUS, vpsNALUS)
+				mp4Video := video.NewMP4(fullName, spsNALUS, ppsNALUS, vpsNALUS, configuration.Config.Capture.MaxLengthRecording)
 				mp4Video.SetWidth(width)
 				mp4Video.SetHeight(height)
 

machinery/src/cloud/Cloud.go

@@ -804,6 +804,12 @@ func HandleLiveStreamHD(livestreamCursor *packets.QueueCursor, configuration *mo
 			streams, _ := rtspClient.GetStreams()
 			videoTrack := webrtc.NewVideoTrack(streams)
 			audioTrack := webrtc.NewAudioTrack(streams)
+
+			if videoTrack == nil && audioTrack == nil {
+				log.Log.Error("cloud.HandleLiveStreamHD(): failed to create both video and audio tracks")
+				return
+			}
+
 			go webrtc.WriteToTrack(livestreamCursor, configuration, communication, mqttClient, videoTrack, audioTrack, rtspClient)
 
 			if config.Capture.ForwardWebRTC == "true" {

machinery/src/routers/mqtt/main.go

@@ -123,7 +123,6 @@ func ConfigureMQTT(configDirectory string, configuration *models.Configuration,
 			opts.SetClientID(mqttClientID)
 			log.Log.Info("routers.mqtt.main.ConfigureMQTT(): Set ClientID " + mqttClientID)
 			rand.Seed(time.Now().UnixNano())
-			webrtc.CandidateArrays = make(map[string](chan string))
 
 			opts.OnConnect = func(c mqtt.Client) {
 				// We managed to connect to the MQTT broker, hurray!

machinery/src/video/mp4.go

@@ -22,52 +22,73 @@ import (
 
 var LastPTS uint64 = 0 // Last PTS for the current segment
 
+// FragmentDurationMs is the target duration for each fragment in milliseconds.
+// Fragments will be flushed at the first keyframe after this duration has elapsed,
+// resulting in ~3 second fragments (assuming a typical GOP interval).
+const FragmentDurationMs = 3000
+
 type MP4 struct {
 	// FileName is the name of the file
-	FileName           string
-	width              int
-	height             int
-	Segments           []*mp4ff.MediaSegment // List of media segments
-	Segment            *mp4ff.MediaSegment
-	MultiTrackFragment *mp4ff.Fragment
-	TrackIDs           []uint32
-	FileWriter         *os.File
-	Writer             *bufio.Writer
-	SegmentCount       int
-	SampleCount        int
-	StartPTS           uint64
-	VideoTotalDuration uint64
-	AudioTotalDuration uint64
-	AudioPTS           uint64
-	Start              bool
-	SPSNALUs           [][]byte // SPS NALUs for H264
-	PPSNALUs           [][]byte // PPS NALUs for H264
-	VPSNALUs           [][]byte // VPS NALUs for H264
-	FreeBoxSize        int64
-	MoofBoxes          int64             // Number of moof boxes in the file
-	MoofBoxSizes       []int64           // Sizes of each moof box
-	StartTime          uint64            // Start time of the MP4 file
-	VideoTrackName     string            // Name of the video track
-	VideoTrack         int               // Track ID for the video track
-	AudioTrackName     string            // Name of the audio track
-	AudioTrack         int               // Track ID for the audio track
-	VideoFullSample    *mp4ff.FullSample // Full sample for video track
-	AudioFullSample    *mp4ff.FullSample // Full sample for audio track
-	LastAudioSampleDTS uint64            // Last PTS for audio sample
-	LastVideoSampleDTS uint64            // Last PTS for video sample
-	SampleType         string            // Type of the sample (e.g., "video", "audio", "subtitle")
+	FileName            string
+	width               int
+	height              int
+	Segments            []*mp4ff.MediaSegment // List of media segments
+	Segment             *mp4ff.MediaSegment
+	MultiTrackFragment  *mp4ff.Fragment
+	TrackIDs            []uint32
+	FileWriter          *os.File
+	Writer              *bufio.Writer
+	SegmentCount        int
+	SampleCount         int
+	StartPTS            uint64
+	VideoTotalDuration  uint64
+	AudioTotalDuration  uint64
+	AudioPTS            uint64
+	Start               bool
+	SPSNALUs            [][]byte // SPS NALUs for H264
+	PPSNALUs            [][]byte // PPS NALUs for H264
+	VPSNALUs            [][]byte // VPS NALUs for H264
+	FreeBoxSize         int64
+	FragmentStartRawPTS uint64            // Raw PTS for timing when to flush fragments
+	FragmentStartDTS    uint64            // Accumulated VideoTotalDuration at fragment start (matches tfdt)
+	MoofBoxes           int64             // Number of moof boxes in the file
+	MoofBoxSizes        []int64           // Sizes of each moof box
+	SegmentDurations    []uint64          // Duration of each segment in timescale units
+	SegmentBaseDecTimes []uint64          // Base decode time of each segment
+	StartTime           uint64            // Start time of the MP4 file
+	VideoTrackName      string            // Name of the video track
+	VideoTrack          int               // Track ID for the video track
+	AudioTrackName      string            // Name of the audio track
+	AudioTrack          int               // Track ID for the audio track
+	VideoFullSample     *mp4ff.FullSample // Full sample for video track
+	AudioFullSample     *mp4ff.FullSample // Full sample for audio track
+	LastAudioSampleDTS  uint64            // Last PTS for audio sample
+	LastVideoSampleDTS  uint64            // Last PTS for video sample
+	SampleType          string            // Type of the sample (e.g., "video", "audio", "subtitle")
 }
 
-// NewMP4 creates a new MP4 object
-func NewMP4(fileName string, spsNALUs [][]byte, ppsNALUs [][]byte, vpsNALUs [][]byte) *MP4 {
+// NewMP4 creates a new MP4 object.
+// maxDurationSec is the maximum expected recording duration in seconds,
+// used to calculate the free-box placeholder size for ftyp+moov+sidx.
+func NewMP4(fileName string, spsNALUs [][]byte, ppsNALUs [][]byte, vpsNALUs [][]byte, maxDurationSec int64) *MP4 {
 
-	init := mp4ff.NewMP4Init()
-
-	// Add a free box to the init segment
-	// Prepend a free box to the init segment with a size of 4096 bytes, so we can overwrite it later with the actual init segment.
-	freeBoxSize := 4096
+	// Calculate the placeholder size needed at the start of the file.
+	// Components:
+	//   ftyp:  ~32 bytes
+	//   moov:  ~1500 bytes (mvhd + mvex + video trak + audio trak + UUID)
+	//   sidx:  24 bytes fixed + 12 bytes per segment reference
+	// Segments are ~FragmentDurationMs each, so:
+	//   numSegments = ceil(maxDurationSec * 1000 / FragmentDurationMs) + 1 (safety margin)
+	//   sidxSize    = 24 + 12 * numSegments
+	baseSize := int64(2560) // ftyp + moov + extra headroom for large UUID signatures
+	numSegments := int64(0)
+	if maxDurationSec > 0 {
+		// Use integer ceiling division to avoid underestimating the number of segments.
+		numSegments = ((maxDurationSec*1000)+FragmentDurationMs-1)/FragmentDurationMs + 1
+	}
+	sidxSize := int64(24 + 12*numSegments)
+	freeBoxSize := int(baseSize + sidxSize)
 	free := mp4ff.NewFreeBox(make([]byte, freeBoxSize))
-	init.AddChild(free)
 
 	// Create a writer
 	ofd, err := os.Create(fileName)
@@ -77,16 +98,15 @@ func NewMP4(fileName string, spsNALUs [][]byte, ppsNALUs [][]byte, vpsNALUs [][]
 	// Create a buffered writer
 	bufferedWriter := bufio.NewWriterSize(ofd, 64*1024) // 64KB buffer
 
-	// We will write the empty init segment to the file
-	// so we can overwrite it later with the actual init segment.
-	err = init.Encode(bufferedWriter)
+	// Write the free box placeholder at the start of the file
+	err = free.Encode(bufferedWriter)
 	if err != nil {
 	}
 
 	return &MP4{
 		FileName:    fileName,
 		StartTime:   uint64(time.Now().Unix()),
-		FreeBoxSize: int64(freeBoxSize),
+		FreeBoxSize: int64(freeBoxSize) + 8, // payload + 8 byte box header
 		FileWriter:  ofd,
 		Writer:      bufferedWriter,
 		SPSNALUs:    spsNALUs,
@@ -130,42 +150,110 @@ func (mp4 *MP4) AddAudioTrack(codec string) uint32 {
 func (mp4 *MP4) AddMediaSegment(segNr int) {
 }
 
+// flushPendingVideoSample writes the pending video sample to the current fragment.
+// If nextPTS is provided (non-zero), it calculates duration from the PTS difference.
+// If nextPTS is 0 (e.g., at Close time), it uses the last known duration.
+// Returns true if a sample was flushed, false if there was no pending sample.
+func (mp4 *MP4) flushPendingVideoSample(nextPTS uint64) bool {
+	if mp4.VideoFullSample == nil || mp4.MultiTrackFragment == nil {
+		return false
+	}
+
+	var duration uint64
+	if nextPTS > 0 && nextPTS > mp4.VideoFullSample.DecodeTime {
+		duration = nextPTS - mp4.VideoFullSample.DecodeTime
+	} else {
+		// No valid nextPTS (Close case) or PTS went backwards (jitter/discontinuity)
+		if nextPTS > 0 {
+			log.Log.Warning(fmt.Sprintf("mp4.flushPendingVideoSample(): video PTS went backwards or zero duration (nextPTS=%d, prevDTS=%d), using last known duration", nextPTS, mp4.VideoFullSample.DecodeTime))
+		}
+		duration = mp4.LastVideoSampleDTS
+		if duration == 0 {
+			duration = 33 // Default ~30fps frame duration
+		}
+	}
+
+	mp4.LastVideoSampleDTS = duration
+	mp4.VideoTotalDuration += duration
+	mp4.VideoFullSample.DecodeTime = mp4.VideoTotalDuration - duration
+	mp4.VideoFullSample.Sample.Dur = uint32(duration)
+
+	err := mp4.MultiTrackFragment.AddFullSampleToTrack(*mp4.VideoFullSample, uint32(mp4.VideoTrack))
+	if err != nil {
+		log.Log.Error("mp4.flushPendingVideoSample(): error adding sample: " + err.Error())
+	}
+
+	mp4.VideoFullSample = nil
+	return true
+}
+
 func (mp4 *MP4) AddSampleToTrack(trackID uint32, isKeyframe bool, data []byte, pts uint64) error {
 
+	if isKeyframe && trackID == uint32(mp4.VideoTrack) {
+		log.Log.Debug(fmt.Sprintf("mp4.AddSampleToTrack(): KEYFRAME received - track=%d, PTS=%d, size=%d, sampleCount=%d",
+			trackID, pts, len(data), mp4.SampleCount))
+	}
+
 	if isKeyframe {
 
-		// Write the segment to the file
+		// Determine whether to start a new fragment.
+		// We only flush at a keyframe boundary once at least FragmentDurationMs
+		// of content has been accumulated, resulting in ~3 second fragments.
+		elapsed := uint64(0)
 		if mp4.Start {
-			mp4.MoofBoxes = mp4.MoofBoxes + 1
-			mp4.MoofBoxSizes = append(mp4.MoofBoxSizes, int64(mp4.Segment.Size()))
-			err := mp4.Segment.Encode(mp4.Writer)
-			if err != nil {
-				log.Log.Error("mp4.AddSampleToTrack(): error encoding segment: " + err.Error())
+			elapsed = pts - mp4.FragmentStartRawPTS
+		}
+		shouldFlush := !mp4.Start || elapsed >= FragmentDurationMs
+
+		if shouldFlush {
+			// Write the previous segment to the file
+			if mp4.Start {
+				// IMPORTANT: Add any pending video sample to the current segment BEFORE flushing.
+				// This ensures the segment contains all frames up to (but not including) this keyframe,
+				// and the new segment will start cleanly with this keyframe.
+				if trackID == uint32(mp4.VideoTrack) {
+					mp4.flushPendingVideoSample(pts)
+				}
+
+				mp4.MoofBoxes = mp4.MoofBoxes + 1
+				mp4.MoofBoxSizes = append(mp4.MoofBoxSizes, int64(mp4.Segment.Size()))
+				// Track the segment's duration and base decode time for sidx.
+				// Use accumulated VideoTotalDuration which matches the tfdt values
+				// in the trun boxes, NOT raw PTS from the camera.
+				segDuration := mp4.VideoTotalDuration - mp4.FragmentStartDTS
+				mp4.SegmentDurations = append(mp4.SegmentDurations, segDuration)
+				mp4.SegmentBaseDecTimes = append(mp4.SegmentBaseDecTimes, mp4.FragmentStartDTS)
+				err := mp4.Segment.Encode(mp4.Writer)
+				if err != nil {
+					log.Log.Error("mp4.AddSampleToTrack(): error encoding segment: " + err.Error())
+				}
+				mp4.Segments = append(mp4.Segments, mp4.Segment)
 			}
-			mp4.Segments = append(mp4.Segments, mp4.Segment)
+
+			mp4.Start = true
+
+			// Increment the segment count
+			mp4.SegmentCount = mp4.SegmentCount + 1
+
+			// Create a new media segment
+			seg := mp4ff.NewMediaSegment()
+
+			// Create a video fragment
+			multiTrackFragment, err := mp4ff.CreateMultiTrackFragment(uint32(mp4.SegmentCount), mp4.TrackIDs)
+			if err != nil {
+				log.Log.Error("mp4.AddSampleToTrack(): error creating multi track fragment: " + err.Error())
+			}
+			mp4.MultiTrackFragment = multiTrackFragment
+			seg.AddFragment(multiTrackFragment)
+
+			// Set to MP4 struct
+			mp4.Segment = seg
+
+			// Set the start PTS for the next segment
+			mp4.StartPTS = pts
+			mp4.FragmentStartRawPTS = pts
+			mp4.FragmentStartDTS = mp4.VideoTotalDuration
 		}
-
-		mp4.Start = true
-
-		// Increment the segment count
-		mp4.SegmentCount = mp4.SegmentCount + 1
-
-		// Create a new media segment
-		seg := mp4ff.NewMediaSegment()
-
-		// Create a video fragment
-		multiTrackFragment, err := mp4ff.CreateMultiTrackFragment(uint32(mp4.SegmentCount), mp4.TrackIDs) // Assuming 1 for video track and 2 for audio track
-		if err != nil {
-			log.Log.Error("mp4.AddSampleToTrack(): error creating multi track fragment: " + err.Error())
-		}
-		mp4.MultiTrackFragment = multiTrackFragment
-		seg.AddFragment(multiTrackFragment)
-
-		// Set to MP4 struct
-		mp4.Segment = seg
-
-		// Set the start PTS for the next segment
-		mp4.StartPTS = pts
 	}
 
 	if mp4.Start {
@@ -182,18 +270,10 @@ func (mp4 *MP4) AddSampleToTrack(trackID uint32, isKeyframe bool, data []byte, p
 			}
 
 			if err == nil {
+				// Flush previous pending sample before storing the new one
 				if mp4.VideoFullSample != nil {
-					duration := pts - mp4.VideoFullSample.DecodeTime
-					log.Log.Debug("Adding sample to track " + fmt.Sprintf("%d, PTS: %d, Duration: %d, size: %d, Keyframe: %t", trackID, pts, duration, len(lengthPrefixed), isKeyframe))
-
-					mp4.LastVideoSampleDTS = duration
-					mp4.VideoTotalDuration += duration
-					mp4.VideoFullSample.DecodeTime = mp4.VideoTotalDuration - duration
-					mp4.VideoFullSample.Sample.Dur = uint32(duration)
-					err := mp4.MultiTrackFragment.AddFullSampleToTrack(*mp4.VideoFullSample, trackID)
-					if err != nil {
-						log.Log.Error("mp4.AddSampleToTrack(): error adding sample to track " + fmt.Sprintf("%d: %v", trackID, err))
-					}
+					log.Log.Debug("Adding sample to track " + fmt.Sprintf("%d, PTS: %d, size: %d, Keyframe: %t", trackID, pts, len(lengthPrefixed), isKeyframe))
+					mp4.flushPendingVideoSample(pts)
 				}
 
 				// Set the sample data
@@ -263,8 +343,47 @@ func (mp4 *MP4) Close(config *models.Config) {
 		log.Log.Error("mp4.Close(): no video or audio samples added, cannot create MP4 file")
 	}
 
+	// Add final pending samples before closing
+	if mp4.Segment != nil {
+		// Add final video sample if pending (pass 0 as nextPTS to use last known duration)
+		mp4.flushPendingVideoSample(0)
+
+		// Add final audio sample if pending
+		if mp4.AudioFullSample != nil && mp4.AudioTrack > 0 {
+			SplitAACFrame(mp4.AudioFullSample.Data, func(started bool, aac []byte) {
+				sampleToAdd := *mp4.AudioFullSample
+				dts := mp4.LastAudioSampleDTS
+				if dts == 0 {
+					dts = 1024 // Default AAC frame duration
+				}
+				mp4.AudioTotalDuration += dts
+				mp4.AudioPTS += dts
+				sampleToAdd.Data = aac[7:]
+				sampleToAdd.DecodeTime = mp4.AudioPTS - dts
+				sampleToAdd.Sample.Dur = uint32(dts)
+				sampleToAdd.Sample.Size = uint32(len(aac[7:]))
+				err := mp4.MultiTrackFragment.AddFullSampleToTrack(sampleToAdd, uint32(mp4.AudioTrack))
+				if err != nil {
+					log.Log.Error("mp4.Close(): error adding final audio sample: " + err.Error())
+				}
+			})
+			mp4.AudioFullSample = nil
+		}
+	}
+
 	// Encode the last segment
 	if mp4.Segment != nil {
+		// Track the last segment's size, duration and base decode time.
+		// Use accumulated VideoTotalDuration which matches tfdt values.
+		mp4.MoofBoxes = mp4.MoofBoxes + 1
+		mp4.MoofBoxSizes = append(mp4.MoofBoxSizes, int64(mp4.Segment.Size()))
+		lastSegDuration := mp4.VideoTotalDuration - mp4.FragmentStartDTS
+		if lastSegDuration == 0 {
+			lastSegDuration = mp4.LastVideoSampleDTS
+		}
+		mp4.SegmentDurations = append(mp4.SegmentDurations, lastSegDuration)
+		mp4.SegmentBaseDecTimes = append(mp4.SegmentBaseDecTimes, mp4.FragmentStartDTS)
+
 		err := mp4.Segment.Encode(mp4.Writer)
 		if err != nil {
 			log.Log.Error("mp4.Close(): error encoding last segment: " + err.Error())
@@ -272,10 +391,14 @@ func (mp4 *MP4) Close(config *models.Config) {
 	}
 
 	mp4.Writer.Flush()
-	defer mp4.FileWriter.Close()
+	// Ensure all segment data is on disk before we overwrite the placeholder at offset 0.
+	if err := mp4.FileWriter.Sync(); err != nil {
+		log.Log.Error("mp4.Close(): error syncing file: " + err.Error())
+	}
 
 	// Now we have all the moof and mdat boxes written to the file.
-	// We can now generate the ftyp and moov boxes, and replace it with the free box we added earlier (size of 2048 bytes).
+	// We build the ftyp + moov init segment and write it at the start,
+	// overwriting the free box placeholder we reserved in NewMP4.
 	init := mp4ff.NewMP4Init()
 
 	// Create a new ftyp box
@@ -316,8 +439,10 @@ func (mp4 *MP4) Close(config *models.Config) {
 		if err != nil {
 		}
 		init.Moov.Traks[0].Tkhd.Duration = mp4.VideoTotalDuration
+		init.Moov.Traks[0].Tkhd.Width = mp4ff.Fixed32(uint32(mp4.width) << 16)
+		init.Moov.Traks[0].Tkhd.Height = mp4ff.Fixed32(uint32(mp4.height) << 16)
 		init.Moov.Traks[0].Mdia.Hdlr.Name = "agent " + utils.VERSION
-		//init.Moov.Traks[0].Mdia.Mdhd.Duration = mp4.VideoTotalDuration
+		init.Moov.Traks[0].Mdia.Mdhd.Duration = mp4.VideoTotalDuration
 	case "H265", "HVC1":
 		init.AddEmptyTrack(videoTimescale, "video", "und")
 		includePS := true
@@ -325,8 +450,10 @@ func (mp4 *MP4) Close(config *models.Config) {
 		if err != nil {
 		}
 		init.Moov.Traks[0].Tkhd.Duration = mp4.VideoTotalDuration
+		init.Moov.Traks[0].Tkhd.Width = mp4ff.Fixed32(uint32(mp4.width) << 16)
+		init.Moov.Traks[0].Tkhd.Height = mp4ff.Fixed32(uint32(mp4.height) << 16)
 		init.Moov.Traks[0].Mdia.Hdlr.Name = "agent " + utils.VERSION
-		//init.Moov.Traks[0].Mdia.Mdhd.Duration = mp4.VideoTotalDuration
+		init.Moov.Traks[0].Mdia.Mdhd.Duration = mp4.VideoTotalDuration
 	}
 
 	// Try adding audio track if available
@@ -345,7 +472,7 @@ func (mp4 *MP4) Close(config *models.Config) {
 		}
 		init.Moov.Traks[1].Tkhd.Duration = mp4.AudioTotalDuration
 		init.Moov.Traks[1].Mdia.Hdlr.Name = "agent " + utils.VERSION
-		//init.Moov.Traks[1].Mdia.Mdhd.Duration = mp4.AudioTotalDuration
+		init.Moov.Traks[1].Mdia.Mdhd.Duration = mp4.AudioTotalDuration
 	}
 
 	// Try adding subtitle track if available
@@ -423,126 +550,94 @@ func (mp4 *MP4) Close(config *models.Config) {
 		}
 	}
 
-	// We will also calculate the SIDX box, which is a segment index box that contains information about the segments in the file.
-	// This is useful for seeking in the file, and for streaming the file.
-	/*sidx := &mp4ff.SidxBox{
-		Version:                  0,
-		Flags:                    0,
-		ReferenceID:              0,
-		Timescale:                videoTimescale,
-		EarliestPresentationTime: 0,
-		FirstOffset:              0,
-		SidxRefs:                 make([]mp4ff.SidxRef, 0),
-	}
-	referenceTrak := init.Moov.Trak
-	trex, ok := init.Moov.Mvex.GetTrex(referenceTrak.Tkhd.TrackID)
-	if !ok {
-		// We have an issue.
+	// Build a Segment Index (sidx) box so players can seek directly to any
+	// fragment without scanning the entire file.
+	if len(mp4.SegmentDurations) > 0 {
+		sidx := &mp4ff.SidxBox{
+			Version:                  1,
+			Flags:                    0,
+			ReferenceID:              uint32(mp4.VideoTrack),
+			Timescale:                videoTimescale,
+			EarliestPresentationTime: 0,
+			FirstOffset:              0,
+			SidxRefs:                 make([]mp4ff.SidxRef, 0, len(mp4.SegmentDurations)),
+		}
+		for i, dur := range mp4.SegmentDurations {
+			sidx.SidxRefs = append(sidx.SidxRefs, mp4ff.SidxRef{
+				ReferenceType:      0, // media reference
+				ReferencedSize:     uint32(mp4.MoofBoxSizes[i]),
+				SubSegmentDuration: uint32(dur),
+				StartsWithSAP:      1,
+				SAPType:            1,
+			})
+		}
+		init.AddChild(sidx)
 	}
 
-	segDatas, err := findSegmentData(mp4.Segments, referenceTrak, trex)
-	if err != nil {
-		// We have an issue.
-	}
-	fillSidx(sidx, referenceTrak, segDatas, true)
-
-	// Add the SIDX box to the moov box
-	init.AddChild(sidx)*/
-
-	// Get a bit slice writer for the init segment
-	// Get a byte buffer of FreeBoxSize bytes to write the init segment
-	buffer := bytes.NewBuffer(make([]byte, 0))
-	init.Encode(buffer)
-
-	// The first FreeBoxSize bytes of the file is a free box, so we can read it and replace it with the moov box.
-	// The init box might not be FreeBoxSize bytes, so we need to read the first FreeBoxSize bytes and then replace it with the moov box.
-	// while the remaining bytes are for a new free box.
-	// Write the init segment at the beginning of the file, replacing the free box
-	if _, err := mp4.FileWriter.WriteAt(buffer.Bytes(), 0); err != nil {
+	// Encode the ftyp + moov + sidx into a buffer to measure the total size.
+	// Then compute the correct sidx.FirstOffset (the gap between the end of
+	// the sidx box and the first moof, occupied by the trailing free box)
+	// and re-encode with the corrected value.
+	var initBuf bytes.Buffer
+	if err := init.Encode(&initBuf); err != nil {
+		log.Log.Error("mp4.Close(): error encoding init segment: " + err.Error())
 	}
 
-	// Calculate the remaining size for the free box
-	remainingSize := mp4.FreeBoxSize - int64(buffer.Len())
-	if remainingSize > 0 {
-		newFreeBox := mp4ff.NewFreeBox(make([]byte, remainingSize))
+	initSize := int64(initBuf.Len())
+
+	// The sidx.FirstOffset is defined as the distance (in bytes) from the
+	// anchor point (first byte after the sidx box) to the first byte of
+	// the first referenced moof/mdat. Since sidx is the last box in init,
+	// the anchor point is at initSize, and the first moof is at FreeBoxSize.
+	if len(mp4.SegmentDurations) > 0 {
+		if mp4.FreeBoxSize < initSize {
+			// Avoid computing a negative offset and wrapping it to uint64.
+			log.Log.Error("mp4.Close(): FreeBoxSize is smaller than initSize; skipping sidx FirstOffset adjustment")
+		} else {
+			firstOffset := uint64(mp4.FreeBoxSize - initSize)
+			// Find the sidx we added and update its FirstOffset
+			for _, child := range init.Children {
+				if sidxBox, ok := child.(*mp4ff.SidxBox); ok {
+					sidxBox.FirstOffset = firstOffset
+					break
+				}
+			}
+			// Re-encode with the corrected FirstOffset (same size, no layout change)
+			initBuf.Reset()
+			if err := init.Encode(&initBuf); err != nil {
+				log.Log.Error("mp4.Close(): error re-encoding init segment: " + err.Error())
+			}
+			initSize = int64(initBuf.Len())
+		}
+	}
+
+	if initSize > mp4.FreeBoxSize {
+		log.Log.Error(fmt.Sprintf("mp4.Close(): init segment (%d bytes) exceeds reserved space (%d bytes), file may be corrupt", initSize, mp4.FreeBoxSize))
+	}
+
+	// Write the init segment at the beginning of the file, overwriting the free box placeholder.
+	if _, err := mp4.FileWriter.WriteAt(initBuf.Bytes(), 0); err != nil {
+		log.Log.Error("mp4.Close(): error writing init segment: " + err.Error())
+	}
+
+	// Fill any remaining reserved space with a new (smaller) free box so
+	// the byte offsets of the moof/mdat boxes that follow are preserved.
+	remainingSize := mp4.FreeBoxSize - initSize
+	if remainingSize >= 8 { // minimum box size is 8 bytes (header only)
+		newFree := mp4ff.NewFreeBox(make([]byte, remainingSize-8))
 		var freeBuf bytes.Buffer
-		if err := newFreeBox.Encode(&freeBuf); err != nil {
+		if err := newFree.Encode(&freeBuf); err != nil {
 			log.Log.Error("mp4.Close(): error encoding free box: " + err.Error())
 		}
-		if _, err := mp4.FileWriter.WriteAt(freeBuf.Bytes(), int64(buffer.Len())); err != nil {
+		if _, err := mp4.FileWriter.WriteAt(freeBuf.Bytes(), initSize); err != nil {
 			log.Log.Error("mp4.Close(): error writing free box: " + err.Error())
 		}
 	}
-}
 
-type segData struct {
-	startPos         uint64
-	presentationTime uint64
-	baseDecodeTime   uint64
-	dur              uint32
-	size             uint32
-}
-
-func fillSidx(sidx *mp4ff.SidxBox, refTrak *mp4ff.TrakBox, segDatas []segData, nonZeroEPT bool) {
-	ept := uint64(0)
-	if nonZeroEPT {
-		ept = segDatas[0].presentationTime
+	if err := mp4.FileWriter.Sync(); err != nil {
+		log.Log.Error("mp4.Close(): error syncing file: " + err.Error())
 	}
-	sidx.Version = 1
-	sidx.Timescale = refTrak.Mdia.Mdhd.Timescale
-	sidx.ReferenceID = 1
-	sidx.EarliestPresentationTime = ept
-	sidx.FirstOffset = 0
-	sidx.SidxRefs = make([]mp4ff.SidxRef, 0, len(segDatas))
-
-	for _, segData := range segDatas {
-		size := segData.size
-		sidx.SidxRefs = append(sidx.SidxRefs, mp4ff.SidxRef{
-			ReferencedSize:     size,
-			SubSegmentDuration: segData.dur,
-			StartsWithSAP:      1,
-			SAPType:            1,
-		})
-	}
-}
-
-// findSegmentData returns a slice of segment media data using a reference track.
-func findSegmentData(segs []*mp4ff.MediaSegment, refTrak *mp4ff.TrakBox, trex *mp4ff.TrexBox) ([]segData, error) {
-	segDatas := make([]segData, 0, len(segs))
-	for _, seg := range segs {
-		var firstCompositionTimeOffest int64
-		dur := uint32(0)
-		var baseTime uint64
-		for fIdx, frag := range seg.Fragments {
-			for _, traf := range frag.Moof.Trafs {
-				tfhd := traf.Tfhd
-				if tfhd.TrackID == refTrak.Tkhd.TrackID { // Find track that gives sidx time values
-					if fIdx == 0 {
-						baseTime = traf.Tfdt.BaseMediaDecodeTime()
-					}
-					for i, trun := range traf.Truns {
-						trun.AddSampleDefaultValues(tfhd, trex)
-						samples := trun.GetSamples()
-						for j, sample := range samples {
-							if fIdx == 0 && i == 0 && j == 0 {
-								firstCompositionTimeOffest = int64(sample.CompositionTimeOffset)
-							}
-							dur += sample.Dur
-						}
-					}
-				}
-			}
-		}
-		sd := segData{
-			startPos:         seg.StartPos,
-			presentationTime: uint64(int64(baseTime) + firstCompositionTimeOffest),
-			baseDecodeTime:   baseTime,
-			dur:              dur,
-			size:             uint32(seg.Size()),
-		}
-		segDatas = append(segDatas, sd)
-	}
-	return segDatas, nil
+	mp4.FileWriter.Close()
 }
 
 // annexBToLengthPrefixed converts Annex B formatted H264 data (with start codes)

machinery/src/webrtc/main.go

@@ -1,6 +1,7 @@
 package webrtc
 
 import (
+	"context"
 	"encoding/base64"
 	"encoding/json"
 	"io"
@@ -22,13 +23,105 @@ import (
 	pionMedia "github.com/pion/webrtc/v4/pkg/media"
 )
 
-var (
-	CandidatesMutex     sync.Mutex
-	CandidateArrays     map[string](chan string)
-	peerConnectionCount int64
-	peerConnections     map[string]*pionWebRTC.PeerConnection
+const (
+	// Channel buffer sizes
+	candidateChannelBuffer = 100
+	rtcpBufferSize         = 1500
+
+	// Timeouts and intervals
+	keepAliveTimeout = 15 * time.Second
+	defaultTimeout   = 10 * time.Second
+
+	// Track identifiers
+	trackStreamID = "kerberos-stream"
 )
 
+// ConnectionManager manages WebRTC peer connections in a thread-safe manner
+type ConnectionManager struct {
+	mu                  sync.RWMutex
+	candidateChannels   map[string]chan string
+	peerConnections     map[string]*peerConnectionWrapper
+	peerConnectionCount int64
+}
+
+// peerConnectionWrapper wraps a peer connection with additional metadata
+type peerConnectionWrapper struct {
+	conn      *pionWebRTC.PeerConnection
+	cancelCtx context.CancelFunc
+	done      chan struct{}
+	closeOnce sync.Once
+}
+
+var globalConnectionManager = NewConnectionManager()
+
+// NewConnectionManager creates a new connection manager
+func NewConnectionManager() *ConnectionManager {
+	return &ConnectionManager{
+		candidateChannels: make(map[string]chan string),
+		peerConnections:   make(map[string]*peerConnectionWrapper),
+	}
+}
+
+// GetOrCreateCandidateChannel gets or creates a candidate channel for a session
+func (cm *ConnectionManager) GetOrCreateCandidateChannel(sessionKey string) chan string {
+	cm.mu.Lock()
+	defer cm.mu.Unlock()
+
+	if ch, exists := cm.candidateChannels[sessionKey]; exists {
+		return ch
+	}
+
+	ch := make(chan string, candidateChannelBuffer)
+	cm.candidateChannels[sessionKey] = ch
+	return ch
+}
+
+// CloseCandidateChannel safely closes and removes a candidate channel
+func (cm *ConnectionManager) CloseCandidateChannel(sessionKey string) {
+	cm.mu.Lock()
+	defer cm.mu.Unlock()
+
+	if ch, exists := cm.candidateChannels[sessionKey]; exists {
+		close(ch)
+		delete(cm.candidateChannels, sessionKey)
+	}
+}
+
+// AddPeerConnection adds a peer connection to the manager
+func (cm *ConnectionManager) AddPeerConnection(sessionID string, wrapper *peerConnectionWrapper) {
+	cm.mu.Lock()
+	defer cm.mu.Unlock()
+	cm.peerConnections[sessionID] = wrapper
+}
+
+// RemovePeerConnection removes a peer connection from the manager
+func (cm *ConnectionManager) RemovePeerConnection(sessionID string) {
+	cm.mu.Lock()
+	defer cm.mu.Unlock()
+
+	if wrapper, exists := cm.peerConnections[sessionID]; exists {
+		if wrapper.cancelCtx != nil {
+			wrapper.cancelCtx()
+		}
+		delete(cm.peerConnections, sessionID)
+	}
+}
+
+// GetPeerConnectionCount returns the current count of active peer connections
+func (cm *ConnectionManager) GetPeerConnectionCount() int64 {
+	return atomic.LoadInt64(&cm.peerConnectionCount)
+}
+
+// IncrementPeerCount atomically increments the peer connection count
+func (cm *ConnectionManager) IncrementPeerCount() int64 {
+	return atomic.AddInt64(&cm.peerConnectionCount, 1)
+}
+
+// DecrementPeerCount atomically decrements the peer connection count
+func (cm *ConnectionManager) DecrementPeerCount() int64 {
+	return atomic.AddInt64(&cm.peerConnectionCount, -1)
+}
+
 type WebRTC struct {
 	Name                  string
 	StunServers           []string
@@ -46,7 +139,7 @@ func CreateWebRTC(name string, stunServers []string, turnServers []string, turnS
 		TurnServers:           turnServers,
 		TurnServersUsername:   turnServersUsername,
 		TurnServersCredential: turnServersCredential,
-		Timer:                 time.NewTimer(time.Second * 10),
+		Timer:                 time.NewTimer(defaultTimeout),
 	}
 }
 
@@ -68,19 +161,14 @@ func (w WebRTC) CreateOffer(sd []byte) pionWebRTC.SessionDescription {
 }
 
 func RegisterCandidates(key string, candidate models.ReceiveHDCandidatesPayload) {
-	// Set lock
-	CandidatesMutex.Lock()
-	_, ok := CandidateArrays[key]
-	if !ok {
-		CandidateArrays[key] = make(chan string, 100)
-	}
-	log.Log.Info("webrtc.main.HandleReceiveHDCandidates(): " + candidate.Candidate)
+	ch := globalConnectionManager.GetOrCreateCandidateChannel(key)
+
+	log.Log.Info("webrtc.main.RegisterCandidates(): " + candidate.Candidate)
 	select {
-	case CandidateArrays[key] <- candidate.Candidate:
+	case ch <- candidate.Candidate:
 	default:
-		log.Log.Info("webrtc.main.HandleReceiveHDCandidates(): channel is full.")
+		log.Log.Info("webrtc.main.RegisterCandidates(): channel is full, dropping candidate")
 	}
-	CandidatesMutex.Unlock()
 }
 
 func RegisterDefaultInterceptors(mediaEngine *pionWebRTC.MediaEngine, interceptorRegistry *interceptor.Registry) error {
@@ -107,12 +195,7 @@ func InitializeWebRTCConnection(configuration *models.Configuration, communicati
 
 	// We create a channel which will hold the candidates for this session.
 	sessionKey := config.Key + "/" + handshake.SessionID
-	CandidatesMutex.Lock()
-	_, ok := CandidateArrays[sessionKey]
-	if !ok {
-		CandidateArrays[sessionKey] = make(chan string, 100)
-	}
-	CandidatesMutex.Unlock()
+	candidateChannel := globalConnectionManager.GetOrCreateCandidateChannel(sessionKey)
 
 	// Set variables
 	hubKey := handshake.HubKey
@@ -178,81 +261,128 @@ func InitializeWebRTCConnection(configuration *models.Configuration, communicati
 
 		if err == nil && peerConnection != nil {
 
-			var videoSender *pionWebRTC.RTPSender = nil
-			if videoSender, err = peerConnection.AddTrack(videoTrack); err != nil {
-				log.Log.Error("webrtc.main.InitializeWebRTCConnection(): something went wrong while adding video track: " + err.Error())
+			// Create context for this connection
+			ctx, cancel := context.WithCancel(context.Background())
+			wrapper := &peerConnectionWrapper{
+				conn:      peerConnection,
+				cancelCtx: cancel,
+				done:      make(chan struct{}),
 			}
+
+			var videoSender *pionWebRTC.RTPSender = nil
+			if videoTrack != nil {
+				if videoSender, err = peerConnection.AddTrack(videoTrack); err != nil {
+					log.Log.Error("webrtc.main.InitializeWebRTCConnection(): error adding video track: " + err.Error())
+					cancel()
+					return
+				}
+			} else {
+				log.Log.Info("webrtc.main.InitializeWebRTCConnection(): video track is nil, skipping video")
+			}
+
 			// Read incoming RTCP packets
 			// Before these packets are returned they are processed by interceptors. For things
 			// like NACK this needs to be called.
-			go func() {
-				rtcpBuf := make([]byte, 1500)
-				for {
-					if _, _, rtcpErr := videoSender.Read(rtcpBuf); rtcpErr != nil {
-						return
+			if videoSender != nil {
+				go func() {
+					defer func() {
+						log.Log.Info("webrtc.main.InitializeWebRTCConnection(): video RTCP reader stopped")
+					}()
+					rtcpBuf := make([]byte, rtcpBufferSize)
+					for {
+						select {
+						case <-ctx.Done():
+							return
+						default:
+							if _, _, rtcpErr := videoSender.Read(rtcpBuf); rtcpErr != nil {
+								return
+							}
+						}
 					}
-				}
-			}()
+				}()
+			}
 
 			var audioSender *pionWebRTC.RTPSender = nil
-			if audioSender, err = peerConnection.AddTrack(audioTrack); err != nil {
-				log.Log.Error("webrtc.main.InitializeWebRTCConnection(): something went wrong while adding audio track: " + err.Error())
-			} // Read incoming RTCP packets
+			if audioTrack != nil {
+				if audioSender, err = peerConnection.AddTrack(audioTrack); err != nil {
+					log.Log.Error("webrtc.main.InitializeWebRTCConnection(): error adding audio track: " + err.Error())
+					cancel()
+					return
+				}
+			} else {
+				log.Log.Info("webrtc.main.InitializeWebRTCConnection(): audio track is nil, skipping audio")
+			}
+
+			// Read incoming RTCP packets
 			// Before these packets are returned they are processed by interceptors. For things
 			// like NACK this needs to be called.
-			go func() {
-				rtcpBuf := make([]byte, 1500)
-				for {
-					if _, _, rtcpErr := audioSender.Read(rtcpBuf); rtcpErr != nil {
-						return
+			if audioSender != nil {
+				go func() {
+					defer func() {
+						log.Log.Info("webrtc.main.InitializeWebRTCConnection(): audio RTCP reader stopped")
+					}()
+					rtcpBuf := make([]byte, rtcpBufferSize)
+					for {
+						select {
+						case <-ctx.Done():
+							return
+						default:
+							if _, _, rtcpErr := audioSender.Read(rtcpBuf); rtcpErr != nil {
+								return
+							}
+						}
 					}
-				}
-			}()
+				}()
+			}
 
 			peerConnection.OnConnectionStateChange(func(connectionState pionWebRTC.PeerConnectionState) {
-				if connectionState == pionWebRTC.PeerConnectionStateDisconnected || connectionState == pionWebRTC.PeerConnectionStateClosed {
-					// Set lock
-					CandidatesMutex.Lock()
-					atomic.AddInt64(&peerConnectionCount, -1)
-					_, ok := CandidateArrays[sessionKey]
-					if ok {
-						close(CandidateArrays[sessionKey])
-						delete(CandidateArrays, sessionKey)
-					}
-					// Not really needed.
-					//senders := peerConnection.GetSenders()
-					//for _, sender := range senders {
-					//	if err := peerConnection.RemoveTrack(sender); err != nil {
-					//		log.Log.Error("webrtc.main.InitializeWebRTCConnection(): something went wrong while removing track: " + err.Error())
-					//	}
-					//}
-					if err := peerConnection.Close(); err != nil {
-						log.Log.Error("webrtc.main.InitializeWebRTCConnection(): something went wrong while closing peer connection: " + err.Error())
-					}
-					peerConnections[handshake.SessionID] = nil
-					delete(peerConnections, handshake.SessionID)
-					CandidatesMutex.Unlock()
-				} else if connectionState == pionWebRTC.PeerConnectionStateConnected {
-					CandidatesMutex.Lock()
-					atomic.AddInt64(&peerConnectionCount, 1)
-					CandidatesMutex.Unlock()
-				} else if connectionState == pionWebRTC.PeerConnectionStateFailed {
-					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): ICEConnectionStateFailed")
-				}
 				log.Log.Info("webrtc.main.InitializeWebRTCConnection(): connection state changed to: " + connectionState.String())
-				log.Log.Info("webrtc.main.InitializeWebRTCConnection(): Number of peers connected (" + strconv.FormatInt(peerConnectionCount, 10) + ")")
+
+				switch connectionState {
+				case pionWebRTC.PeerConnectionStateDisconnected, pionWebRTC.PeerConnectionStateClosed:
+					wrapper.closeOnce.Do(func() {
+						count := globalConnectionManager.DecrementPeerCount()
+						log.Log.Info("webrtc.main.InitializeWebRTCConnection(): Peer disconnected. Active peers: " + string(rune(count)))
+
+						// Clean up resources
+						globalConnectionManager.CloseCandidateChannel(sessionKey)
+
+						if err := peerConnection.Close(); err != nil {
+							log.Log.Error("webrtc.main.InitializeWebRTCConnection(): error closing peer connection: " + err.Error())
+						}
+
+						globalConnectionManager.RemovePeerConnection(handshake.SessionID)
+						close(wrapper.done)
+					})
+
+				case pionWebRTC.PeerConnectionStateConnected:
+					count := globalConnectionManager.IncrementPeerCount()
+					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): Peer connected. Active peers: " + string(rune(count)))
+
+				case pionWebRTC.PeerConnectionStateFailed:
+					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): ICE connection failed")
+				}
 			})
 
 			go func() {
+				defer func() {
+					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): candidate processor stopped for session: " + handshake.SessionID)
+				}()
+
 				// Iterate over the candidates and send them to the remote client
-				// Non blocking channe
-				for candidate := range CandidateArrays[sessionKey] {
-					CandidatesMutex.Lock()
-					log.Log.Info(">>>> webrtc.main.InitializeWebRTCConnection(): Received candidate from channel: " + candidate)
-					if candidateErr := peerConnection.AddICECandidate(pionWebRTC.ICECandidateInit{Candidate: string(candidate)}); candidateErr != nil {
-						log.Log.Error("webrtc.main.InitializeWebRTCConnection(): something went wrong while adding candidate: " + candidateErr.Error())
+				for {
+					select {
+					case <-ctx.Done():
+						return
+					case candidate, ok := <-candidateChannel:
+						if !ok {
+							return
+						}
+						log.Log.Info("webrtc.main.InitializeWebRTCConnection(): Received candidate from channel: " + candidate)
+						if candidateErr := peerConnection.AddICECandidate(pionWebRTC.ICECandidateInit{Candidate: candidate}); candidateErr != nil {
+							log.Log.Error("webrtc.main.InitializeWebRTCConnection(): error adding candidate: " + candidateErr.Error())
+						}
 					}
-					CandidatesMutex.Unlock()
 				}
 			}()
 
@@ -270,22 +400,56 @@ func InitializeWebRTCConnection(configuration *models.Configuration, communicati
 
 			// When an ICE candidate is available send to the other peer using the signaling server (MQTT).
 			// The other peer will add this candidate by calling AddICECandidate
+			var hasRelayCandidates bool
 			peerConnection.OnICECandidate(func(candidate *pionWebRTC.ICECandidate) {
+
 				if candidate == nil {
+					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): ICE gathering complete (candidate is nil)")
+					if !hasRelayCandidates {
+						log.Log.Error("webrtc.main.InitializeWebRTCConnection(): WARNING - No TURN (relay) candidates were gathered! TURN servers: " +
+							config.TURNURI + ", Username: " + config.TURNUsername + ", ForceTurn: " + config.ForceTurn)
+					}
 					return
 				}
 
+				// Log candidate details for debugging
+				candidateJSON := candidate.ToJSON()
+				candidateStr := candidateJSON.Candidate
+
+				// Determine candidate type from the candidate string
+				candidateType := "unknown"
+				if candidateJSON.Candidate != "" {
+					switch candidate.Typ {
+					case pionWebRTC.ICECandidateTypeRelay:
+						candidateType = "relay"
+					case pionWebRTC.ICECandidateTypeSrflx:
+						candidateType = "srflx"
+					case pionWebRTC.ICECandidateTypeHost:
+						candidateType = "host"
+					case pionWebRTC.ICECandidateTypePrflx:
+						candidateType = "prflx"
+					}
+				}
+
+				// Track if we received any relay (TURN) candidates
+				if candidateType == "relay" {
+					hasRelayCandidates = true
+				}
+
+				log.Log.Info("webrtc.main.InitializeWebRTCConnection(): ICE candidate received - Type: " + candidateType +
+					", Candidate: " + candidateStr)
+
 				//  Create a config map
 				valueMap := make(map[string]interface{})
-				candateJSON := candidate.ToJSON()
-				candateBinary, err := json.Marshal(candateJSON)
+				candateBinary, err := json.Marshal(candidateJSON)
 				if err == nil {
 					valueMap["candidate"] = string(candateBinary)
 					// SDP is not needed to be send..
 					//valueMap["sdp"] = []byte(base64.StdEncoding.EncodeToString([]byte(answer.SDP)))
 					valueMap["session_id"] = handshake.SessionID
+					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): sending " + candidateType + " candidate to hub")
 				} else {
-					log.Log.Info("webrtc.main.InitializeWebRTCConnection(): something went wrong while marshalling candidate: " + err.Error())
+					log.Log.Error("webrtc.main.InitializeWebRTCConnection(): failed to marshal candidate: " + err.Error())
 				}
 
 				// We'll send the candidate to the hub
@@ -305,8 +469,8 @@ func InitializeWebRTCConnection(configuration *models.Configuration, communicati
 				}
 			})
 
-			// Create a channel which will be used to send candidates to the other peer
-			peerConnections[handshake.SessionID] = peerConnection
+			// Store peer connection in manager
+			globalConnectionManager.AddPeerConnection(handshake.SessionID, wrapper)
 
 			if err == nil {
 				//  Create a config map
@@ -339,7 +503,11 @@ func InitializeWebRTCConnection(configuration *models.Configuration, communicati
 
 func NewVideoTrack(streams []packets.Stream) *pionWebRTC.TrackLocalStaticSample {
 	mimeType := pionWebRTC.MimeTypeH264
-	outboundVideoTrack, _ := pionWebRTC.NewTrackLocalStaticSample(pionWebRTC.RTPCodecCapability{MimeType: mimeType}, "video", "pion124")
+	outboundVideoTrack, err := pionWebRTC.NewTrackLocalStaticSample(pionWebRTC.RTPCodecCapability{MimeType: mimeType}, "video", trackStreamID)
+	if err != nil {
+		log.Log.Error("webrtc.main.NewVideoTrack(): error creating video track: " + err.Error())
+		return nil
+	}
 	return outboundVideoTrack
 }
 
@@ -354,161 +522,245 @@ func NewAudioTrack(streams []packets.Stream) *pionWebRTC.TrackLocalStaticSample
 			mimeType = pionWebRTC.MimeTypePCMA
 		}
 	}
-	outboundAudioTrack, _ := pionWebRTC.NewTrackLocalStaticSample(pionWebRTC.RTPCodecCapability{MimeType: mimeType}, "audio", "pion124")
+	if mimeType == "" {
+		log.Log.Error("webrtc.main.NewAudioTrack(): no supported audio codec found")
+		return nil
+	}
+	outboundAudioTrack, err := pionWebRTC.NewTrackLocalStaticSample(pionWebRTC.RTPCodecCapability{MimeType: mimeType}, "audio", trackStreamID)
+	if err != nil {
+		log.Log.Error("webrtc.main.NewAudioTrack(): error creating audio track: " + err.Error())
+		return nil
+	}
 	return outboundAudioTrack
 }
 
+// streamState holds state information for the streaming process
+type streamState struct {
+	lastKeepAlive    int64
+	peerCount        int64
+	start            bool
+	receivedKeyFrame bool
+	lastAudioSample  *pionMedia.Sample
+	lastVideoSample  *pionMedia.Sample
+}
+
+// codecSupport tracks which codecs are available in the stream
+type codecSupport struct {
+	hasH264      bool
+	hasPCM_MULAW bool
+	hasAAC       bool
+	hasOpus      bool
+}
+
+// detectCodecs examines the stream to determine which codecs are available
+func detectCodecs(rtspClient capture.RTSPClient) codecSupport {
+	support := codecSupport{}
+	streams, _ := rtspClient.GetStreams()
+
+	for _, stream := range streams {
+		switch stream.Name {
+		case "H264":
+			support.hasH264 = true
+		case "PCM_MULAW":
+			support.hasPCM_MULAW = true
+		case "AAC":
+			support.hasAAC = true
+		case "OPUS":
+			support.hasOpus = true
+		}
+	}
+
+	return support
+}
+
+// hasValidCodecs checks if at least one valid video or audio codec is present
+func (cs codecSupport) hasValidCodecs() bool {
+	hasVideo := cs.hasH264
+	hasAudio := cs.hasPCM_MULAW || cs.hasAAC || cs.hasOpus
+	return hasVideo || hasAudio
+}
+
+// shouldContinueStreaming determines if streaming should continue based on keepalive and peer count
+func shouldContinueStreaming(config models.Config, state *streamState) bool {
+	if config.Capture.ForwardWebRTC != "true" {
+		return true
+	}
+
+	now := time.Now().Unix()
+	hasTimedOut := (now - state.lastKeepAlive) > int64(keepAliveTimeout.Seconds())
+	hasNoPeers := state.peerCount == 0
+
+	return !hasTimedOut && !hasNoPeers
+}
+
+// updateStreamState updates keepalive and peer count from communication channels
+func updateStreamState(communication *models.Communication, state *streamState) {
+	select {
+	case keepAliveStr := <-communication.HandleLiveHDKeepalive:
+		if val, err := strconv.ParseInt(keepAliveStr, 10, 64); err == nil {
+			state.lastKeepAlive = val
+		}
+	default:
+	}
+
+	select {
+	case peerCountStr := <-communication.HandleLiveHDPeers:
+		if val, err := strconv.ParseInt(peerCountStr, 10, 64); err == nil {
+			state.peerCount = val
+		}
+	default:
+	}
+}
+
+// writeFinalSamples writes any remaining buffered samples
+func writeFinalSamples(state *streamState, videoTrack, audioTrack *pionWebRTC.TrackLocalStaticSample) {
+	if state.lastVideoSample != nil && videoTrack != nil {
+		if err := videoTrack.WriteSample(*state.lastVideoSample); err != nil && err != io.ErrClosedPipe {
+			log.Log.Error("webrtc.main.writeFinalSamples(): error writing final video sample: " + err.Error())
+		}
+	}
+
+	if state.lastAudioSample != nil && audioTrack != nil {
+		if err := audioTrack.WriteSample(*state.lastAudioSample); err != nil && err != io.ErrClosedPipe {
+			log.Log.Error("webrtc.main.writeFinalSamples(): error writing final audio sample: " + err.Error())
+		}
+	}
+}
+
+// processVideoPacket processes a video packet and writes samples to the track
+func processVideoPacket(pkt packets.Packet, state *streamState, videoTrack *pionWebRTC.TrackLocalStaticSample, config models.Config) {
+	if videoTrack == nil {
+		return
+	}
+
+	// Start at the first keyframe
+	if pkt.IsKeyFrame {
+		state.start = true
+	}
+
+	if !state.start {
+		return
+	}
+
+	sample := pionMedia.Sample{Data: pkt.Data, PacketTimestamp: uint32(pkt.Time)}
+
+	if config.Capture.ForwardWebRTC == "true" {
+		// Remote forwarding not yet implemented
+		log.Log.Debug("webrtc.main.processVideoPacket(): remote forwarding not implemented")
+		return
+	}
+
+	if state.lastVideoSample != nil {
+		duration := sample.PacketTimestamp - state.lastVideoSample.PacketTimestamp
+		state.lastVideoSample.Duration = time.Duration(duration) * time.Millisecond
+
+		if err := videoTrack.WriteSample(*state.lastVideoSample); err != nil && err != io.ErrClosedPipe {
+			log.Log.Error("webrtc.main.processVideoPacket(): error writing video sample: " + err.Error())
+		}
+	}
+
+	state.lastVideoSample = &sample
+}
+
+// processAudioPacket processes an audio packet and writes samples to the track
+func processAudioPacket(pkt packets.Packet, state *streamState, audioTrack *pionWebRTC.TrackLocalStaticSample, hasAAC bool) {
+	if audioTrack == nil {
+		return
+	}
+
+	if hasAAC {
+		// AAC transcoding not yet implemented
+		// TODO: Implement AAC to PCM_MULAW transcoding
+		return
+	}
+
+	sample := pionMedia.Sample{Data: pkt.Data, PacketTimestamp: uint32(pkt.Time)}
+
+	if state.lastAudioSample != nil {
+		duration := sample.PacketTimestamp - state.lastAudioSample.PacketTimestamp
+		state.lastAudioSample.Duration = time.Duration(duration) * time.Millisecond
+
+		if err := audioTrack.WriteSample(*state.lastAudioSample); err != nil && err != io.ErrClosedPipe {
+			log.Log.Error("webrtc.main.processAudioPacket(): error writing audio sample: " + err.Error())
+		}
+	}
+
+	state.lastAudioSample = &sample
+}
+
 func WriteToTrack(livestreamCursor *packets.QueueCursor, configuration *models.Configuration, communication *models.Communication, mqttClient mqtt.Client, videoTrack *pionWebRTC.TrackLocalStaticSample, audioTrack *pionWebRTC.TrackLocalStaticSample, rtspClient capture.RTSPClient) {
 
 	config := configuration.Config
 
-	// Make peerconnection map
-	peerConnections = make(map[string]*pionWebRTC.PeerConnection)
-
-	// Set the indexes for the video & audio streams
-	// Later when we read a packet we need to figure out which track to send it to.
-	hasH264 := false
-	hasPCM_MULAW := false
-	hasAAC := false
-	hasOpus := false
-	streams, _ := rtspClient.GetStreams()
-	for _, stream := range streams {
-		if stream.Name == "H264" {
-			hasH264 = true
-		} else if stream.Name == "PCM_MULAW" {
-			hasPCM_MULAW = true
-		} else if stream.Name == "AAC" {
-			hasAAC = true
-		} else if stream.Name == "OPUS" {
-			hasOpus = true
-		}
+	// Check if at least one track is available
+	if videoTrack == nil && audioTrack == nil {
+		log.Log.Error("webrtc.main.WriteToTrack(): both video and audio tracks are nil, cannot proceed")
+		return
 	}
 
-	if !hasH264 && !hasPCM_MULAW && !hasAAC && !hasOpus {
-		log.Log.Error("webrtc.main.WriteToTrack(): no valid video codec and audio codec found.")
-	} else {
-		if config.Capture.TranscodingWebRTC == "true" {
-			// Todo..
-		} else {
-			//log.Log.Info("webrtc.main.WriteToTrack(): not using a transcoder.")
-		}
+	// Detect available codecs
+	codecs := detectCodecs(rtspClient)
 
-		var cursorError error
-		var pkt packets.Packet
-		var lastAudioSample *pionMedia.Sample = nil
-		var lastVideoSample *pionMedia.Sample = nil
-
-		start := false
-		receivedKeyFrame := false
-		lastKeepAlive := "0"
-		peerCount := "0"
-
-		for cursorError == nil {
-
-			pkt, cursorError = livestreamCursor.ReadPacket()
-
-			//if config.Capture.ForwardWebRTC != "true" && peerConnectionCount == 0 {
-			//	start = false
-			//	receivedKeyFrame = false
-			//	continue
-			//}
-
-			select {
-			case lastKeepAlive = <-communication.HandleLiveHDKeepalive:
-			default:
-			}
-
-			select {
-			case peerCount = <-communication.HandleLiveHDPeers:
-			default:
-			}
-
-			now := time.Now().Unix()
-			lastKeepAliveN, _ := strconv.ParseInt(lastKeepAlive, 10, 64)
-			hasTimedOut := (now - lastKeepAliveN) > 15 // if longer then no response in 15 sec.
-			hasNoPeers := peerCount == "0"
-
-			if config.Capture.ForwardWebRTC == "true" && (hasTimedOut || hasNoPeers) {
-				start = false
-				receivedKeyFrame = false
-				continue
-			}
-
-			if len(pkt.Data) == 0 || pkt.Data == nil {
-				receivedKeyFrame = false
-				continue
-			}
-
-			if !receivedKeyFrame {
-				if pkt.IsKeyFrame {
-					receivedKeyFrame = true
-				} else {
-					continue
-				}
-			}
-
-			//if config.Capture.TranscodingWebRTC == "true" {
-			// We will transcode the video
-			// TODO..
-			//}
-
-			if pkt.IsVideo {
-
-				// Start at the first keyframe
-				if pkt.IsKeyFrame {
-					start = true
-				}
-				if start {
-					sample := pionMedia.Sample{Data: pkt.Data, PacketTimestamp: uint32(pkt.Time)}
-					//sample = pionMedia.Sample{Data: pkt.Data, Duration: time.Second}
-					if config.Capture.ForwardWebRTC == "true" {
-						// We will send the video to a remote peer
-						// TODO..
-					} else {
-						if lastVideoSample != nil {
-							duration := sample.PacketTimestamp - lastVideoSample.PacketTimestamp
-							bufferDurationCasted := time.Duration(duration) * time.Millisecond
-							lastVideoSample.Duration = bufferDurationCasted
-							if err := videoTrack.WriteSample(*lastVideoSample); err != nil && err != io.ErrClosedPipe {
-								log.Log.Error("webrtc.main.WriteToTrack(): something went wrong while writing sample: " + err.Error())
-							}
-						}
-
-						lastVideoSample = &sample
-					}
-				}
-			} else if pkt.IsAudio {
-
-				// @TODO: We need to check if the audio is PCM_MULAW or AAC
-				// If AAC we need to transcode it to PCM_MULAW
-				// If PCM_MULAW we can send it directly.
-
-				if hasAAC {
-					// We will transcode the audio from AAC to PCM_MULAW
-					// Not sure how to do this yet, but we need to use a decoder
-					// and then encode it to PCM_MULAW.
-					// TODO..
-					//d := fdkaac.NewAacDecoder()
-					continue
-				}
-
-				// We will send the audio
-				sample := pionMedia.Sample{Data: pkt.Data, PacketTimestamp: uint32(pkt.Time)}
-
-				if lastAudioSample != nil {
-					duration := sample.PacketTimestamp - lastAudioSample.PacketTimestamp
-					bufferDurationCasted := time.Duration(duration) * time.Millisecond
-					lastAudioSample.Duration = bufferDurationCasted
-					if err := audioTrack.WriteSample(*lastAudioSample); err != nil && err != io.ErrClosedPipe {
-						log.Log.Error("webrtc.main.WriteToTrack(): something went wrong while writing sample: " + err.Error())
-					}
-				}
-
-				lastAudioSample = &sample
-			}
-		}
+	if !codecs.hasValidCodecs() {
+		log.Log.Error("webrtc.main.WriteToTrack(): no valid video or audio codec found")
+		return
 	}
 
-	peerConnectionCount = 0
-	log.Log.Info("webrtc.main.WriteToTrack(): stop writing to track.")
+	if config.Capture.TranscodingWebRTC == "true" {
+		log.Log.Info("webrtc.main.WriteToTrack(): transcoding enabled but not yet implemented")
+	}
+
+	// Initialize streaming state
+	state := &streamState{
+		lastKeepAlive: time.Now().Unix(),
+		peerCount:     0,
+	}
+
+	defer func() {
+		writeFinalSamples(state, videoTrack, audioTrack)
+		log.Log.Info("webrtc.main.WriteToTrack(): stopped writing to track")
+	}()
+
+	var pkt packets.Packet
+	var cursorError error
+
+	for cursorError == nil {
+		pkt, cursorError = livestreamCursor.ReadPacket()
+
+		if cursorError != nil {
+			break
+		}
+
+		// Update state from communication channels
+		updateStreamState(communication, state)
+
+		// Check if we should continue streaming
+		if !shouldContinueStreaming(config, state) {
+			state.start = false
+			state.receivedKeyFrame = false
+			continue
+		}
+
+		// Skip empty packets
+		if len(pkt.Data) == 0 || pkt.Data == nil {
+			state.receivedKeyFrame = false
+			continue
+		}
+
+		// Wait for first keyframe before processing
+		if !state.receivedKeyFrame {
+			if pkt.IsKeyFrame {
+				state.receivedKeyFrame = true
+			} else {
+				continue
+			}
+		}
+
+		// Process video or audio packets
+		if pkt.IsVideo {
+			processVideoPacket(pkt, state, videoTrack, config)
+		} else if pkt.IsAudio {
+			processAudioPacket(pkt, state, audioTrack, codecs.hasAAC)
+		}
+	}
 }