Design

Architecture diagram

digraph shaka_packager { label=<<u>Shaka Packager Architecture</u>> labelloc=t subgraph cluster_media { label=<<u>Media Processing Pipeline</u>> Demuxer, ChunkingHandler, EncryptionHandler, Replicator, TrickplayHandler, Muxer [shape=rectangle] Demuxer -> ChunkingHandler [style=bold headlabel="many" taillabel="1"] ChunkingHandler -> EncryptionHandler -> Replicator -> TrickplayHandler -> Muxer [style=bold] ChunkingHandler -> Replicator -> Muxer [style=bold] } MuxerListener, MpdNotifyMuxerListener, HlsNotifyMuxerListener, MpdNotifier, HlsNotifier [shape=rectangle style=rounded] Muxer -> MuxerListener MuxerListener -> MpdNotifyMuxerListener, HlsNotifyMuxerListener [dir=back arrowtail=onormal] subgraph cluster_manifest { label=<<u>Manifest Generation</u>> HlsNotifyMuxerListener -> HlsNotifier [headlabel="1" taillabel="many"] MpdNotifyMuxerListener -> MpdNotifier [headlabel="1" taillabel="many"] MasterPlaylist, MediaPlaylist, MpdBuilder, AdaptationSet, Representation [shape=trapezium] HlsNotifier -> MasterPlaylist MasterPlaylist -> MediaPlaylist [dir=back arrowtail=diamond headlabel="many" taillabel="1"] MpdNotifier -> MpdBuilder MpdBuilder -> AdaptationSet -> Representation [dir=back arrowtail=diamond headlabel="many" taillabel="1"] {rank=same; MasterPlaylist, MpdBuilder} {rank=same; MediaPlaylist, Representation} } }
digraph shaka_packager { subgraph cluster_demuxer { style=rounded label=<<u> </u>> Demuxer2 [label="Demuxer" shape=rectangle] Demuxer2 -> MediaParser [dir=back arrowtail=diamond headlabel="1" taillabel="1"] MediaParser -> Mp4MediaParser, WebMMediaParser, Mp2tMediaParser, WvmMediaParser [dir=back arrowtail=onormal] } subgraph cluster_muxer { style=rounded label=<<u> </u>> Muxer2 [label="Muxer" shape=rectangle] Muxer2 -> Mp4Muxer, WebMMuxer, Mp2tMuxer [dir=back arrowtail=onormal] } }
digraph shaka_packager { subgraph cluster_legend { style=rounded label=<<u>Legend</u>> node [shape=plaintext] blank1 [label="" height=0] blank2 [label="" height=0] blank3 [label="" height=0] "Composition" -> blank1 [dir=back arrowtail=diamond] "Inheritance" -> blank2 [dir=back arrowtail=onormal] "MediaHandler data flow" -> blank3 [style=bold] "Bridge Class" [shape=rectangle style=rounded] "Manifest Class" [shape=trapezium] MediaHandler [shape=rectangle] } }

Media handler data flow

digraph g { rankdir=LR StreamData [ label="{... | SegmentInfo | MediaSample ... | SegmentInfo | MediaSample ... | StreamInfo}" shape=record style=rounded ]; MediaHandler [shape=rectangle] MediaHandler2 [shape=rectangle, label=MediaHandler] MediaHandler -> StreamData -> MediaHandler2 }

Teletext processing pipeline

When processing MPEG-TS input with DVB-Teletext subtitles, a specialized pipeline ensures text segments align with video/audio segment boundaries:

digraph teletext_pipeline { label=<<u>Teletext Processing Pipeline</u>> labelloc=t rankdir=TB compound=true subgraph cluster_demuxer { label=<<u>Demuxer</u>> style=rounded Mp2tMediaParser [shape=rectangle] EsParserTeletext [shape=rectangle] EsParserH264 [shape=rectangle label="EsParserH26x"] Mp2tMediaParser -> EsParserH264 [label="video PID"] Mp2tMediaParser -> EsParserTeletext [label="teletext PID"] Mp2tMediaParser -> EsParserTeletext [label="heartbeat\n(video PTS)" style=dashed] } subgraph cluster_coordinator { label=<<u>Segment Coordination</u>> style=rounded SegmentCoordinator [shape=rectangle] } subgraph cluster_chunking { label=<<u>Chunking</u>> style=rounded ChunkingHandler [shape=rectangle] TextChunker [shape=rectangle] } subgraph cluster_output { label=<<u>Output</u>> style=rounded VideoMuxer [shape=rectangle label="Muxer\n(video)"] TextMuxer [shape=rectangle label="Muxer\n(text)"] } EsParserH264 -> SegmentCoordinator [style=bold label="video stream"] EsParserTeletext -> SegmentCoordinator [style=bold label="text stream"] SegmentCoordinator -> ChunkingHandler [style=bold label="video/audio"] SegmentCoordinator -> TextChunker [style=bold label="teletext"] ChunkingHandler -> SegmentCoordinator [style=dashed label="SegmentInfo" constraint=false] SegmentCoordinator -> TextChunker [style=dashed label="replicate\nSegmentInfo" constraint=false] ChunkingHandler -> VideoMuxer [style=bold] TextChunker -> TextMuxer [style=bold] }

The key components are:

  • Mp2tMediaParser: Sends periodic “heartbeat” signals (video PTS timestamps) to the teletext parser to drive segment generation even when no subtitle data is present.

  • EsParserTeletext: Parses DVB-Teletext PES packets and emits text samples with special roles (CueStart, CueEnd, TextHeartBeat, MediaHeartBeat).

  • SegmentCoordinator: An N-to-N handler that passes all streams through unchanged, but also replicates SegmentInfo from video/audio streams to registered teletext streams. This ensures text segment boundaries match video segment boundaries exactly.

  • TextChunker: In “coordinator mode”, receives SegmentInfo events from the SegmentCoordinator and uses them to determine segment boundaries instead of calculating boundaries from text sample timestamps.