SolarOS 4.17.7 manual · job

rtspd

Publish explicitly selected sources as a standard single-client RTSP session. Video uses RTP/JPEG; audio uses RTP/L16 PCM. Each enabled track has its own RTP/RTCP UDP port pair, SSRC, and sender reports, with a shared RTCP CNAME and session start clock.

job start rtspd [video=<stream>|none] [audio=<stream>|none] [size=qvga|vga] [fps=0..30] [port=<port>]
job start rtspd video=camera0 audio=none size=qvga fps=10
job start rtspd video=none audio=audio0.capture
job start rtspd video=camera0 audio=audio0.capture
job status rtspd
job stop rtspd

Defaults are video=camera0 audio=none size=qvga fps=5 port=554. video=camera is a compatibility alias for camera0. Video selects a typed JPEG frame source from stream list; opening it holds the same exclusive hardware lease as the camera service. At least one source must be enabled. fps= is a maximum video publication rate, not a capture timer; fps=0 removes the cap. size= and fps= are invalid with video=none. JPEG quality is fixed at 12.

The publisher keeps Wi-Fi awake while the job runs. Stopping it releases that request; other active low-latency services can continue to keep Wi-Fi awake.

Use stream list to find available video and audio source IDs. audio0.capture is the board codec's PCM microphone endpoint when available; mic0 is a scalar level sensor, not a PCM source. Audio must be a source or duplex S16LE PCM endpoint with 16-bit samples, 1..8 channels, and a native rate of 8000..192000 Hz. The job advertises the source's native rate and channels; it does not resample. Only selected hardware is leased as job:rtspd. Tab completion after job start rtspd offers unused option keys. After audio= or video=, double Tab lists compatible registered source IDs and none; partial IDs complete normally. Selection does not open or lease the source. Audio-only publishing requires neither a camera nor the camera package. For playback on another SolarOS device, use JPEG video and L16 audio at 8000..48000 Hz with one or two channels. A 16000 Hz mono source is the recommended low-bandwidth audio format; native 16000 Hz stereo capture also works. The publisher's wider native-format support does not imply that every receiver supports those formats, and selecting a source never resamples it. External publishers should use baseline JPEG with both quantization tables; 160x120 at 10 fps is a useful initial ESP32 receiver workload. RTP packets must fit the path MTU; SolarOS publishes packets of at most 1200 bytes. Selecting an absent, busy, sink-only, or incompatible source fails startup; there is no silent fallback to another source.

Reader workers block on camera/audio availability and immediately publish or discard the completed frame/block. They continuously drain idle sources, so connecting does not expose a stale camera image or audio backlog. The video cap drops newly captured images rather than holding them until a timer expires. The publisher leases one PSRAM camera framebuffer at a time and keeps no additional JPEG copy or video queue. OV5640 capture uses two PSRAM framebuffers in latest-frame mode so capture can continue while a frame is being sent; OV2640 capture uses one. There is at most one MTU-sized PCM block. Disabled sources allocate no reader stack. Temporary UDP transmit pressure drops the current frame/block and newly captured data during a short backoff instead of disconnecting or building a queue. Audio sample timestamps advance across dropped blocks. Status reports congestion drops and the last transmit errno separately from fatal send errors. Each enabled source uses an additional 4096-byte internal worker stack, shown by job status rtspd, and is closed by its owning reader during cancellation. Source stack minimum-free values are reported in bytes, including microphone startup and publishing. Packet scratch and RTSP session storage are allocated only when the job starts, using the PSRAM-preferred memory policy; disabled sources allocate no scratch. Audio scratch is one 1188-byte PCM block plus one 1200-byte RTP packet, off the reader's stack. I2S DMA buffers and task stacks remain internal. Runtime buffers are freed after workers and leases have closed, including failed startup; a pending stop retains them safely. An unused/stopped job reserves no session or media buffers, only its small internal control state and lock. job status rtspd reports runtime buffer sizes.

Open the single RTSP client session:

vlc rtsp://device/media
rtsp rtsp://device/media
ffplay -rtsp_transport udp -fflags nobuffer -probesize 32 -analyzeduration 1 -max_delay 100000 rtsp://device/media

FFplay's default probing/playback buffers can add seconds of latency; the example reduces client-side probing and UDP reordering delay. It does not guarantee a particular glass-to-glass latency. Capture timestamps determine video RTP timing; audio timestamps advance by sample frames. Images and audio blocks predating each PLAY are discarded.

The camera lease excludes cam-webd, shell camera capture, and other camera owners only when a camera source is selected. Each packet is at most 1200 bytes. Unsupported JPEG modes are counted and dropped, not sent using private payloads. Transport is UDP; RTSP-over-TCP interleaving is unsupported. VLC requires a build with Live555: if its log reports satip or access_realrtsp failures, check for --disable-live555 and use FFplay or a compatible VLC build.

The stream is unauthenticated and unencrypted; use it only on a trusted LAN or protected network path.