sb-im / sphinx Goto Github PK

There are chances when a signal publisher's WebRTC signaling and connecting fails, especialy for ICE candidate exchanging. A publlisher should retry these steps at certain times to ensure success.

The currrent yaml config file contains only plain key value maps, and lacks structure.

A Toml config file not only keeps key value pairs simplicity but also adds structure with section name. And thus a better choice.

Publisher misses ICE candidates MQTT message before subscribing to receiving topic. Because skywalker is too fast to publish those candidates before livestream publisher could make subscription done. It causes the ICE connection failure.

Therefore, all MQTT subscriptions must be done upon service starting up to catch all messages.

When sphinx livestream streams from a proxied usb-webcamera, the video losses quality greatly.

Per 4ed1074
Context is cancel instantly due to its surrounding function returns immediately as a fact, current code is not affected mainly becasuse of all sub-commands code ignores context.

Solution, use context.Backgourd only.

The increase of unhandled disconnected orphan TrackLocal will increase memory usage and incur memory leak. Must tidy these resource after WebRTC disconnected.

Nano use this script to do video stream multicast as udp client:
/usr/bin/gst-launch-1.0 udpsrc port=5500 ! multiudpsink clients=127.0.0.1:5600,127.0.0.1:5601,127.0.0.1:5602,192.168.100.52:5004,192.168.100.25:5004,127.0.0.1:5004

if sphinx livestream is ran after this script as udp server, it can't play video.

Still needs to find whether if it can receive any udp data or just can't play video or both.

See: https://github.com/SB-IM/sphinx/blob/63fd93422c3fefc241b7f75bc70794d0940d8317/internal/livestream/signal.go#L23-L23

It blocks after reconnecting to mqtt broker.