Shitty service to proxy data from OBS into an HTTP Live Streaming (HLS) feed VRChat understands.

## Cost

- One connection is about 7 Mb/s
- GCE charges $0.05 per GiB

($0.05 $/GiB) * (1 byte / 8 bits) = 0.00625 $/Gib
(0.00625 $/Gib) * 0.007 Gib/s = $4.375 * 10^-5 $/sec (per connection)
[$4.375 * 10^-5 / (sec * connection)] * 100 connections = $0.004375 / sec

7 Mb/s is 3.076 GiB/hr

## Streamer instructions

1. Configure OBS with a custom server pointing at `rtmps://<your-domain>:1935/live`
    in the stream settings and enter your stream key.
2. Press stream. Check status on bottom right of obs.

## Dev instructions

1. Get ssh perms to yummers.dev. `ssh yummers.dev` should log you in as yum if
   not, you'll have to update the ssh commands below.
2. Run ./push.sh
3. Add this to your .bashrc:
```bash
function check_live {
  ssh yummers.dev 'sudo journalctl -u obsproxy -f'
}
function start_live {
  ssh yummers.dev 'sudo systemctl start obsproxy'
}
function stop_live {
  ssh yummers.dev 'sudo systemctl stop obsproxy'
}
function reset_live {
  ssh yummers.dev 'sudo systemctl restart obsproxy'
  get_live_logs
}
function get_live {
  live_url=$(ssh yummers.dev "journalctl -u obsproxy -n 10000 | grep 'HLS:' | awk '{print \$NF}'" | tail -n 1)
  echo $live_url | clip.exe
  echo "Copied to clipboard: $live_url"
}
```
... and source it with `source ~/.bashrc`.
4. 

2. Install runtime dependencies: `pip install -r opt/obsproxy/requirements.txt`.
3. Start the Python service (see `etc/systemd/system/obsproxy.service` for a sample unit). The bundled entrypoint now runs under [Waitress](https://docs.pylonsproject.org/projects/waitress/en/stable/) so you get a production-grade WSGI server out of the box.
4. When the service starts it prints a session-specific playlist URL like `https://<your-domain>/hls/<session-hex>/stream.m3u8`; share that exact URL with your VRChat video player. The manifest now advertises AES-128 encryption with a companion key URL in the same session-scoped directory, so the player must support standard HLS key retrieval. Multiple viewers can consume the feed concurrently.

RTMPS termination happens at nginx (default port `1935`) via the `ngx_stream_module`, which proxies plain RTMP to the local nginx-rtmp listener on `127.0.0.1:1936`. Update the `INGEST_RTMP_*` settings if you run the backend elsewhere, and make sure the stream module is installed/enabled (on Debian/Ubuntu install `libnginx-mod-stream`).

Environmental knobs:

- `OBS_STREAM_KEY` / `STREAM_PSK`: required PSK for the single ingest client.
- `INGEST_THREAD_QUEUE_SIZE`: RTMP demux queue depth before frames are dropped (defaults to `4096`).
- `INGEST_RTMP_HOST`: host FFmpeg pulls RTMP from (defaults to `127.0.0.1`).
- `INGEST_RTMP_PORT`: port FFmpeg pulls RTMP from (defaults to `1936`).
- `HLS_SEGMENT_TIME`: length (in seconds) of the `.ts` segments emitted by FFmpeg (defaults to `2`).
- `HLS_PLAYLIST_SIZE`: number of segments retained in the rolling playlist (defaults to `6`).
- `HLS_DELETE_THRESHOLD`: additional historical HLS segments to keep on disk before pruning (defaults to `2`).
- `HOST`: interface Waitress binds to (defaults to `0.0.0.0`).
- `PORT`: TCP port Waitress listens on (defaults to `5000`).
- `FFMPEG_LOGLEVEL`: override the log verbosity passed to FFmpeg (defaults to `info`).

The server seeds a fresh 128-bit session ID on every restart and writes HLS artifacts under `<STREAM_DIR>/live/<session-hex>`. The playlist and segments are only exposed under `/hls/<session-hex>/`, making it infeasible to guess a live session path.