Hardware encoding
- Detect the best usable encoder at startup by encoding a test frame with each
candidate in turn (nvenc, qsv, amf, then libx264). Being compiled into ffmpeg
is not enough: h264_amf fails without amfrt64.dll and h264_qsv fails without an
Intel iGPU, so the probe is the only reliable answer.
- Translate the quality target into the chosen encoder's own vocabulary. libx264's
-crf and "ultrafast" mean nothing to h264_nvenc, which takes -cq and p1..p7, so
passing the configured values straight through would have been silently wrong.
- On the reference GTX 1060 a 1080p->720p rendition costs ~62% of a core on NVENC
versus ~109% on libx264 ultrafast.
Per-rendition supervision
- Track encoder processes per rendition, not per stream. is_running answered
"any alive", so when the source ("原画") encoder died while 1080p and 720p kept
running, the supervisor saw a healthy stream and never restarted it — the
source URL then 404'd permanently. It now requires every rendition to be alive
and reports which ones are down.
- Restart only the dead renditions, so a single failure no longer interrupts
viewers on the renditions that were fine.
Production serving
- Serve the built frontend from FastAPI with SPA fallback, so production is one
process and one port with no Node runtime. Node becomes a build-time dependency.
- start.ps1/start.sh build the frontend on first run instead of failing to a 503.
Cleanup
- Drop the duplicated top-left LIVE badge; the control strip already carries it.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Frontend
- Replace the Ant Design app with the shadcn/ui app as the single frontend,
keeping the portfolio side-by-side revertible in this commit.
- Drop the abandoned mdui experiment and the AntD tree so there is one app,
one router and one design system to maintain.
Live playback
- Stop racing the encoder. lowLatencyMode + liveSyncDurationCount:1 targeted a
single 4s segment, so playback ran at up to 1.10x and the forward buffer
bottomed out at 0.1s, re-buffering every segment. Measured over 40s:
buffer min 0.1s -> 8.9s, rate 1.00 constant, 0 stalls.
- Remove the 落后 labels and the page-level edge-snapping timer that fought
hls.js over two different definitions of "latency".
- Live can no longer be paused: the control is not rendered while it plays, so
the icon can never misreport state.
- Fix autoplay: muted autoplay started silently and never unmuted, and the
gesture that restored sound also toggled playback off. One owner now.
Source ("原画") rendition
- Publish the publisher's stream remuxed losslessly via -c:v copy alongside the
transcoded ones, so a source option exists even when every ladder entry
downsizes.
- Give renditions an ASCII path slug: the HLS path is part of the URL and
MediaMTX rejects non-alphanumeric paths, so 原画 could never be served.
Backend
- Supervise MediaMTX from the backend: start it at boot and restart it on exit,
never killing an origin it did not start.
- Clear stale HLS output before restarting an encoder; an ENDLIST-less playlist
left by a dead encoder hangs hls.js forever.
- Mark JSON timestamps as UTC. utcnow().isoformat() emitted a naive string that
JS read as local time, so every viewer heartbeat looked 8 hours old in UTC+8
and the live board's active-viewer count was permanently zero.
- Rebind a viewer session when it arrives on another stream instead of failing,
and survive the StrictMode double-heartbeat race on first insert.
- Replace the Linux-only /proc and os.getloadavg metrics with psutil.
- Complete MediaMTX's cookieCheck redirect inside /hls proxying, and hold the
httpx client open for the streamed body.
Housekeeping
- Ignore RainOSS (it stores accessKeySecret), runtime output, TLS keys and
.venv; remove generated screenshots, load test scripts and stray files.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
