You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
proposal: Preflight, a "try before you go live" test in the Auto-Configuration Wizard (early engine only, no UI yet)
#2413
Moved from #752 (opened by @chriscoveries). The maintainer's decision is on the issue.
Proposal. What we have so far is weak: a headless engine only, no UI. Our CI is in the repair shop (see our note on #699), so none of it has been through CI. We'd like your view before we build more.
The idea
A "try before you go live" window. It would replace the test page of today's Auto-Configuration Wizard. You pick where you stream and record. Preflight then really encodes, really records and, where a service allows it, really sends to the ingest server. It shows live charts and a 400% loupe on the encoded picture. You change any stream or record setting and watch the effect, then press Apply.
Why: what today's wizard can't tell you
Line numbers are against master at a48a9a13d.
Hardware encoders are chosen by formula and never actually encoded. See FindIdealHardwareResolution (frontend/wizards/AutoConfigTestPage.cpp:771-875).
Only x264 is really tested (AutoConfigTestPage.cpp:584 onward). The bandwidth test also always uses x264 (:121), not the encoder you'll stream with.
Some encoders are never probed. VAAPI is never probed (frontend/wizards/AutoConfig.cpp:211-234), and native Intel Macs never get VideoToolbox (:225). AMD and QSV users with more than four cores default to x264 (:193).
Apply always switches you to Simple output mode (AutoConfig.cpp:373).
There are no tests for any of it.
A "use a faster preset" warning is no use if you can't try the faster preset first. Preflight is meant to be that test.
Plan
Destinations. Add one or more stream targets, plus "Record locally". Each is labelled with the kind of test it gets:
an ingest test with nothing public: Twitch and Amazon IVS ?bandwidthtest, which the wizard already uses (AutoConfigTestPage.cpp:141);
a private preview: YouTube, using the wizard's existing throwaway-broadcast path (AutoConfigStreamPage.cpp:300);
an uplink speed test only: services with no test mode. These are never streamed to publicly.
Live bench. The source and encoded previews with a movable loupe, an editable settings panel, and charts. The charts cover CPU, render time against the frame interval, skipped frames, dropped frames, bitrate, congestion and RTT.
Bake-off. Every available encoder at each candidate resolution and frame rate, with a pass, marginal or fail verdict and a plain-words reason. Quality would be scored with SSIM on luma, not VMAF, to avoid the dependency.
Apply. Shows exactly which config keys change, and keeps your Simple or Advanced mode.
Architecture: the logic lives in a new headless Rust crate. It takes stat samples in and gives decisions out, so it can be tested with recorded traces and no GPU or network. Thin C glue in frontend/preflight/ reads the stats OBS already exposes. The Qt UI reuses the wizard shell.
rust/obs-preflight: the engine. It rolls OBS's cumulative counters up into 5-second windows and gives a verdict with a reason for each. It handles counter resets after a restart, and it fails closed when an encoder produces no frames. A CLI replays a recorded trace through it.
A C ABI and a collector (frontend/preflight/) that sample a live output and video mix.
37 Rust tests pass, and the code went through four rounds of our own review.
Weak spots, plainly:
No UI at all. None of the screens above exist yet. There's nothing to click.
The thresholds are our guesses. Only the skipped-frames warning comes from OBS: the status bar warns at 0.1% (frontend/widgets/OBSBasicStatusBar.cpp:492). Every other band is a provisional default, for example CPU fails at 98% and dropped frames at 1%. We'd want maintainers to pick these.
The collector's live Linux test has never run. It needs a display (Xvfb), and our CI is down.
It conflicts with current master in Cargo.lock and the three symbol-export lists. The fix is mechanical, but it's not done.
No performance or real-stream numbers yet.
We're not opening a PR. A PR makes sense once there's a usable first screen.
Questions for maintainers
Is a preflight like this something you'd want in obs-rust, replacing the wizard's test page?
Should the verdict thresholds be fixed defaults, configurable, or derived from OBS's existing warnings only?
For services with no test mode, is a neutral uplink speed test acceptable? Candidates are M-Lab ndt7 or Cloudflare's speed test, and each sees the user's IP.
Should Apply keep Advanced mode? Today it always forces Simple.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Moved from #752 (opened by @chriscoveries). The maintainer's decision is on the issue.
The idea
A "try before you go live" window. It would replace the test page of today's Auto-Configuration Wizard. You pick where you stream and record. Preflight then really encodes, really records and, where a service allows it, really sends to the ingest server. It shows live charts and a 400% loupe on the encoded picture. You change any stream or record setting and watch the effect, then press Apply.
Why: what today's wizard can't tell you
Line numbers are against
masterata48a9a13d.FindIdealHardwareResolution(frontend/wizards/AutoConfigTestPage.cpp:771-875).AutoConfigTestPage.cpp:584onward). The bandwidth test also always uses x264 (:121), not the encoder you'll stream with.frontend/wizards/AutoConfig.cpp:211-234), and native Intel Macs never get VideoToolbox (:225). AMD and QSV users with more than four cores default to x264 (:193).AutoConfig.cpp:373).A "use a faster preset" warning is no use if you can't try the faster preset first. Preflight is meant to be that test.
Plan
?bandwidthtest, which the wizard already uses (AutoConfigTestPage.cpp:141);AutoConfigStreamPage.cpp:300);Architecture: the logic lives in a new headless Rust crate. It takes stat samples in and gives decisions out, so it can be tested with recorded traces and no GPU or network. Thin C glue in
frontend/preflight/reads the stats OBS already exposes. The Qt UI reuses the wizard shell.What exists today, and what's weak about it
Branch:
feat/preflight-m0, 7 commits, about 3,200 lines.rust/obs-preflight: the engine. It rolls OBS's cumulative counters up into 5-second windows and gives a verdict with a reason for each. It handles counter resets after a restart, and it fails closed when an encoder produces no frames. A CLI replays a recorded trace through it.frontend/preflight/) that sample a live output and video mix.Weak spots, plainly:
frontend/widgets/OBSBasicStatusBar.cpp:492). Every other band is a provisional default, for example CPU fails at 98% and dropped frames at 1%. We'd want maintainers to pick these.masterinCargo.lockand the three symbol-export lists. The fix is mechanical, but it's not done.We're not opening a PR. A PR makes sense once there's a usable first screen.
Questions for maintainers
All reactions