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
export ts: after #8449, a padded export under a sustained slow reader keeps about a quarter of the video an unpadded one keeps #8557
Measured on main at f190d90b34, which carries #8449; nothing under rs/moq-mux/src/container/ts has changed up to 75f00f92c8.
#8449 fixes the bursts from #8383, re-run with every setting as before. After stdout blocks for 60 s, the first second behind tsp -P regulate --pcr-synchronous carries the line rate, against 539 Mb/s before. After a 30 s freeze and a followed return on moq-lite-07, the peak second is 1.0× the line rate, against 30.9×. Under a sustained slow reader the export now sheds and stays bounded, but the reader keeps less of the video than before.
Repro:#8383's slow reader. A constant-rate 10 Mb/s source, --mux-rate from the catalog, moq … export ts --delay 2s | <reader that drains stdout at half the line rate for 60 s>.
Observed, two runs alike: the lag holds at 3.6–3.7 s and the export no longer exits. It sheds 29 times, every 2.0–2.2 s, each time 1.00–1.06 s behind (half its delay), and drops 3,911 frames. The reader gets 11–12 % of the video PES a full-rate subscriber gets. Before #8449 the padded export kept 20 % (and then exited). With --mux-rate 0 (at bad28cefd5) the same reader kept 47 % at a flat 4 s lag.
Why it matters: a padded export is the one a constant-rate receiver needs. Under a slow sink it now keeps about a quarter of the video an unpadded export keeps, at the same bound.
Not traced. One guess: the budget's slots laid on each resume take the reader's bandwidth ahead of media, so each 2 s cycle carries mostly padding. Is that the intended trade-off, or worth a follow-up?
#8449 - fix(moq-mux): a padded TS export sheds when its sink falls behind [closed] #7012 - fix(mux): bound jitter admission [merged]
📝 Issue Planner
Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.
Create Plan
🧪 Issue enrichment is currently in open beta.
You can configure auto-planning by selecting labels in the issue_enrichment configuration.
To disable automatic issue enrichment, add the following to your .coderabbit.yaml:
issue_enrichment:
auto_enrich:
enabled: false
💬 Have feedback or questions? Drop into our discord!
Thanks for the careful measurement. We're closing this as working as intended.
The cause you guessed is close. After each shed, the resumed grid opens up to the budget before the next keyframe is due. That stretch is mostly null packets, and a half-rate reader spends its next ~2 s writing it, falls ~1 s behind, and sheds again. It isn't back-fill for the skipped time; Schedule::clear resets the grid.
A padded export exists for a constant-rate receiver, and that receiver can't use a feed arriving at half its PCR rate, whatever the feed carries. Against a sink slower than the mux rate, the padded export can promise a bounded lag, a valid mux (flagged discontinuities, correct continuity counters), no bursts, and no false clock-drift error. #8449 delivers all four. Periodic jumps under a sustained slow sink were accepted when #8449 was planned. A sink that can't keep up with the mux rate should use --mux-rate 0, which sheds whole groups at the source without spending bytes on padding.
Activity
coderabbitai commentedon Nov 12, 2026
🔗 Related PRs
#8449 - fix(moq-mux): a padded TS export sheds when its sink falls behind [closed]
#7012 - fix(mux): bound jitter admission [merged]
📝 Issue Planner
Check the box below or use the
@coderabbitai plancommand to generate an implementation plan and prompts that you can use with your favorite coding assistant.🧪 Issue enrichment is currently in open beta.
You can configure auto-planning by selecting labels in the issue_enrichment configuration.
To disable automatic issue enrichment, add the following to your
.coderabbit.yaml:💬 Have feedback or questions? Drop into our discord!
kixelated commentedon Nov 12, 2026
Thanks for the careful measurement. We're closing this as working as intended.
The cause you guessed is close. After each shed, the resumed grid opens up to the budget before the next keyframe is due. That stretch is mostly null packets, and a half-rate reader spends its next ~2 s writing it, falls ~1 s behind, and sheds again. It isn't back-fill for the skipped time;
Schedule::clearresets the grid.A padded export exists for a constant-rate receiver, and that receiver can't use a feed arriving at half its PCR rate, whatever the feed carries. Against a sink slower than the mux rate, the padded export can promise a bounded lag, a valid mux (flagged discontinuities, correct continuity counters), no bursts, and no false clock-drift error. #8449 delivers all four. Periodic jumps under a sustained slow sink were accepted when #8449 was planned. A sink that can't keep up with the mux rate should use
--mux-rate 0, which sheds whole groups at the source without spending bytes on padding.Recorded in the import PR: #8582
(Written by Opus 5.5)