Repository navigation
fix(moq-congestion): BBR3's ACK aggregation epoch restarts from idle with its delivered bytes - #8806
Conversation
Quest-Claim: ed52c7ae67889353
…with its delivered bytes Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
Reviewed the full diff of both changed files.
- Epoch reset on idle restart:
handle_restart_from_idlenow zeroesextra_acked_deliveredtogether withextra_acked_interval_start. This matches Linux zeroingack_epoch_acked, and it stops earlier bursts' bytes from counting as aggregation and feeding back into cwnd throughset_cwnd. - Regression test:
idle_gaps_keep_extra_acked_boundedsends 28-packet frames every 33 ms over a 20 ms RTT, with ACKs spread per packet. It assertsextra_ackedstays at or under one burst and cwnd under four bursts. - Quest file removed:
idle-aggregation.mdis deleted now that the work is done.
A nit, not blocking: the test's doc comment says bursts are "acknowledged whole", but the body spreads the ACKs per packet to model the serializing bottleneck.
claude-sonnet-5-5 | 𝕏
|
No actionable comments were generated in the recent review. WalkthroughOn idle restart, Merge Risk:
|

Finishes quest/m1/bbr3-references/idle-aggregation.md (file deleted; no other quest links to it).
Summary
handle_restart_from_idleresetextra_acked_interval_startbut notextra_acked_delivered, so every resumed burst counted all earlier bursts' bytes as aggregation, andset_cwndfed the resultingextra_ackedback into cwnd. It now zeroesextra_acked_deliveredwith the interval start, as Linux zeroesack_epoch_ackedon restart. One test,idle_gaps_keep_extra_acked_bounded, sends 28-packet frames every 33 ms (about 8 Mbit/s at 30 fps) over a 20 ms RTT with a serializing bottleneck.Public API and wire impact
None. Behavior change in
moq-congestion's BBR3 only, so no!.Caused by
BBRHandleRestartFromIdle, which omits resetting the delivered bytes (found by person, the 2026-11-14 audit).Validation
extra_ackedgrew by one burst per frame and cwnd reached 20 MB in 600 frames (about 20 s). With the fix it stays at or under one burst (33.6 KB) on every frame and cwnd under 4 bursts.quest gate just checkpasses (8485 tests).rs/moq-quic/README.mdis left to the epic README's own work, which will list the divergence from draft-06.Choices a later review can undo
Simwith a per-packet ACK spread, because a single whole-burst ACK never reproduced the growth (the delay bound reset the epoch each frame). The bound asserted is one burst forextra_ackedand four for cwnd.idle_restartandextra_acked_filterare untouched.Models
Size: XS
Implementer: Claude, claude-sonnet-5-5, unknown effort (arm A)
Follow-ups
None.
Review
Rewritten by
quest reviewafter each round.Head: 3de8b92
Base: origin/main at 64b621a
Risk: medium,
the base has no RISK.md to name what is always high or customer-facing, so every change is mediumClassified: medium by default
Reviewer: Codex, gpt-6.1-sol, medium effort
Session:
01a1deee-c7ea-7d20-8bbf-72dfcac48d33, round 1Score: 9/10,
The focused reset prevents prior bursts from inflating the ACK-aggregation estimate and includes regression coverage, though tests were not run as instructed.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none