Repository navigation
fix(moq-congestion)!: BBR3 counts Startup loss events per lost packet - #8848
Conversation
Quest-Claim: 4d77964dd95ecea3
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
ℹ️ No critical issues — one test comment nit inline.
Reviewed changes
I reviewed the switch from draft-06's per-run loss counting to draft-07's per-packet counting in BBR3, along with the removal of Lost::new_burst from the public API.
- Per-packet loss events:
note_lossnow adds one for every lost packet. The count resets inadvance_latest_delivery_signals(aftercheck_startup_done), inreset_congestion_signals, and inon_spurious_loss. I checked these four sites against the draft-07 pseudocode forNoteLoss,AdvanceLatestDeliverySignals,ResetCongestionSignals, andHandleSpuriousLossDetection, and they match. Lost::new_burstremoved: the field and its computation in moq-quic'sdetect_lost_packetsare gone. moq-quic keeps thelost_packetsmap because spurious-loss detection uses it. Nonew_burstreferences remain in the repo, and the migration note is in the PR's impact section.- Tail-drop Startup test:
startup_exits_to_drain_on_loss_when_app_limitednow drives an application-limited frame sender into a 6-packet tail-drop queue. It asserts that loss arrives in runs, so the old run-based count could not have ended Startup. - Reset coverage test:
loss_events_count_packets_and_reset_per_round_and_undocovers the per-packet count, the reset at a loss-round boundary, and the reset on spurious-loss undo. - Quest cleanup: deletes
quest/m2/bbr3-startup-loss.mdand its link in thebbr3-referencesREADME.
claude-opus-5-5 | 𝕏
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 16 minutes. View limit details
No actionable comments were generated in the recent review. WalkthroughThe Merge Risk:
|
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
I reviewed the one commit added since the prior Pullfrog review.
- Test comment fix: reworded the loss-round boundary comment in
loss_events_count_packets_and_reset_per_round_and_undoas suggested. No code changed.
claude-sonnet-5-5 | 𝕏
|
quest land watch: the merge queue removed this pull request at 33245c4 because it conflicts with The land fixer works on it in the background, and the watcher re-enqueues it at the back of the queue once it is fixed or judged flaky, or hands it back here with what it needs. (written by quest land watch) |
|
quest land watch: armed auto-merge on this pull request at 41f9c56, so it joins the back of the merge queue once its required checks pass, with the land fixer's fix ( (written by quest land watch) |
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
I reviewed the merge commit added since the prior Pullfrog review. It brings in main and resolves the conflicts in rs/moq-congestion/src/bbr3/mod.rs.
- Conflict resolution: kept the PR's per-packet
note_loss(), under base'scheck_probe_lossguard inprocess_lost_packet. - Resets preserved:
loss_events_in_roundstill resets inadvance_latest_delivery_signals,reset_congestion_signals, andon_spurious_loss. Base'sprobe_rtt_skip_congestedundo line now sits beside theon_spurious_lossreset. - Doc links: took base's draft-07 links. No
new_burstreferences remain inrs/.
claude-opus-5-5 | 𝕏

Finishes quest/m2/bbr3-startup-loss.md and deletes it.
Summary
BBR3's Startup high-loss exit now counts every lost packet toward
STARTUP_FULL_LOSS_CNT, as draft-07 section 5.3.1.3 specifies, instead of counting discontiguous lost ranges as draft-06 read. A media sender bursting a frame into a tail-drop link loses one run per frame, so a round saw one to three events and the exit never fired; its application-limited rounds also block the plateau exit, so Startup never ended.rs/moq-congestion/src/bbr3/mod.rs:note_lossadds one per lost packet.loss_events_in_roundresets where draft-07 does: inadvance_latest_delivery_signals(the loss-round boundary, aftercheck_startup_high_lossran, asAdvanceLatestDeliverySignalsdoes), inreset_congestion_signals, and inon_spurious_loss. Before, it was reset insidecheck_startup_high_lossand nowhere else.Lost::new_burstis removed (rs/moq-congestion/src/lib.rs), with its computation inrs/moq-quic/src/connection/mod.rsand its fixtures in SCReAM, GoogCC, and BBR3's test driver.lost_packetsin moq-quic stays: spurious-loss detection uses it.startup_exits_to_drain_on_loss_when_app_limitednow drives a tail-drop bottleneck instead of an even 1-in-25 drop. A new test,loss_events_count_packets_and_reset_per_round_and_undo, covers the count and its resets.bbr3-referencesREADME.Public API and wire impact
Public API: removes
moq_congestion::Lost::new_burst; aControllerimplementation that buildsLostdrops the field. Wire: none.Migration: delete the
new_burstfield wherever aLostis built.Decisions
Mine, made while the maintainer was away; a review can undo any of them.
advance_latest_delivery_signals, as draft-07 puts it, rather than keeping the reset insidecheck_startup_high_loss. The old placement skipped the reset once Startup had ended, so the count could carry stale losses; the new one resets on every loss round.recovery_start_round). This change does not need it: the new test passes onmainas is. I trial-merged fix(quic): controllers hear an ACK's losses before the ACK #8807 into this branch: no conflicts, and moq-congestion tests, thetests::sim,tests::probe, andtests::screamsuites pass.Validation
startup_exits_to_drain_on_loss_when_app_limitedfails on the unchanged code ("BBR never left STARTUP on loss") and passes now. It asserts the samples stay application-limited, so only loss ends Startup.on_spurious_loss, or the one inadvance_latest_delivery_signals, failsloss_events_count_packets_and_reset_per_round_and_undo. Thereset_congestion_signalsreset has no direct test.just checkpasses: 8482 tests.tests::sim::(seed 12345, 1024 cases) passes; the ignoredtests::probe::,tests::scream::, andtests::realtime::suites pass (17 tests).tests::realtime::compare(cellular presets) prints an identical table before and after, since those runs drop no packets in Startup; the change only matters under tail-drop loss, which the unit test exercises.git merge-tree: no conflicts with #8805, #8806, #8807, #8810. #8809 conflicts in five adjacent hunks ofrs/moq-congestion/src/bbr3/mod.rs(itscheck_probe_lossrename,note_loss(new_burst)and the comments that describe range counting, an undo line, a doc link); whichever lands second resolves them by taking this PR's counting and fix(moq-congestion): BBR3 schedules ProbeRTT as draft-07 does #8809's naming.Caused by
Follow-ups
None.
Models
Size: S
Implementer: Claude Code, claude-sonnet-5-5, high effort (iterate, S, arm B)
🤖 Generated with Claude Code
(Written by Claude Sonnet 5.5)
Review
mechanical fix by codex, gpt-6.1-sol, medium effort:
Merged base and mechanically re·solved BBR3 conflicts preserving packet loss counting, the check_probe_loss guard, and ProbeRTT undo; passed git diff --check and nix develop --no-update-lock-file --command just ci base (8488 tests passed).(head 41f9c56)Rewritten by
quest reviewafter each round.Head: 863df06
Base: origin/main at ca88fc1
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:
01a1dfba-f789-7ef2-a196-b1c604baac9d, round 1Score: 9/10,
No actionable regressions were found; packet-based loss counting and reset placement match the intended behavior, though builds and tests were not run as instructed.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none
Re-review skipped for 33245c4: it only rewords one test comment (Pullfrog finding), no code changed; the round-1 review of 863df06 stands.