Repository navigation
fix(moq-mux): the SI renewal bench times a renewal, not the first catalog advertise - #8970
Conversation
…alog advertise Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
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 1 minute. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI) · 60c264429c944267c646989dd91da382bab8f7ac
No blocker. The added PCR and SI replay are in iter_batched_ref setup, so the timed closure still measures one PCR interval plus the configured number of arrivals. The change is scoped to the renewal fixture; the pending and history fixtures are untouched. I'd land this and treat the first paired Nightly comparison against the old fixture as a baseline reset, not an importer speedup. The same-fixture figures are the useful comparison across #8829. Static review; I did not rerun the benchmark.
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes: the single bench change in rs/moq-mux/benches/containers.rs.
- Untimed catalog advertise:
SiImport::identitiesnow ticks a PCR and replays the same SI sections once in setup, so the one-time catalog publish happens before timing rather than in every sample.
claude-sonnet-5-5 | 𝕏

Problem
Nightly run 43909101837 confirmed
moq-mux/containers ts_si_renewal_pcr/*regressions of +24 to +341% between 505a86f and 13e593c. Bisecting the bench binaries points to moq-dev/moq#8829 (5b4c01b fast, 2c3eb75 slow).The renewal itself did not get slower. Each timed iteration starts from a fresh importer whose setup sees each SI section only once. The importer advertises the SI tracks in the catalog on the section's next arrival, which is the timed renewal, and that first publish allocates the catalog track's deflate encoder. Before #8829, the mux-rate hold kept the catalog unpublished for the whole bench, so the publish never ran. Now it adds a constant ~33 µs to every sample.
Approach
SiImport::identitiesrenews every section once in setup (a PCR, then the same sections again), so the advertise happens before timing. Counted with temporary prints: the timed tick and renewal now publish no catalog. The only remaining publish comes from the importer's drop, which is outside timing.Validation
Patched bench built at each revision, run on one host (criterion, 1 s warmup, 2 s measurement), median of
ts_si_renewal_pcr/<retained>/<touches>:What remains of #8829 is under 1.5 µs per PCR, and visible only in the 1-touch cases. I tried an extra PCR alone in setup first, and it changed nothing: the advertise waits for the section's next arrival, not the next PCR.
just checkpasses.Impact
Alternatives
Follow-ups
None.
Caused by
Models
Size: XS
Implementer: Claude, claude-opus-5-5, unknown effort
🤖 Generated with Claude Code
(Written by Claude Opus 5.5)
Review
Rewritten by
quest reviewafter each round.Head: 60c2644
Base: origin/main at 13e593c
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:
01a1e5b1-c1de-72e2-972b-049a51d60233, round 1Score: 10/10,
The additional untimed PCR interval and SI replay establish the advertised tracks before renewal is measured, matching the importer’s clock and catalog behavior with no identified defects.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none