[kixelated]

Completes /quest/m1/test-flakes-2/wasm-ietf-read.md (reopened after #5650).

Cause

The fixture published a 128 KiB group every 50 ms (2.6 MB/s). The wasm subscription uses the default Latency::REAL_TIME budget (max_delay 0; the binding hard-codes subscribe(None)). Under that budget, a reader that parks for more bytes after a newer group has landed gets old (model/group.rs poll_expired_if_blocked, model/track.rs superseded). On a loaded runner the browser and relay deliver a group slower than 50 ms. With TRACE on, the relay took ~150 ms to forward one group, so every group is superseded mid-read and none arrives whole.

This isn't IETF-specific:

• CI history: setup (lite-02) failed the same way on 2026-09-29 (run 36504066809, read 1 whole groups in 8 attempts). Its log has the same shape: group 0, an 890 ms gap, then groups 1-7 within 6 ms.
• Local reproduction (suite pinned to 2 cores with busy loops competing): old skips on all three flavours.
• Why a burst kills groups: the stall-then-burst is delivery lag on the runner. When it releases, the newer groups' headers land while older groups are still arriving, so the older groups are retired. That is correct real-time behaviour: a group that had fully arrived is always drained, and only a group the reader is still waiting on is dropped.

The earlier unsupported signature (2026-10-09 to 10-12) was a relay REQUEST_ERROR on SUBSCRIBE from TRACK_STATUS. It last appeared before #5409 and #5468 merged and hasn't recurred.

Change

• The publisher appends one group once the track has demand, then one more each time the reader calls advance(). No group is ever superseded while it is being read.
• expectGroups reads exactly count groups and requires consecutive sequences. Any skip or short group fails. This restores the bound, stricter than count * 4: the cause leaves no legal skip.
• Removed isOld, GROUP_INTERVAL_MS, and sleep. The README case description now matches what the case does.

Public API: none. Wire: none.

Evidence

Interleaved A/B on a loaded shared host (load average ~36 on 32 cores), Nix Chromium 155, the suite pinned with taskset alongside competing busy loops. A is main with the old count * 4 bound restored; B is this PR.

setup	A fails	B fails
2 cores, 6 busy loops	1/6 (setup: read 0 whole groups in 8 attempts)	0/6
1 core, 3 busy loops	3/6 (lite, ietf, and setup each failed)	0/6

just check passes.

Decision

Settled by the maintainer (2026-10-17): land option 1, the reader-paced fixture. The alternatives were a latency-budget parameter on @moq/wasm subscribe (public API, and any finite budget can still be outrun) or asserting only what REAL_TIME promises (drops "whole groups").

Follow-ups

• The 500-900 ms delivery stalls themselves (loopback WebTransport on CI, where the relay warns its UDP send buffer was granted 208 KiB) are not explained here. Relay QUIC stats in the WASM job's failure artifact would show whether it's loss recovery.
• On a NixOS host, Playwright's bundled Chromium can't load libglib. An env override for the browser's executablePath in test/wasm/driver.ts would let local runs reproduce (test: let real-time wasm reads skip freely #5650 couldn't).

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[kixelated]

Outcome: cause found. The REAL_TIME (zero) budget drops a group superseded mid-read, and a 50 ms timer outruns a loaded runner, on lite-02 as well as IETF. Fixed by pacing the fixture's publisher by the reader. The bound is back and stricter: exactly two consecutive whole groups. Interleaved A/B on a loaded host: the old bound failed 4 of 12 runs, this PR 0 of 12. It's a draft for one maintainer decision (reader pacing vs. a wasm latency-budget API vs. a weaker assertion); see the description.

(Written by Claude Opus 5.5)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-17T14:45:43.876428Z	b1aa995	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Warning

Review limit reached

You'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 19 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 4 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: edc36617-6685-4444-8ac0-e1052c5df623

📥 Commits

Reviewing files that changed from the base of the PR and between e93391d and b1aa995.

📒 Files selected for processing: 4

• quest/m1/test-flakes-2/README.md
• quest/m1/test-flakes-2/wasm-ietf-read.md
• test/wasm/README.md
• test/wasm/src/main.ts

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[kixelated]

Merge summary: rebased on main (one conflict in the epic README's Required list; resolved by keeping main's entries and removing this quest's). The decision is settled: the maintainer chose the reader-paced fixture (option 1). CI is green, including the WASM job. Codex reviewed b1aa995 with no findings; CodeRabbit was rate-limited and didn't review. Two follow-ups stay open, both listed in the description: the unexplained 500-900 ms CI delivery stalls, and a Chromium executable override for NixOS hosts.

(Written by Claude Opus 5.5)