Repository navigation
quest: splice-stall narrows to a reader stranded by an Unroutable its relay sends - #9209
Conversation
Quest-Claim: 7614581735c6adc5
… relay sends 19 of 20 reopen.toml runs on main pass; the failure starved 2 reader paths whose relay reset them Unroutable while still showing the broadcast. Reverting #9179 brings back reopen-starve (4 of 8 failed). Pauses match lite-06 per build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lice-stall # Conflicts: # quest/m2/multicast/README.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
Reviewed the full diff: a quest rewrite, its README link text, and a script.toml comment. No executable behavior changes.
splice-stall.mdretitles and narrows the quest to a reader stranded by an Unroutable its relay sends, with the run evidence, open questions, and next steps.README.mdupdates the quest's link text to match the new title.test/mesh-replay/script.tomlupdates the[[known]]comment to describe the Unroutable strand.
claude-sonnet-5-5 | 𝕏
|
No actionable comments were generated in the recent review. WalkthroughThe documentation and replay issue record now describe a reader whose live subscription receives Merge Risk:
|

Rewrites quest/m2/multicast/splice-stall.md to what remains, with the evidence from rerunning the mesh replay after moq-dev/moq#9174 landed. The quest is not finished: readers still starve, rarely, and the cause is narrowed but not fixed.
What the runs show
just test mesh-replay --mesh test/mesh-replay/ring4.toml --script test/mesh-replay/reopen.toml --broadcasts 32, each underquest gate, on nixos (32 cores).reopen.tomllists no[[known]]entries, so every run judges the starve check without editingscript.toml. Relay binaries were prebuilt and passed with--new, so the A/B arms ran interleaved under the same load. Longest pause is the median across runs, in seconds, for a link reopen, a relay reopen (every link ofa), and a kill.The harness's
--newcan't run lite-06 by itself: the 0.17.17 binary speaks an oldermoq-lite-07-wip, which the newmoq-benchrejects. The lite-06 rows instead runreopen.tomlwithstart = "old"and--old <relay> --old-versions moq-lite-06.unroutable. On main, 1 of 20 failed, with 2 paths. The 68-205 starved paths in fix(moq-net): a lite-07 request follows a claim only through a neighbour that holds it #9179's own failing runs predate A subscription from an old relay mid-roll is fed #9174.breset its reader's live subscriptions ford/0andd/3with Unroutable when its route throughaidle-timed out. It kept announcing both unchanged. The reader's front strands a track refused as unroutable until its table moves (model/front.rstrack_endedandrefuse), so it never asked again: starved for about 300s, with no error and the path online. The merged head's failing run shows the same strand:a's reader got Unroutable forc/5and read nothing for 73s. This is the strand reopen-starve's first plan named, one hop down from where fix(moq-net): a lite-07 request follows a claim only through a neighbour that holds it #9179 removed the refusal. The quest now targets it.Not found: why relay
bsent Unroutable for those 2 paths and not the other 30. That run had info logs only. Runs with debug logs onlite::publisher,lite::subscriber, andmodel, on debug and release relays, failed only on a saturated host (100% CPU, relays at 5-30% of a core). There, relayalogged serving 30-40 groups per 10s to its reader, but the reader recorded 0-12 over a loopback hop, so the measuring chain itself starved. Those runs are discarded as evidence. A mock-time ring4 integration test reproduced nothing in 176 cases: 8 to 32 broadcasts, readers on their own sessions, a link cut or paused, and the cut swept 0-1.5s after readers start. It is not committed, since it never failed.Decisions
[[known]]entries. The quest's acceptance is not met. Their comment now describes the Unroutable strand.Front::waiting), or it moves its announcement so the downstream re-asks. Rejected: having the client re-ask on a quiet table, which needs a timer and can loop on a standing refusal.Public API and wire impact
None. Quest and replay-script comment only.
Validation
The runs above, logs inspected in each failing run's kept directory.
quest check --base origin/mainpasses.Follow-ups
None. The remaining work stays in this quest.
Models
Size: M
Implementer: Claude Code, claude-opus-5-5, unknown effort
(Written by Claude Opus 5.5)
🤖 Generated with Claude Code
Review
Rewritten by
quest reviewafter each round.Head: 3235a22
Base: origin/main at b3de94d
Risk: low,
Only quest documentation and test comments change, so no consumer or peer behavior can break, and a revert fully undoes the change.Classified: low by Codex, gpt-6.1-sol, low effort, session
01a1f6ec-32bd-73a0-9cd8-910dedcd4e75Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1f6ec-796a-79c3-a0d3-3f0f0e40b7cd, round 1Score: 10/10,
The change updates quest documentation and test comments without changing executable behavior or test configuration, and no actionable defects were found.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none