Repository navigation
quest: close the cpp -> js interop budget flake, fixed by #6231 - #6442
Conversation
Quest-Claim: 1f719bcfbd274311
The 38 s cell in run 39734065658 was a stalled player, not a slow one: the browser decoded 3 video frames and no audio for 30 s. Under load the C++ publisher fell back to WebSocket, and crates.io qmux 0.6 never returned connection credit for bytes a stopped or dropped stream left unread. Every subscription the relay cancelled as the fast subscribers left leaked its window, until the publisher could not send. The in-tree qmux on the QUIC stream state machine (#6231) returns that credit, and `credit_returns_for_unread_and_stopped_streams` covers it. Reproduced with a WebSocket publisher, eight short-lived subscribers, and one browser on two CPUs: 3/12 browser cells stall at #6231's parent and at the failing commit, 0/20 at #6231, 0/28 on main. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EeyiEy5FWz3z44DYZVo7vU
9a37edb to
8b3211f
Compare
kixelated
changed the title
|
Outcome: the cause was a qmux 0.6 connection-credit leak behind a WebSocket publisher, already fixed by #6231. I reproduced it with a forced-WebSocket publisher on 2 CPUs: 3/12 cells stalled at #6231's parent, 0/20 at #6231, and 0/28 on main. This PR only deletes the quest. One follow-up is suggested in the body (the tone-grace check under extreme CPU starvation), not filed. (Written by Claude Opus 5.5) |
|
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 5 seconds. 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. Comment |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
@codex review (Written by Claude Opus 5.5) |
|
Codex Review: Didn't find any major issues. Delightful! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
# Conflicts: # quest/m1/test-flakes-2/README.md
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: eda6b04691
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Merging. Closes the cpp to js interop budget flake quest, fixed by #6231, and records the tone-check grace follow-up as a quest. CI is green, the current head has a non-Grok review, and (Written by Claude Opus 5.5) |
Closes quest/m1/test-flakes-2/interop-cpp-js.md with no code change: the cause was already fixed by moq-dev/moq#6231. This PR deletes the quest and its link.
What overran
None of the three phases the quest guessed. In run 39734065658 the browser started fast, painted 3 video frames, then got no more video and no audio for the whole 30 s audio wait (
videoFrames: 3,audioBytes: 0). That was a stalled player, not a slow one.Cause
The run's artifacts show the C++ publisher fell back to WebSocket under load (
using WebSocket fallback). Its log then shows "subscribed error ... internal error" each time the relay cancelled an upstream subscription as a fast subscriber left. The relay's WebSocket sessions ran on crates.io qmux 0.6, which returned connection credit (MAX_DATA) only for bytes the application read (RecvStream::report_consumed). A stream that was stopped or dropped with unread bytes never gave its credit back. So every cancelled group stream used up part of the publisher's connection window. By the time the browser subscribed, the window was empty: the catalog got through, plus a few frames, then nothing.#6231 replaced qmux 0.6 with the in-tree qmux on the QUIC stream state machine, which credits unread data on stop.
credit_returns_for_unread_and_stopped_streamsinrs/moq-quic/src/connection/mux/tests.rsalready covers this.Reproduction
The harness rebuilds every client on each run, so I used a scratch loop instead. It ran a relay, a C++ publisher forced onto
ws://, eight short-livedcpp-interopsubscribers, and one browserdriver.ts subscribe, all pinned to 2 CPUs:c59c3ad5172259ab3c~172259ab3cOn the two older builds the publisher logged about 7 "subscribed error" lines per run. On #6231 and main it logged none. Without the WebSocket publisher, 500+ browser cells on main at 2 to 4 CPUs showed no stall (7 fast tone-check failures, see follow-ups).
Public API / wire impact
None.
Checks
just check: passed.quest check --base origin/main: passed.just test interop --allwas not rerun, since this PR changes no code.Follow-ups (not filed)
TONE_GRACE_MSsees silence. In one case it saw stuttered audio and reported a false "wrong tone". This was never seen in CI. A fix would charge the grace only for audible playback.(Written by Claude Opus 5.5)
🤖 Generated with Claude Code
https://claude.ai/code/session_01EeyiEy5FWz3z44DYZVo7vU