From b079d519bd03d5750b767929c7a140d9eb200cc4 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Wed, 11 Nov 2026 00:46:19 -0800 Subject: [PATCH] quest: plan the gst sibling-reannounce flake and a WebTransport close reason that always arrives Co-Authored-By: Claude Opus 5.5 --- quest/m0/test-flakes-2/README.md | 2 +- .../test-flakes-2/gst-sibling-reannounce.md | 19 ++++++++++ quest/m1/webtransport-close-reason.md | 36 +++++++++++++++++++ 3 files changed, 56 insertions(+), 1 deletion(-) create mode 100644 quest/m0/test-flakes-2/gst-sibling-reannounce.md create mode 100644 quest/m1/webtransport-close-reason.md diff --git a/quest/m0/test-flakes-2/README.md b/quest/m0/test-flakes-2/README.md index 5b5bc2b25db..ba440e0fce6 100644 --- a/quest/m0/test-flakes-2/README.md +++ b/quest/m0/test-flakes-2/README.md @@ -1,4 +1,4 @@ -# [S] More tests hold up under load +# More tests hold up under load ## Goal diff --git a/quest/m0/test-flakes-2/gst-sibling-reannounce.md b/quest/m0/test-flakes-2/gst-sibling-reannounce.md new file mode 100644 index 00000000000..25d3f97b178 --- /dev/null +++ b/quest/m0/test-flakes-2/gst-sibling-reannounce.md @@ -0,0 +1,19 @@ +# [S] A reannounced gst sibling never replays its finish under load + +## Goal + +`moq-gst`'s `source::imp::session_tests::a_reannounced_sibling_instance_does_not_replay_its_finish` +passes reliably under a loaded `just check --all`, fixed at its cause, and is +re-enabled: its `#[ignore]` and its `sh/gh/test-inventory.allow` entry go. + +## Plan + +[moq-dev/moq#8438](https://github.com/moq-dev/moq/pull/8438) disabled it after +one failure in a loaded `just check`: the finished track replayed on its pad +(2 starts instead of 1). It passed 3/3 alone. No single PR is known to have +caused it; find out whether the test races real time or `moqsrc` really can +replay a finished track when the sibling's re-announce wins the race against +its end, and fix whichever it is. If it is a product bug, the PR is a `fix`. + +Decided 2026-11-11 (maintainer, follow-up planning in /quest-complete): ✅ +plan it as a child of this m0 epic (rejected: leave the test ignored). diff --git a/quest/m1/webtransport-close-reason.md b/quest/m1/webtransport-close-reason.md new file mode 100644 index 00000000000..ed980b71553 --- /dev/null +++ b/quest/m1/webtransport-close-reason.md @@ -0,0 +1,36 @@ +# [S] A WebTransport client always reads the peer's close reason + +## Goal + +When a peer closes a WebTransport session, the client's error carries the +peer's close code and reason, never `cancelled`, whichever arrives first: a +stream reset or the CLOSE_WEBTRANSPORT_SESSION capsule. That covers every +handshake refusal, such as a version mismatch or a refused token, in +`moq-tokio` and `moq-uring`. + +## Plan + +Follow-up of [moq-dev/moq#8478](https://github.com/moq-dev/moq/pull/8478): +the relay closes the session with the version code (0x15), but its SETUP +stream's reset (code 0) reaches the client before the capsule, so the client +reports `cancelled`. Closing the session before releasing its streams won the +race only sometimes, so ordering on the sender is not the fix. + +Recommended shape (draft-ietf-webtrans-http3 §4.4): a closing session resets +and stops its open streams with WT_SESSION_GONE, and a reader that sees +WT_SESSION_GONE before the capsule waits for it, ending on the connection's +close if none comes. Reading WT_SESSION_GONE already exists +(`rs/moq-tokio/src/quic/web/error.rs`, `session_gone`), but it reports +`Error::SessionGone` until the capsule lands, and nothing sends it. The same +code lives in `rs/moq-uring/src/quic/web/`; keep the two in step, and check +whether `js/net`'s native WebTransport path has the same race. + +Test the race both ways on the paused clock: the reset first and the capsule +first both yield the peer's code and reason. + +Decided 2026-11-11 (maintainer, follow-up planning in /quest-complete): ✅ m1 +(rejected: m3, drop): a non-breaking fix that every handshake refusal's +diagnosis needs. + +Public API: none. Wire: WT_SESSION_GONE on stream resets, as the WebTransport +draft specifies.