From 3771d326f5c0109e0f1b221339bbec97b717e490 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 10:17:41 -0700 Subject: [PATCH 1/5] quest: plan the merge-queue Platform fixes and auth serve address log Two m0/ci quests fix the macOS and Windows failures that would block every merge once Platform is a required gate; platform-queue now requires them. An m2 XS quest logs the address `moq auth serve` actually bound. Co-Authored-By: Claude Opus 5.5 --- quest/m0/ci/README.md | 2 ++ quest/m0/ci/platform-macos-zero-rtt.md | 16 ++++++++++++++++ quest/m0/ci/platform-queue.md | 5 +++++ quest/m0/ci/platform-windows-drill.md | 19 +++++++++++++++++++ quest/m2/README.md | 1 + quest/m2/auth-serve-bound-addr.md | 13 +++++++++++++ 6 files changed, 56 insertions(+) create mode 100644 quest/m0/ci/platform-macos-zero-rtt.md create mode 100644 quest/m0/ci/platform-windows-drill.md create mode 100644 quest/m2/auth-serve-bound-addr.md diff --git a/quest/m0/ci/README.md b/quest/m0/ci/README.md index 1b311c5a471..fae0693d6ca 100644 --- a/quest/m0/ci/README.md +++ b/quest/m0/ci/README.md @@ -66,6 +66,8 @@ The historical samples locate candidates; they are not controlled benchmarks. - [Share local check capacity](/quest/m0/ci/local-gate.md) - gate expensive phases and bound internal fanout - [Reduce test cost](/quest/m0/ci/test-cost.md) - measure and retire low-value expensive work - [Reduce setup cost](/quest/m0/ci/setup-cost.md) - select cleanup and setup from actual workload needs +- [Windows restart drill](/quest/m0/ci/platform-windows-drill.md) - the relay restart drill passes on Windows in the merge queue +- [macOS zero-RTT test](/quest/m0/ci/platform-macos-zero-rtt.md) - the moq-tokio 0-RTT driver test passes on macOS in the merge queue - [Require platform gates](/quest/m0/ci/platform-queue.md) - activate the consolidated platform checks in the main and release rulesets after the workflow lands - [mbx on R2](/quest/m0/ci/mbx-r2.md) - Rust builds on every OS read an R2 cache that only `main` writes, escaping the 10 GB Actions cache - [Nix cache](/quest/m0/ci/nix-cache.md) - `nix.moq.dev` on R2 replaces Cachix for CI and releases diff --git a/quest/m0/ci/platform-macos-zero-rtt.md b/quest/m0/ci/platform-macos-zero-rtt.md new file mode 100644 index 00000000000..d5b09a5ec43 --- /dev/null +++ b/quest/m0/ci/platform-macos-zero-rtt.md @@ -0,0 +1,16 @@ +# [S] macOS zero-RTT driver test passes in the merge queue + +## Goal + +`moq-tokio quic::driver::tests::zero_rtt` passes on the macOS Platform job, so +Platform can become a required merge-queue gate. + +## Plan + +- It fails on macOS merge-queue runs since #5664 moved moq-tokio onto + `moq-quic` (runs 38924430657 and 38931667318). The connect future at + `rs/moq-tokio/src/quic/driver/tests.rs:733` fails (`expect("connect")`) on + the 0-RTT attempt. Linux passes. +- The test is a verbatim copy of quinn's; find what differs on macOS in the + adapted driver (socket options, GSO/ECN, ticket handling) and fix the cause. + Do not ignore the test on macOS. diff --git a/quest/m0/ci/platform-queue.md b/quest/m0/ci/platform-queue.md index 1a1444bda39..8e81d044d64 100644 --- a/quest/m0/ci/platform-queue.md +++ b/quest/m0/ci/platform-queue.md @@ -33,6 +33,11 @@ the implementation PR. - Nightly and releases share the five-job macOS cap. Measure overlap with queue runs before reserving capacity or reducing the queue build limit. +## Required + +- [Windows restart drill](/quest/m0/ci/platform-windows-drill.md) - a required Platform gate would block every merge while it fails +- [macOS zero-RTT test](/quest/m0/ci/platform-macos-zero-rtt.md) - a required Platform gate would block every merge while it fails + ## Related - [mbx on R2](/quest/m0/ci/mbx-r2.md) - gives platform jobs a Rust cache diff --git a/quest/m0/ci/platform-windows-drill.md b/quest/m0/ci/platform-windows-drill.md new file mode 100644 index 00000000000..3700662a2ac --- /dev/null +++ b/quest/m0/ci/platform-windows-drill.md @@ -0,0 +1,19 @@ +# [S] Windows relay restart drill passes in the merge queue + +## Goal + +`moq-relay::drills relay_killed_mid_group_aborts_then_resumes` (both the +`loopback` and `impaired` lanes) passes on the Windows Platform job, so +Platform can become a required merge-queue gate. + +## Plan + +- It fails on every merge-queue Platform run that executes Windows, back to + at least 2026-10-16 05:41 UTC (runs 38857400879 through 38931667318). Both + lanes panic in `expect_status` at `rs/moq-relay/tests/drills.rs:661`: + "no status change within 10s, wanted Connected". Linux and macOS pass. +- Find when it started (bisect the merge-queue history) and the root cause: + why the reconnect loop never reports `Connected` after the relay restarts on + Windows. Fix the cause; do not raise the timeout or skip the test on Windows. +- Land a regression check that fails without the fix on Windows. The drill + itself is that check if it runs in the queue. diff --git a/quest/m2/README.md b/quest/m2/README.md index 2db979d6b19..71eddadd1e3 100644 --- a/quest/m2/README.md +++ b/quest/m2/README.md @@ -166,3 +166,4 @@ ranked just below deterministic TS legs, which it requires. - [Release execution](/quest/m2/release/README.md) - publication, installed-artifact proof, cleanup, and downstream adoption after m1 - [Play resume](/quest/m2/play-resume.md) - `moq-cli play` plays out pictures buffered before a declared end when new media arrives quickly - [Task supervision](/quest/m2/task-supervision.md) - the auth decider loop, stats publisher, and moq-tokio's token and WebSocket internals are supervised like #5447's tasks +- [Auth serve bound address](/quest/m2/auth-serve-bound-addr.md) - `moq auth serve` logs the port it actually bound diff --git a/quest/m2/auth-serve-bound-addr.md b/quest/m2/auth-serve-bound-addr.md new file mode 100644 index 00000000000..ffa4fbdc2fd --- /dev/null +++ b/quest/m2/auth-serve-bound-addr.md @@ -0,0 +1,13 @@ +# [XS] Auth serve logs the address it bound + +## Goal + +`moq auth serve --listen 127.0.0.1:0` logs the port the OS actually assigned, +not port 0, so scripts and tests can find the server. + +## Plan + +- In `Serve::run` (`rs/moq-cli/src/auth.rs`), log `listener.local_addr()` + after binding instead of the configured `addr`. +- Found while abandoning `quest/m1/test-flakes-2/auth-serve-signals.md` + (#5707). From 249fdd3f2cf33e0004a95309c08b70eab408581c Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 10:20:35 -0700 Subject: [PATCH 2/5] quest: refuse compact nonzero subgroups and REQUEST_UPDATE filters Two XS fail-loud quests in the ietf-extension epic, found while abandoning subgroups in #5704. Co-Authored-By: Claude Opus 5.5 --- quest/m1/ietf-extension/README.md | 2 ++ quest/m1/ietf-extension/compact-subgroup.md | 23 +++++++++++++++++++++ quest/m1/ietf-extension/update-filters.md | 15 ++++++++++++++ 3 files changed, 40 insertions(+) create mode 100644 quest/m1/ietf-extension/compact-subgroup.md create mode 100644 quest/m1/ietf-extension/update-filters.md diff --git a/quest/m1/ietf-extension/README.md b/quest/m1/ietf-extension/README.md index b3082738d76..b32fbc277f3 100644 --- a/quest/m1/ietf-extension/README.md +++ b/quest/m1/ietf-extension/README.md @@ -83,6 +83,8 @@ Work owned by this README, after the children land: ## Required +- [Compact subgroup refusal](/quest/m1/ietf-extension/compact-subgroup.md) - a compact-form subgroup other than 0 is refused, not relabelled as subgroup 0 +- [Update filters](/quest/m1/ietf-extension/update-filters.md) - REQUEST_UPDATE with a Range Filter ends the subscription instead of being served unfiltered - [Dynamic groups](/quest/m1/ietf-extension/dynamic-groups.md) - NEW_GROUP_REQUEST reaches the original publisher, coalesced across downstream subscribers - [Track properties](/quest/m1/ietf-extension/track-properties.md) - unknown track properties are forwarded unchanged in SUBSCRIBE_OK, FETCH_OK, and TRACK_STATUS_OK - [Relay hops](/quest/m1/ietf-extension/lite.md) - the `ietf` SETUP extension carries every crate-private field across a lite-07 hop between relays diff --git a/quest/m1/ietf-extension/compact-subgroup.md b/quest/m1/ietf-extension/compact-subgroup.md new file mode 100644 index 00000000000..cdadd3ce91c --- /dev/null +++ b/quest/m1/ietf-extension/compact-subgroup.md @@ -0,0 +1,23 @@ +# [XS] Refuse a compact-form subgroup other than 0 + +## Goal + +A moq-transport subgroup stream whose header names its subgroup by its first +object ID (the compact form) is refused when that ID is nonzero, exactly as +an explicit nonzero Subgroup ID already is. Nothing is relabelled as +subgroup 0. + +## Plan + +- Today `recv_group` in `rs/moq-net/src/ietf/subscriber.rs` refuses an + explicit nonzero Subgroup ID (`Error::Unsupported`, stream stopped, session + kept; test `a_non_zero_subgroup_is_stopped_without_closing_the_session`), + but never reads the compact flag (`has_subgroup_object`). A compact stream + for subgroup N > 0 is taken as subgroup 0 starting at object N, and if it + opens the group it is forwarded downstream as subgroup 0. moq-test mode 1 + (one subgroup per object, compact form) hits this. +- Refuse it the same way, with a regression test that fails today. A compact + stream with no objects names no subgroup; treat it as subgroup 0. +- Check whether NOT_SUPPORTED is a better stop code than INTERNAL_ERROR in + every supported draft, and use it if so. +- State in `doc/concept/standard.md` that only subgroup 0 is accepted. diff --git a/quest/m1/ietf-extension/update-filters.md b/quest/m1/ietf-extension/update-filters.md new file mode 100644 index 00000000000..c441d1078f7 --- /dev/null +++ b/quest/m1/ietf-extension/update-filters.md @@ -0,0 +1,15 @@ +# [XS] Refuse filters on REQUEST_UPDATE + +## Goal + +A moq-transport REQUEST_UPDATE that carries a non-empty Range Filter (a +Subgroup Filter included) ends that subscription with an error instead of +being applied unfiltered, matching how SUBSCRIBE and FETCH already refuse +filters with INVALID_FILTER. + +## Plan + +- `rs/moq-net/src/ietf/subscribe.rs` decodes the filters on REQUEST_UPDATE + (`_subgroup_filter` and friends, around line 516) and drops them. +- End the subscription the way other supported-but-unimplemented update + preferences end it, with a regression test that fails today. From 1e008af3a442e67416363f554424888065426071 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 10:21:25 -0700 Subject: [PATCH 3/5] quest: place the macOS zero_rtt failure in the initial handshake Co-Authored-By: Claude Opus 5.5 --- quest/m0/ci/platform-macos-zero-rtt.md | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/quest/m0/ci/platform-macos-zero-rtt.md b/quest/m0/ci/platform-macos-zero-rtt.md index d5b09a5ec43..d6aa2458228 100644 --- a/quest/m0/ci/platform-macos-zero-rtt.md +++ b/quest/m0/ci/platform-macos-zero-rtt.md @@ -8,9 +8,11 @@ Platform can become a required merge-queue gate. ## Plan - It fails on macOS merge-queue runs since #5664 moved moq-tokio onto - `moq-quic` (runs 38924430657 and 38931667318). The connect future at - `rs/moq-tokio/src/quic/driver/tests.rs:733` fails (`expect("connect")`) on - the 0-RTT attempt. Linux passes. + `moq-quic` (runs 38924430657 and 38931667318). The failure is in the + initial full-handshake connection, before any 0-RTT attempt: the server's + 0.5-RTT `open_uni` (`rs/moq-tokio/src/quic/driver/tests.rs:710`) and the + client's first `accept_uni` (`:733`) both end with `TimedOut`. Linux + passes. - The test is a verbatim copy of quinn's; find what differs on macOS in the - adapted driver (socket options, GSO/ECN, ticket handling) and fix the cause. + adapted driver (socket options, GSO/ECN, stream credit during the handshake) and fix the cause. Do not ignore the test on macOS. From e57430b192c13f33d9540f4a303eeee453ab2bf1 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 10:32:01 -0700 Subject: [PATCH 4/5] quest: drop the update-filters quest; REQUEST_UPDATE already refuses filters request_stream::Update::decode_body marks any nonempty Range Filter unsupported and apply_subscription_update ends the subscription. Co-Authored-By: Claude Opus 5.5 --- quest/m1/ietf-extension/README.md | 1 - quest/m1/ietf-extension/update-filters.md | 15 --------------- 2 files changed, 16 deletions(-) delete mode 100644 quest/m1/ietf-extension/update-filters.md diff --git a/quest/m1/ietf-extension/README.md b/quest/m1/ietf-extension/README.md index b32fbc277f3..c4521e3ed9a 100644 --- a/quest/m1/ietf-extension/README.md +++ b/quest/m1/ietf-extension/README.md @@ -84,7 +84,6 @@ Work owned by this README, after the children land: ## Required - [Compact subgroup refusal](/quest/m1/ietf-extension/compact-subgroup.md) - a compact-form subgroup other than 0 is refused, not relabelled as subgroup 0 -- [Update filters](/quest/m1/ietf-extension/update-filters.md) - REQUEST_UPDATE with a Range Filter ends the subscription instead of being served unfiltered - [Dynamic groups](/quest/m1/ietf-extension/dynamic-groups.md) - NEW_GROUP_REQUEST reaches the original publisher, coalesced across downstream subscribers - [Track properties](/quest/m1/ietf-extension/track-properties.md) - unknown track properties are forwarded unchanged in SUBSCRIBE_OK, FETCH_OK, and TRACK_STATUS_OK - [Relay hops](/quest/m1/ietf-extension/lite.md) - the `ietf` SETUP extension carries every crate-private field across a lite-07 hop between relays diff --git a/quest/m1/ietf-extension/update-filters.md b/quest/m1/ietf-extension/update-filters.md deleted file mode 100644 index c441d1078f7..00000000000 --- a/quest/m1/ietf-extension/update-filters.md +++ /dev/null @@ -1,15 +0,0 @@ -# [XS] Refuse filters on REQUEST_UPDATE - -## Goal - -A moq-transport REQUEST_UPDATE that carries a non-empty Range Filter (a -Subgroup Filter included) ends that subscription with an error instead of -being applied unfiltered, matching how SUBSCRIBE and FETCH already refuse -filters with INVALID_FILTER. - -## Plan - -- `rs/moq-net/src/ietf/subscribe.rs` decodes the filters on REQUEST_UPDATE - (`_subgroup_filter` and friends, around line 516) and drops them. -- End the subscription the way other supported-but-unimplemented update - preferences end it, with a regression test that fails today. From bc6276ed79d1a30e8572bdf856c2f9bd72930359 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 10:36:15 -0700 Subject: [PATCH 5/5] quest: refuse an empty compact subgroup stream as malformed Co-Authored-By: Claude Opus 5.5 --- quest/m1/ietf-extension/compact-subgroup.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/quest/m1/ietf-extension/compact-subgroup.md b/quest/m1/ietf-extension/compact-subgroup.md index cdadd3ce91c..89c11742da8 100644 --- a/quest/m1/ietf-extension/compact-subgroup.md +++ b/quest/m1/ietf-extension/compact-subgroup.md @@ -16,8 +16,10 @@ subgroup 0. for subgroup N > 0 is taken as subgroup 0 starting at object N, and if it opens the group it is forwarded downstream as subgroup 0. moq-test mode 1 (one subgroup per object, compact form) hits this. -- Refuse it the same way, with a regression test that fails today. A compact - stream with no objects names no subgroup; treat it as subgroup 0. +- Refuse it the same way, with a regression test that fails today. +- A compact stream that ends after its header names no subgroup, and the + stream grammar requires at least one object (draft-14 10.4, draft-19 + 11.4). Refuse it as malformed rather than reading it as subgroup 0. - Check whether NOT_SUPPORTED is a better stop code than INTERNAL_ERROR in every supported draft, and use it if so. - State in `doc/concept/standard.md` that only subgroup 0 is accepted.