From 45c5314efb92600c01642c74a6d3505575d2dc82 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sat, 24 Oct 2026 22:44:57 -0700 Subject: [PATCH 1/4] chore(quest): claim quest/m1/ietf-extension/README.md Quest-Claim: 4e5c0817378dc132 From 1cf9096b39ebe238e7e849535a531904c002f29d Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 25 Oct 2026 09:06:01 -0700 Subject: [PATCH 2/4] test(ietf): answer TRACK_STATUS in the pinned test origin and re-add the two-relay cells A local Moxygen patch makes MoQTestPublisher answer TRACK_STATUS with what its SUBSCRIBE_OK carries, mirroring facebookexperimental/moxygen#264. The harness runs the cluster topology again on drafts 16 and 18; both cells still fail at the relay, recorded in the passthrough quest. Co-Authored-By: Claude Opus 5.5 --- nix/moxygen-track-status.patch | 85 +++++++++++++++++++++++++++++++ nix/moxygen.nix | 12 +++-- quest/m1/ietf-extension/README.md | 35 +++++++------ test/ietf-conformance/README.md | 16 +++--- test/ietf-conformance/run.sh | 69 ++++++++++++++++--------- 5 files changed, 167 insertions(+), 50 deletions(-) create mode 100644 nix/moxygen-track-status.patch diff --git a/nix/moxygen-track-status.patch b/nix/moxygen-track-status.patch new file mode 100644 index 00000000000..0e7963b506c --- /dev/null +++ b/nix/moxygen-track-status.patch @@ -0,0 +1,85 @@ +--- a/moxygen/moqtest/MoQTestPublisher.cpp ++++ b/moxygen/moqtest/MoQTestPublisher.cpp +@@ -88,18 +88,25 @@ void MoQTestPublisher::cancelAll() { + } + } + +-std::shared_ptr MoQTestPublisher::makeForwarder( +- const FullTrackName& ftn, +- uint64_t /*version*/) { +- auto forwarder = std::make_shared(ftn); ++namespace { + +- // Advertise the priority even-numbered groups are published at, so the +- // subgroup and datagram encodings elide it for those and write it +- // explicitly for the odd ones. ++// Advertise the priority even-numbered groups are published at, so the ++// subgroup and datagram encodings elide it for those and write it explicitly ++// for the odd ones. ++Extensions makeTrackProperties() { + Extensions trackProperties; + trackProperties.insertMutableExtension( + Extension{kPublisherPriorityExtensionType, kMoQTestPublisherPriority}); +- forwarder->setExtensions(std::move(trackProperties)); ++ return trackProperties; ++} ++ ++} // namespace ++ ++std::shared_ptr MoQTestPublisher::makeForwarder( ++ const FullTrackName& ftn, ++ uint64_t /*version*/) { ++ auto forwarder = std::make_shared(ftn); ++ forwarder->setExtensions(makeTrackProperties()); + + auto cb = std::make_shared(); + cb->publisher = shared_from_this(); +@@ -121,6 +128,34 @@ void MoQTestPublisher::retireTrack( + tracks_.erase(it); + } + ++folly::coro::Task MoQTestPublisher::trackStatus( ++ TrackStatus trackStatus) { ++ XLOG(INFO) << "Received Track Status"; ++ ++ auto res = moxygen::convertTrackNamespaceToMoqTestParam( ++ &trackStatus.fullTrackName.trackNamespace); ++ if (res.hasError()) { ++ co_return folly::makeUnexpected( ++ TrackStatusError{ ++ trackStatus.requestID, ++ TrackStatusErrorCode::NOT_SUPPORTED, ++ "Invalid Parameters"}); ++ } ++ ++ TrackStatusOk ok; ++ ok.requestID = trackStatus.requestID; ++ auto trackIt = tracks_.find(trackStatus.fullTrackName); ++ if (trackIt == tracks_.end()) { ++ ok.trackProperties = makeTrackProperties(); ++ co_return ok; ++ } ++ const auto& forwarder = trackIt->second.forwarder; ++ ok.groupOrder = forwarder->groupOrder(); ++ ok.largest = forwarder->largest(); ++ ok.trackProperties = forwarder->extensions(); ++ co_return ok; ++} ++ + folly::coro::Task MoQTestPublisher::subscribe( + SubscribeRequest sub, + std::shared_ptr callback) { +--- a/moxygen/moqtest/MoQTestPublisher.h ++++ b/moxygen/moqtest/MoQTestPublisher.h +@@ -71,6 +71,11 @@ class MoQTestPublisher : public Publisher, + // ask for data -- otherwise it would hold the session open forever. + void cancelAll(); + ++ // Answers with what a SUBSCRIBE_OK for the track would carry, without ++ // starting it. A track that is not being generated has no largest location. ++ virtual folly::coro::Task trackStatus( ++ TrackStatus trackStatus) override; ++ + // Subscribing Methods. Subscribers to the same track share one forwarder + // and one generator, so a late subscriber joins the track in progress. + virtual folly::coro::Task subscribe( diff --git a/nix/moxygen.nix b/nix/moxygen.nix index ef8879bc552..02d157c8ba7 100644 --- a/nix/moxygen.nix +++ b/nix/moxygen.nix @@ -30,9 +30,15 @@ stdenv.mkDerivation { version = "unstable-2026-10-23"; src = sources.moxygen; cmakeDir = "../standalone"; - # The relay conformance profile: the client accepts track-level priority and - # an END_OF_TRACK directly after the last object. Every other check stays. - patches = [ ./moxygen-profile.patch ]; + patches = [ + # The relay conformance profile: the client accepts track-level priority and + # an END_OF_TRACK directly after the last object. Every other check stays. + ./moxygen-profile.patch + # The test origin answers TRACK_STATUS, which a relay cluster's edge sends + # before subscribing. Drop at the next pin bump once + # https://github.com/facebookexperimental/moxygen/pull/264 lands. + ./moxygen-track-status.patch + ]; nativeBuildInputs = [ cmake ninja diff --git a/quest/m1/ietf-extension/README.md b/quest/m1/ietf-extension/README.md index 251d8fc8ff2..87cda05e25c 100644 --- a/quest/m1/ietf-extension/README.md +++ b/quest/m1/ietf-extension/README.md @@ -103,27 +103,30 @@ directly after the last object, whose stream ends no payload group. All other upstream checks stay, including object IDs, properties, END_OF_GROUP and FIRST_OBJECT. +`nix/moxygen-track-status.patch` makes the test origin answer TRACK_STATUS +with what its SUBSCRIBE_OK carries, until +[facebookexperimental/moxygen#264](https://github.com/facebookexperimental/moxygen/pull/264) +lands and the pins move past it. + Measured on the profile over six runs: the direct-origin controls and the -single-relay cells pass 29/29 on drafts 16 and 18, and the nightly job gates on -them. Two-relay cells failed 29/29 before media, so the harness does not run -them yet. Remaining, in order: - -- **A publisher that answers metadata queries.** Cluster cells fail because the - edge's TRACK_INFO query becomes TRACK_STATUS upstream, and the pinned - MoQTestPublisher inherits NOT_SUPPORTED. Preserve that refusal: widening - query-only demand into a SUBSCRIBE is not an authorized fallback. - No cold request needs that metadata to pick its start, since - [moq-dev/moq#6465](https://github.com/moq-dev/moq/pull/6465) starts every - subscription at a group's first frame. Needs a Moxygen revision whose test - publisher answers TRACK_STATUS; then bump the pins, rebase the profile, and - add the two-relay cell back to `test/ietf-conformance/run.sh` (it ran in - [d96126c65a](https://github.com/moq-dev/moq/commit/d96126c65a52027820884cee7cb9741e79fb674b)). +single-relay cells pass 29/29 on drafts 16 and 18. With TRACK_STATUS answered, +the two-relay cells reach the origin but still fail 29/29. Remaining, in order: + +- **Track properties learned from a draft-16 TRACK_STATUS.** The edge's + TRACK_INFO query becomes TRACK_STATUS at the origin, and a draft-15/16 + TRACK_STATUS_OK has no Track Properties block, so the cluster forwards the + track at the default priority (128, not the publisher's 200) and would drop + DYNAMIC_GROUPS and unknown track properties the same way. Draft-17 already + SUBSCRIBEs for this reason; whether 15/16 should too is open. +- **GROUP_ORDER in a draft-18 TRACK_STATUS_OK.** Moxygen writes GROUP_ORDER as + a REQUEST_OK parameter, which the draft-18 decoder refuses as a known + parameter on the wrong message, so the origin fails every TRACK_INFO. - **The direct draft-18 control.** One checkpoint run failed upstream's "Large scale: Many groups and objects" with `Group Mismatch: Actual=1 Expected=0` without any relay; it has not recurred in six runs since. The verifier expects groups in order although each arrives on its own stream, which its - own comment calls a spurious failure. Resolve it at the verifier, not with a - retry. + own comment calls a spurious failure. Wait for a recurrence in nightly, then + resolve it at the verifier, not with a retry. Then run both direct-origin controls and all four relay cells, fix any independently justified forwarding failures at source, and keep the nightly diff --git a/test/ietf-conformance/README.md b/test/ietf-conformance/README.md index b089af5a52d..9c0f88cbfda 100644 --- a/test/ietf-conformance/README.md +++ b/test/ietf-conformance/README.md @@ -6,7 +6,8 @@ builds the Moxygen revision and matching Meta dependencies pinned in `conformance_test.sh` through this checkout's relay. This is separate from the published-package smoke matrix. -Sections 1-7 run on both shared drafts, 16 and 18, through one relay. Each cell +Sections 1-7 run on both shared drafts, 16 and 18, through one relay and +through a two-relay cluster joined by native lite-07. Each cell runs 29 upstream SUBSCRIBE cases. A direct-origin control runs the same cases once per draft, separating suite or pin failures from relay behavior. FETCH is skipped using upstream's `SKIP_FETCH`; section 8 uses PUBLISH and is outside the @@ -60,9 +61,10 @@ Measured over six consecutive runs: | 16 | 29 pass | 29 pass | | 18 | 29 pass | 29 pass | -A two-relay cell is not run yet. The edge needs metadata from TRACK\_STATUS -before subscribing, and the pinned MoQTestPublisher inherits NOT\_SUPPORTED for -that request, so every case fails before media. The relay preserves that -refusal, rather than widening query-only demand into SUBSCRIBE. The -[passthrough quest](../../quest/m1/ietf-extension/README.md) tracks adding the -cell back once a pinned publisher answers it. +The two-relay cells need the origin to answer TRACK\_STATUS, which the edge's +metadata query becomes upstream. The pinned MoQTestPublisher inherits +NOT\_SUPPORTED for it, so +[`nix/moxygen-track-status.patch`](../../nix/moxygen-track-status.patch) answers +with what its SUBSCRIBE\_OK carries until the upstream fix is pinned. The +[passthrough quest](../../quest/m1/ietf-extension/README.md) tracks the +cluster failures that remain. diff --git a/test/ietf-conformance/run.sh b/test/ietf-conformance/run.sh index 77dfc024294..85804719bdd 100755 --- a/test/ietf-conformance/run.sh +++ b/test/ietf-conformance/run.sh @@ -1,5 +1,6 @@ #!/usr/bin/env bash -# Pinned upstream conformance under the relay profile, subgroup zero, one relay. +# Pinned upstream conformance under the relay profile, subgroup zero, through one +# relay and through a two-relay cluster. set -euo pipefail if [[ "${GITHUB_ACTIONS:-}" != true && "${LOCAL_SUPERVISED:-}" != 1 ]]; then @@ -37,7 +38,7 @@ openssl req -config "$HARNESS_RUN/openssl.cnf" -x509 -newkey rsa:2048 -nodes \ >"$HARNESS_RUN/certificate.log" 2>&1 start_relay() { - local label="$1" version="$2" + local label="$1" version="$2" seed="${3:-}" harness_port "$label" local port="$HARNESS_PORT" config="$HARNESS_RUN/$label.toml" cat >"$config" <>"$config" <&2 exit 1 fi -echo "Moxygen conformance: all shared drafts passed through the relay" +echo "Moxygen conformance: all shared drafts and both topologies passed" From 6d001757c068dcecd37104c861660e906abb4825 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 25 Oct 2026 09:47:54 -0700 Subject: [PATCH 3/4] test(ietf): pass moq-test through a two-relay cluster on draft 18 Moxygen wrote GROUP_ORDER in TRACK_STATUS_OK, which draft-18 forbids outside SUBSCRIBE, PUBLISH_OK and FETCH, so the origin refused every cluster metadata query. The local patch drops it, as facebookexperimental/moxygen#264 now does. Draft 16 has no cluster cell: its TRACK_STATUS_OK cannot carry the track properties, and that draft is being dropped. All cells pass over three runs, which finishes the passthrough quest. The reorder flake it tracked is documented with the harness. Co-Authored-By: Claude Opus 5.5 --- nix/moxygen-track-status.patch | 16 ++++ nix/moxygen.nix | 3 +- quest/m1/README.md | 1 - quest/m1/api-audit/model-private.md | 4 +- quest/m1/ietf-extension/README.md | 138 ---------------------------- quest/m1/interop-coverage/README.md | 1 - quest/m1/smoke-moxygen.md | 4 - test/ietf-conformance/README.md | 45 ++++++--- test/ietf-conformance/run.sh | 10 +- 9 files changed, 58 insertions(+), 164 deletions(-) delete mode 100644 quest/m1/ietf-extension/README.md diff --git a/nix/moxygen-track-status.patch b/nix/moxygen-track-status.patch index 0e7963b506c..ab6583a156b 100644 --- a/nix/moxygen-track-status.patch +++ b/nix/moxygen-track-status.patch @@ -1,3 +1,19 @@ +--- a/moxygen/MoQFramer.cpp ++++ b/moxygen/MoQFramer.cpp +@@ -3362,10 +3362,9 @@ RequestOk RequestOk::fromTrackStatusOk(const TrackStatusOk& trackStatusOk) { + folly::to_underlying(TrackRequestParamKey::EXPIRES), + static_cast(trackStatusOk.expires.count())); + +- // Add group order parameter +- requestOk.requestSpecificParams.emplace_back( +- folly::to_underlying(TrackRequestParamKey::GROUP_ORDER), +- folly::to_underlying(trackStatusOk.groupOrder)); ++ // No GROUP_ORDER: it is only a SUBSCRIBE, PUBLISH_OK or FETCH parameter, ++ // and the receiver closes the session on it anywhere else. As in ++ // SUBSCRIBE_OK, a publisher's group order is a track property. + + // Add the LARGEST_OBJECT param if present + if (trackStatusOk.largest) { --- a/moxygen/moqtest/MoQTestPublisher.cpp +++ b/moxygen/moqtest/MoQTestPublisher.cpp @@ -88,18 +88,25 @@ void MoQTestPublisher::cancelAll() { diff --git a/nix/moxygen.nix b/nix/moxygen.nix index 02d157c8ba7..cec2a02af83 100644 --- a/nix/moxygen.nix +++ b/nix/moxygen.nix @@ -35,7 +35,8 @@ stdenv.mkDerivation { # an END_OF_TRACK directly after the last object. Every other check stays. ./moxygen-profile.patch # The test origin answers TRACK_STATUS, which a relay cluster's edge sends - # before subscribing. Drop at the next pin bump once + # before subscribing, without the GROUP_ORDER parameter draft-18 forbids in + # TRACK_STATUS_OK. Drop at the next pin bump once # https://github.com/facebookexperimental/moxygen/pull/264 lands. ./moxygen-track-status.patch ]; diff --git a/quest/m1/README.md b/quest/m1/README.md index 84cf6e6c7b1..5bb2c0d2954 100644 --- a/quest/m1/README.md +++ b/quest/m1/README.md @@ -109,7 +109,6 @@ Catalog HLS field. - [Generated C bindings](/quest/m1/c/README.md) - C generated from moq-ffi ships as `moq-c` 0.8.0 and replaces the hand-written libmoq - [Exact timestamps in every binding](/quest/m1/binding-timestamps.md) - bindings carry `{value, scale}` timestamps instead of lossy microseconds - [Binding durations](/quest/m1/ffi-durations.md) - moq-ffi takes and returns durations as a typed duration instead of `_us` integers, so wrappers stop converting by hand -- [moq-transport passthrough](/quest/m1/ietf-extension/README.md) - object IDs, properties, end status, datagram objects, and unknown track properties survive moq-transport -> moq-relay (and a relay cluster) -> moq-transport; moxygen's moq-test sections 1-7 pass nightly under the relay conformance profile, except cases that use a subgroup other than 0 - [Announcement takeover](/quest/m1/announce-takeover.md) - a new independent route explicitly continues instances served through its predecessor, with shared Rust/JS semantics and lite-07 wire - [Subscribe ranges](/quest/m1/subscribe-ranges/README.md) - a lite-07 SUBSCRIBE asks for past and live ranges with absolute or relative ends, in either order, and replaces FETCH; merged ranges never starve a subscriber; relays fill misses by range, including over moq-transport - [Updates reach fills](/quest/m1/fill-updates.md) - a reader's new priority, delay, order, or earlier start reaches fills already queued or taken diff --git a/quest/m1/api-audit/model-private.md b/quest/m1/api-audit/model-private.md index 5c91fd59224..b81e7313043 100644 --- a/quest/m1/api-audit/model-private.md +++ b/quest/m1/api-audit/model-private.md @@ -23,8 +23,8 @@ for a named consumer. `Track.Producer.dropGroups` become internal again (`dropGroups` returns to `hooks.dropTrackGroups`). Keep `create_frame(self)`: hang and moq-mux use it. - The `// pending:` and `// private:` marks and their lint are already gone. - The moq-transport passthrough exception stays documented in - [moq-transport passthrough](/quest/m1/ietf-extension/README.md). + The moq-transport passthrough fields stay crate-private, as + [moq-transport stays private](/quest/m1/api-audit/ietf-private.md) requires. Rejected: the toy-transport integration tests proving a third party could build from the public model; the goal they proved is dropped. diff --git a/quest/m1/ietf-extension/README.md b/quest/m1/ietf-extension/README.md deleted file mode 100644 index 87cda05e25c..00000000000 --- a/quest/m1/ietf-extension/README.md +++ /dev/null @@ -1,138 +0,0 @@ -# [M] moq-transport passthrough - -## Goal - -A moq-transport track forwarded moq-transport -> moq-relay -> moq-transport, -including across a relay cluster, arrives with the publisher's object IDs, -object properties, END_OF_GROUP status objects, datagram object IDs, and -unknown track properties intact. -Tracks advertising DYNAMIC_GROUPS relay NEW_GROUP_REQUEST back to the -original publisher, with redundant requests coalesced. Across the cluster, -refresh demand carries them on native lite-07. -moxygen's moq-test conformance (sections 1-7, except cases that use a -subgroup other than 0) passes through moq-relay in a nightly CI job under the -relay conformance profile. The profile accepts the track's publisher priority -on every group and an END_OF_TRACK directly after the last object. It still -checks forwarded object IDs, properties, END_OF_GROUP and FIRST_OBJECT. - -Boundaries: - -- The public API does not change, except one documented field on - `setup::Extensions`. Every carried field is crate-private in moq-net. -- Plain readers are unchanged: the public Rust and JS APIs, the media layer, - and a moq-lite session without the extension see frames in order with gaps - invisible, and no properties. -- Non-goals: full moq-transport parity, subgroups other than 0, accepting - PUBLISH (moq-test section 8 needs it), stitching a FETCH across unmarked - gaps, JS support, and an application API for any of these fields. - -## Plan - -Decided with the maintainer on 2026-10-13. This revisits the -[#4926](https://github.com/moq-dev/moq/issues/4926) decline for relays only. -That decline's cache-fragmentation concern does not apply here, because -private Object IDs distinguish gaps from frame ordinals, and a FETCH it -cannot answer unambiguously is refused. - -- **Carried:** object ID gaps as crate-private absolute Object IDs, object - properties as a decoded KVP list, END_OF_GROUP status objects at their - object ID, datagrams past object 0, and unknown track properties. -- **Subgroup 0 only.** Decided 2026-10-16 in - [#5704](https://github.com/moq-dev/moq/pull/5704), which built and then - reverted a subgroup relay: "don't implement sub-groups. I can't see how - they could work. Anybody that wants to use an IETF client needs to use - sub-group 0." Other subgroups are not carried, and their per-stream - publisher priority with them; egress ranks within a subscription by the - track's publisher priority. -- **FETCH** serves a group only if every object ID jump in it is marked by a - Prior Object ID Gap property. Anything else gets `NOT_SUPPORTED`. -- **Relay hops** carry the fields over lite-07 through a SETUP extension - named `ietf`, not a separate ALPN. It is off by default, and moq-relay - turns it on. A non-relay peer that negotiated it would receive properties - and status objects it can only discard. -- **Dynamic groups** are supported rather than treated as an opaque track - property. The public `Demand::wanted` path (#5930, through relays in - #6030) carries NEW_GROUP_REQUEST upstream and coalesces redundant requests - until the largest group advances, so dynamic groups are no longer part of - the private carve-out below. -- **Immutable properties** keep their serialized inner bytes, and the track - its ingress moq-transport version, including across relay hops. Refuse forwarding between - the absolute-key/QUIC-varint drafts 14/15, delta-key/QUIC-varint draft 16, - and delta-key/leading-ones-varint drafts 17+ when immutable properties - are present. Draft 17 is a separate family because it forbids 7-byte varints. - Rewriting them would violate immutability; forwarding unchanged - across those families would change their meaning. -- **Crate-private on purpose.** These fields never become public API - ([moq-transport stays private](/quest/m1/api-audit/ietf-private.md)); a - relay that is not moq-relay drops them, as plain lite does. -- **Verification:** Rust integration tests per feature in each child, plus - the nightly moxygen job owned here. - -Work owned by this README, after the children land: - -- A nightly CI job runs moxygen's `conformance_test.sh` sections 1-7 with - `moqtest_server --relay_url` publishing the `moq-test-00` prefix to - moq-relay. It runs from a pinned moxygen build, at every draft both moxygen - and moq-relay support, and again across a two-relay cluster. -- The moq-test client fails on any added or removed extension, a missing - END_OF_GROUP status object, and a FIRST_OBJECT bit that differs. Every child must - preserve these byte for byte. -- Check routing first. Sections 1-7 send a SUBSCRIBE for a 16-field namespace - under the announced `moq-test-00` prefix. Mock-clock integration tests now - prove that path on drafts 16 and 18, through one relay and through two relays - joined by lite-07; the live harness reaches the publisher too. -- FETCH cases are skipped upstream. Section 8 stays out while PUBLISH is - declined. - -### Conformance profile and remaining work - -Decided with the maintainer on 2026-10-23: keep the relay's track-level egress -priority and its END_OF_TRACK, and verify under a relay conformance profile -instead of the literal upstream suite. Changing per-group priority transport or -terminal representation to satisfy one verifier would cost scheduling and wire -complexity for no behavioral gain. - -The pinned harness is in -[`test/ietf-conformance`](../../../test/ietf-conformance/README.md), with Moxygen -`c27ae6358df62a1195da479b2e07c50dd8174279` and its matching dependency revisions. -It keeps all selected case arguments and deadlines, excludes only the agreed -nonzero-subgroup cases, and fails on client crashes or publisher reconnects. -`nix/moxygen-profile.patch` applies the profile to the client verifier: it also -accepts the track's advertised priority on every group, and one END_OF_TRACK -directly after the last object, whose stream ends no payload group. All other -upstream checks stay, including object IDs, properties, END_OF_GROUP and -FIRST_OBJECT. - -`nix/moxygen-track-status.patch` makes the test origin answer TRACK_STATUS -with what its SUBSCRIBE_OK carries, until -[facebookexperimental/moxygen#264](https://github.com/facebookexperimental/moxygen/pull/264) -lands and the pins move past it. - -Measured on the profile over six runs: the direct-origin controls and the -single-relay cells pass 29/29 on drafts 16 and 18. With TRACK_STATUS answered, -the two-relay cells reach the origin but still fail 29/29. Remaining, in order: - -- **Track properties learned from a draft-16 TRACK_STATUS.** The edge's - TRACK_INFO query becomes TRACK_STATUS at the origin, and a draft-15/16 - TRACK_STATUS_OK has no Track Properties block, so the cluster forwards the - track at the default priority (128, not the publisher's 200) and would drop - DYNAMIC_GROUPS and unknown track properties the same way. Draft-17 already - SUBSCRIBEs for this reason; whether 15/16 should too is open. -- **GROUP_ORDER in a draft-18 TRACK_STATUS_OK.** Moxygen writes GROUP_ORDER as - a REQUEST_OK parameter, which the draft-18 decoder refuses as a known - parameter on the wrong message, so the origin fails every TRACK_INFO. -- **The direct draft-18 control.** One checkpoint run failed upstream's "Large - scale: Many groups and objects" with `Group Mismatch: Actual=1 Expected=0` - without any relay; it has not recurred in six runs since. The verifier - expects groups in order although each arrives on its own stream, which its - own comment calls a spurious failure. Wait for a recurrence in nightly, then - resolve it at the verifier, not with a retry. - -Then run both direct-origin controls and all four relay cells, fix any -independently justified forwarding failures at source, and keep the nightly -gate passing before deleting this README. - -## Related - -- [#4926](https://github.com/moq-dev/moq/issues/4926) - the declined request for an object-level API; this carries the fields for relays only -- [moq-transport stays private](/quest/m1/api-audit/ietf-private.md) - no public moq-net item exists only for moq-transport diff --git a/quest/m1/interop-coverage/README.md b/quest/m1/interop-coverage/README.md index 287c479fc0c..a975434102b 100644 --- a/quest/m1/interop-coverage/README.md +++ b/quest/m1/interop-coverage/README.md @@ -53,4 +53,3 @@ coverage on `main` at `3ab6cbf81d`. - [Runner client bump](/quest/m2/release/runner-client-bump.md) - after the cut publishes #5083, the runner's Rust client stops stamping every object - [Interop audit](/quest/m2/audit-interop.md) - classifies the red cells these cases produce -- [moq-transport passthrough](/quest/m1/ietf-extension/README.md) - Moxygen's moq-test conformance through moq-relay diff --git a/quest/m1/smoke-moxygen.md b/quest/m1/smoke-moxygen.md index 70d1243490e..270a8a35314 100644 --- a/quest/m1/smoke-moxygen.md +++ b/quest/m1/smoke-moxygen.md @@ -32,7 +32,3 @@ Reported upstream as acknowledged 2026-10-16; check it and the Smoke job. The fix is a registered default (Largest Object or Next Group Start) for drafts that lack LargestGroup. Do not accept 0xFA in moq-relay. - -## Related - -- [moq-transport passthrough](/quest/m1/ietf-extension/README.md) - broader object forwarding and moq-test conformance diff --git a/test/ietf-conformance/README.md b/test/ietf-conformance/README.md index 9c0f88cbfda..6a02f5a3c5e 100644 --- a/test/ietf-conformance/README.md +++ b/test/ietf-conformance/README.md @@ -6,8 +6,10 @@ builds the Moxygen revision and matching Meta dependencies pinned in `conformance_test.sh` through this checkout's relay. This is separate from the published-package smoke matrix. -Sections 1-7 run on both shared drafts, 16 and 18, through one relay and -through a two-relay cluster joined by native lite-07. Each cell +Sections 1-7 run on both shared drafts, 16 and 18, through one relay, and on +draft 18 through a two-relay cluster joined by native lite-07. Draft 16 has no +cluster cell: its TRACK\_STATUS\_OK has no Track Properties block, so the +cluster's origin cannot learn the track's priority from it. Each cell runs 29 upstream SUBSCRIBE cases. A direct-origin control runs the same cases once per draft, separating suite or pin failures from relay behavior. FETCH is skipped using upstream's `SKIP_FETCH`; section 8 uses PUBLISH and is outside the @@ -54,17 +56,32 @@ Every other check is upstream's: forwarded object IDs, properties, payloads, END\_OF\_GROUP status objects and header bits, and FIRST\_OBJECT. Any other priority, a second END\_OF\_TRACK, or one anywhere else still fails. -Measured over six consecutive runs: +Measured over three consecutive runs, after six earlier runs of the +direct-origin and one-relay cells: -| Draft | Direct origin | One relay | -| --- | --- | --- | -| 16 | 29 pass | 29 pass | -| 18 | 29 pass | 29 pass | +| Draft | Direct origin | One relay | Two relays | +| --- | --- | --- | --- | +| 16 | 29 pass | 29 pass | not run | +| 18 | 29 pass | 29 pass | 29 pass | -The two-relay cells need the origin to answer TRACK\_STATUS, which the edge's -metadata query becomes upstream. The pinned MoQTestPublisher inherits -NOT\_SUPPORTED for it, so -[`nix/moxygen-track-status.patch`](../../nix/moxygen-track-status.patch) answers -with what its SUBSCRIBE\_OK carries until the upstream fix is pinned. The -[passthrough quest](../../quest/m1/ietf-extension/README.md) tracks the -cluster failures that remain. +## Local Moxygen patches + +A cluster's edge asks the origin about a track with TRACK\_STATUS before +subscribing. +[`nix/moxygen-track-status.patch`](../../nix/moxygen-track-status.patch) makes +the pinned MoQTestPublisher answer it with what its SUBSCRIBE\_OK carries, +instead of NOT\_SUPPORTED, and stops Moxygen writing a GROUP\_ORDER parameter +in TRACK\_STATUS\_OK, which draft 18 forbids there. Drop it at the next pin +bump once +[facebookexperimental/moxygen#264](https://github.com/facebookexperimental/moxygen/pull/264) +is in the pinned revision. + +## Known verifier flake + +One early run of the direct draft-18 control failed upstream's "Large scale: +Many groups and objects" with `Group Mismatch: Actual=1 Expected=0`, without +any relay. It has not recurred since. The verifier keeps one cursor per +subscription and expects groups in order, although each group arrives on its +own stream; its own comment calls that a spurious failure. If it recurs in +nightly, fix the verifier in the profile patch with per-group cursors rather +than retrying. diff --git a/test/ietf-conformance/run.sh b/test/ietf-conformance/run.sh index 85804719bdd..0b17af77f82 100755 --- a/test/ietf-conformance/run.sh +++ b/test/ietf-conformance/run.sh @@ -1,6 +1,6 @@ #!/usr/bin/env bash # Pinned upstream conformance under the relay profile, subgroup zero, through one -# relay and through a two-relay cluster. +# relay, and on draft 18 through a two-relay cluster. set -euo pipefail if [[ "${GITHUB_ACTIONS:-}" != true && "${LOCAL_SUPERVISED:-}" != 1 ]]; then @@ -115,7 +115,11 @@ run_suite() { # The pinned MoQVersions.h supports drafts 16 and 18; this relay supports both. failed=0 for version in 16 18; do - for topology in single cluster; do + # A draft-16 TRACK_STATUS_OK has no Track Properties, so a cluster's origin + # cannot learn the track's priority from it; draft 16 is not a cluster target. + topologies=(single) + [[ "$version" == 16 ]] || topologies+=(cluster) + for topology in "${topologies[@]}"; do label="$topology-$version" start_relay "$label-origin" "$version" ingress="$HARNESS_PORT" @@ -149,4 +153,4 @@ if ((failed != 0)); then echo "Moxygen conformance failed; see each cell's client and endpoint logs" >&2 exit 1 fi -echo "Moxygen conformance: all shared drafts and both topologies passed" +echo "Moxygen conformance: every draft and topology passed" From 1c0fe3bbe4037834a31beb93872435a9cad968b6 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 25 Oct 2026 15:43:55 -0700 Subject: [PATCH 4/4] chore(quest): drop the link to the finished ietf-private quest Co-Authored-By: Claude Opus 5.5 --- quest/m1/api-audit/model-private.md | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/quest/m1/api-audit/model-private.md b/quest/m1/api-audit/model-private.md index b81e7313043..1d6dfa258ec 100644 --- a/quest/m1/api-audit/model-private.md +++ b/quest/m1/api-audit/model-private.md @@ -23,8 +23,7 @@ for a named consumer. `Track.Producer.dropGroups` become internal again (`dropGroups` returns to `hooks.dropTrackGroups`). Keep `create_frame(self)`: hang and moq-mux use it. - The `// pending:` and `// private:` marks and their lint are already gone. - The moq-transport passthrough fields stay crate-private, as - [moq-transport stays private](/quest/m1/api-audit/ietf-private.md) requires. + The moq-transport passthrough fields stay crate-private. Rejected: the toy-transport integration tests proving a third party could build from the public model; the goal they proved is dropped.