Skip to content
12 changes: 12 additions & 0 deletions quest/m0/nightly/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,18 @@ to the PR gate; the full run costs about 17 minutes per PR.

Public API: none. Wire: none.

Decided 2026-11-09 (maintainer, /quest-complete), from the triage of runs
[42377867523](https://github.com/moq-dev/moq/actions/runs/42377867523) and
[42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545): four
more children for the causes nothing owned. Dropped: one-night bench moves
near the paired threshold (re-check if one repeats), a one-off
`remote error: 52` in the browser bench smoke, and failures main or
[moq-dev/moq#8290](https://github.com/moq-dev/moq/pull/8290) already fixes.

## Required

- [Nightly's macOS platform job passes](/quest/m0/nightly/macos.md) - `moq-audio`'s tests compile on macOS again and the websocket version test's timeout is fixed if it recurs
- [The qmux startup benchmark's fake peer sends transport parameters](/quest/m0/nightly/qmux-bench-peer.md) - Audio quality's WebSocket startup step stops timing out
- [The moq-net session bench completes moq-transport-22 at 16 broadcasts](/quest/m0/nightly/session-bench-handshake.md) - the bench workload stops panicking on a handshake timeout
- [Bisect the moq-audio bus and moq-mux parameter_sets regressions](/quest/m0/nightly/bench-regressions.md) - two real benchmark regressions are explained and fixed or accepted
- [Moxygen conformance cells pass again](/quest/m0/nightly/moxygen-conformance.md) - the first-object signal and integer extensions survive, or moxygen's divergence is skipped with an upstream link
32 changes: 32 additions & 0 deletions quest/m0/nightly/bench-regressions.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
# [M] Bisect the moq-audio bus and moq-mux parameter_sets regressions

## Goal

Two benchmark regressions Nightly flagged are explained and either fixed or
accepted with a recorded reason:

- 13 `moq-audio/bus` cases (`bus_minus`, `bus_mixes`, `bus_sources`), +16 to
+29% together, between `fbef872` and `b5805859`.
- `moq-mux/parameter_sets h264-split-pps/distinct`, +14% / +26% / +46% at
n = 16 / 256 / 4096, so the cost grows with n.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete from the triage of Nightly
runs [42377867523](https://github.com/moq-dev/moq/actions/runs/42377867523)
(the bus set, in "Rust supplemental benchmarks") and
[42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545) (the
parameter sets, in "Rust shared")). Both look real: the bus cases moved
together, and the parameter_sets cost scales with n, which points at a
per-item cost that grew rather than noise.

Suspects from the triage, to confirm by bisecting with the paired
comparison (`bench/README.md#paired-comparison`), not by reading:
[moq-dev/moq#8121](https://github.com/moq-dev/moq/pull/8121) (encoders take a
join) and [moq-dev/moq#8010](https://github.com/moq-dev/moq/pull/8010)
(comfort noise) for the bus; [moq-dev/moq#8140](https://github.com/moq-dev/moq/pull/8140)
and [moq-dev/moq#8199](https://github.com/moq-dev/moq/pull/8199) for the
parameter sets. Name the culprit PR under `## Caused by`. A regression that
buys correctness may stay, with the trade recorded in the PR.

Public API: none expected. Wire: none.
25 changes: 25 additions & 0 deletions quest/m0/nightly/moxygen-conformance.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
# [M] Moxygen conformance cells pass again

## Goal

Nightly's `ietf-conformance / moxygen` job passes. Today its cells fail with
`Missing Begins With First Object Signal for group=0 subgroup=0`, and with
`Extension Error=2 ... Expected size: 1 Actual size: 0` in the Extensions 1/2
and Object/Group Counts 3/4 cells.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete from the triage of Nightly
runs [42377867523](https://github.com/moq-dev/moq/actions/runs/42377867523)
and [42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545);
also red on 2026-11-06). ✅ Plan it (rejected: drop the lane), since it
keeps Nightly red, though moq-transport compatibility is low priority by the
repository's rules.

Find which side is wrong before changing anything: our relay dropping the
subgroup's first-object flag and an integer extension's value, or moxygen's
expectations diverging from the draft we implement. Fix ours when it is
ours. When moxygen is wrong, record it upstream and skip just those cells
with a link, rather than adding cost to match it.

Public API: none expected. Wire: moq-transport only, if ours is wrong.
23 changes: 23 additions & 0 deletions quest/m0/nightly/qmux-bench-peer.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
# [XS] The qmux startup benchmark's fake peer sends transport parameters

## Goal

Nightly's Audio quality job passes its "WebSocket fallback startup benchmark"
step: `just js qmux-bench` completes instead of failing with
`error: evaluate: Error: handshake timeout`.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete from the triage of Nightly
runs [42377867523](https://github.com/moq-dev/moq/actions/runs/42377867523)
and [42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545)).
Failed both nights.

The Bun server in `js/net/bench/qmux.ts` never sends QMux transport
parameters, and the JS session now waits for them before it opens.
[moq-dev/moq#8290](https://github.com/moq-dev/moq/pull/8290) fixes the same
fake-peer handshake in `js/qmux/bench/datagrams.ts`, but Nightly runs the
`js/net` copy. Make the fake peer send them the way the real server does,
ideally through the same helper as `datagrams.ts` so the two cannot drift.

Public API: none. Wire: none.
19 changes: 19 additions & 0 deletions quest/m0/nightly/session-bench-handshake.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
# [S] The moq-net session bench completes moq-transport-22 at 16 broadcasts

## Goal

Nightly's "Rust bench targets" workload runs `moq-net`'s session bench to the
end: `session_delivery_broadcasts/moq-transport-22/.../broadcasts=16` no
longer panics with `server handshake failed: Timeout`
(`harness.rs:233`).

## Plan

Planned 2026-11-09 (maintainer, /quest-complete from the triage of Nightly
run [42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545),
where it first failed this way). Reproduce it first, then fix the cause:
a handshake that stalls under that load is a bug in the session or the
harness, never a reason to raise the timeout. Check whether the harness
reuses state across cells and whether lite versions pass the same cell.

Public API: none expected. Wire: none.
17 changes: 15 additions & 2 deletions quest/m0/test-flakes-2/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -216,8 +216,18 @@ OS picked for TCP and failed a local run while other checks held that UDP port.
Added 2026-11-08 (maintainer's standing rule that flakes join this epic): moq-ffi's
raw media warmup test, which failed in the merge queue for
[moq-dev/moq#8272](https://github.com/moq-dev/moq/pull/8272) when its reader skipped one of three
buffered groups on the FFI runtime's real clock. It is quarantined with `#[ignore]` until its
child fixes it.
buffered groups on the FFI runtime's real clock.
[moq-dev/moq#8329](https://github.com/moq-dev/moq/pull/8329) moved it to the paused test
clock, re-enabled it, and deleted its child.

Added 2026-11-09 (maintainer, /quest-complete): `js/net`'s compressed-frame
cap test and lite-07 re-dropped re-offer test, which failed under load during
the datagram-queues runs and were quarantined with `test.skip` by
[moq-dev/moq#8333](https://github.com/moq-dev/moq/pull/8333); and moq-net's
two random-mesh withdrawal tests, which hit nextest's 60 s kill in
[moq-dev/moq#8247](https://github.com/moq-dev/moq/pull/8247)'s local
`just check` and are quarantined with `#[ignore]` by
[moq-dev/moq#8336](https://github.com/moq-dev/moq/pull/8336).

Public API: none. Wire: none.

Expand All @@ -227,3 +237,6 @@ Public API: none. Wire: none.
- [js/net's bandwidth sampling test holds up under load](/quest/m0/test-flakes-2/bandwidth-sampling.md) - the sampling test counts ticks without racing the mocked clock, and its `test.skip` is removed
- [js/net's widened max-age re-drop test ends cleanly, not with a reset](/quest/m0/test-flakes-2/widened-max-age-reset.md) - the test finishes the track only after the re-drop reply arrives, so the end is a FIN, and drops its wall-clock assert
- [bench/relay.test.py's stream-transport test gets a port free for TCP and UDP](/quest/m0/test-flakes-2/relay-port-pair.md) - the test binds TCP and UDP to one number free in both, so parallel checks can't collide with it
- [js/net's compressed-frame cap test holds up under load](/quest/m0/test-flakes-2/compressed-frame-cap.md) - the cap is proven without a 32 MiB payload, and its `test.skip` is removed
- [js/net's lite-07 re-dropped re-offer test holds up under load](/quest/m0/test-flakes-2/redropped-reoffer.md) - update 3 is ordered before the re-offer opens by an event, not a yield count, and its `test.skip` is removed
- [moq-net's random-mesh withdrawal tests fit the gate](/quest/m0/test-flakes-2/mesh-withdraw-runtime.md) - the 200-seed mesh tests cost seconds of CPU, and their `#[ignore]` is removed
27 changes: 27 additions & 0 deletions quest/m0/test-flakes-2/compressed-frame-cap.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
# [S] js/net's compressed-frame cap test holds up under load

## Goal

`js/net/src/group.test.ts`'s "a compressed frame is refused unless its
worst-case slice fits a peer's frame cap" passes on a loaded host, and its
`test.skip` is removed. It still proves the cap: a frame whose worst-case
slice exceeds the peer's cap is refused, and one that fits is accepted.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete), under the repository's
flaky-test rule. Quarantined by
[moq-dev/moq#8333](https://github.com/moq-dev/moq/pull/8333) after the
whole `js/net` package failed it 2 of 30 times under 48 CPU burners on 24
cores, first seen in [moq-dev/moq#8326](https://github.com/moq-dev/moq/pull/8326)'s
loaded runs.

The failure is bun's per-test limit ("timed out after 5000ms", at 5840 ms and
10442 ms). Alone it takes about 0.85 s: it deflates and inflates a random
payload of almost 32 MiB, so the CPU work, not a race, runs past the limit
under load. This resembles the flate tail-shed test fixed in
[moq-dev/moq#8137](https://github.com/moq-dev/moq/pull/8137). Prove the cap
with a payload sized relative to a small cap (or a cap the test sets low)
rather than a full-size one; never raise the timeout.

Public API: none. Wire: none.
37 changes: 37 additions & 0 deletions quest/m0/test-flakes-2/mesh-withdraw-runtime.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
# [S] moq-net's random-mesh withdrawal tests fit the gate

## Goal

`moq-net`'s `mesh_withdraw::random_meshes_retract_once` and
`capped_random_meshes_retract_once` run in a few seconds on a loaded host,
well under nextest's 60 s kill (`.config/nextest.toml`, whose stated ceiling
for an honest test is about 9 s), and their `#[ignore]` and
`sh/gh/test-inventory.allow` entries are removed. They still check every one
of the 200 seeded meshes retracts exactly once.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete): ✅ quarantine now and plan
this quest (rejected: leave them in the gate and only plan; move them to
nightly without a quest). Reported by
[moq-dev/moq#8247](https://github.com/moq-dev/moq/pull/8247)'s iterate, whose
`just check` saw both hit the 60 s kill. On `main` they take about 40 s and
45 s even alone, so any load pushes them over. CI passed them, so this is cost,
not a race.

Find where the time goes before changing anything. The tests run on
`#[kio::test]`'s single-threaded simulated clock, where every wait is
`kio::clock::sim::idle()`, so the 300 ms hold costs no wall time: the 40 s is
CPU spent across the 200 seeds in `withdrawal_counts`, each building a mesh,
publishing, closing, and settling. Profile one seed (`cargo flamegraph` or
`perf` on the test binary) and cut the per-seed cost it shows, such as mesh
setup the seeds could share or work that grows with the mesh on each route
change. Cutting seeds is the last resort: the counts in each test's doc
comment came from all 200. If neither fits, move them to the nightly suite
with a caller instead of ignoring them. Never raise the timeout.

[Route layer acceptance](/quest/m2/cluster-routing/routes/acceptance.md)
owns the `residual_*` counterexamples in the same file; keep those as they
are.

Public API: none. Wire: none.
29 changes: 29 additions & 0 deletions quest/m0/test-flakes-2/redropped-reoffer.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# [S] js/net's lite-07 re-dropped re-offer test holds up under load

## Goal

`js/net/src/lite/publisher.test.ts`'s "lite draft-07: a re-offer dropped
again before it opens stays stale at the end" passes on a loaded host, and
its `test.skip` is removed. Update 3 is guaranteed to land before the
re-offer of group 1 opens, so the end still reports group 1 as stale.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete), under the repository's
flaky-test rule. Quarantined by
[moq-dev/moq#8333](https://github.com/moq-dev/moq/pull/8333) after the whole
`js/net` package failed it 1 of 30 times under 48 CPU burners on 24 cores,
first seen in [moq-dev/moq#8326](https://github.com/moq-dev/moq/pull/8326)'s
loaded runs.

The failure: after 22 ms, `holesToEnd` returned `[]` where
`[["drop", 1n, 1n, 3n]]` was expected (`publisher.test.ts:1572`). Suspected,
not confirmed: the write budget is already stubbed, but the test sends update
2, waits one `yieldNow()`, then sends update 3, assuming update 3 lands
before the re-offer opens; under load the re-offer is served first. Reproduce
it first. Then order the test on an event (the update's reply or the
re-offer's queueing), not a count of yields. If the publisher can serve a
re-offer an already-sent update dropped, that is a publisher bug: fix it there,
mirroring Rust's `a_redropped_reoffer_stays_stale_at_the_end_on_lite07`.

Public API: none. Wire: none.
1 change: 1 addition & 0 deletions quest/m1/audio-jitter-target/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,7 @@ native playback against its budgets.
- [Capped budgets](/quest/m1/audio-jitter-target/capped-budgets.md) - the AAC browser budgets are re-measured from nightlies after the TS importer groups each PES
- [Native A/V lag](/quest/m1/audio-jitter-target/native-av-lag.md) - `moq play` reports how far audio trails its presentation clock, and native stops expanding into the gap between arrivals if it settles audio late
- [Native rendition delay](/quest/m1/audio-jitter-target/native-rendition-delay.md) - `moq play` holds each subscribed rendition's catalog `delay` plus its target, as the browser does
- [Chromium Opus stalled share](/quest/m1/audio-jitter-target/opus-stalled-share.md) - the Chromium Opus browser rows stay under the 0.01 stalled-share budget every night

## Closes

Expand Down
2 changes: 1 addition & 1 deletion quest/m1/audio-jitter-target/native-rendition-delay.md
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ behind it.
Planned 2026-11-08 (maintainer, approved follow-up of
[moq-dev/moq#8272](https://github.com/moq-dev/moq/pull/8272), which gives
moq-mux's playout sync per-track targets and the largest-target delay). Milestone:
m1, an additive fix to shipped playback; start after #8272 merges.
m1, an additive fix to shipped playback. #8272 has merged.

On `main`, nothing under `rs/moq-cli/src/play/` reads a rendition's
`delay`; the browser does (`js/watch/src/video/playhead.ts`,
Expand Down
25 changes: 25 additions & 0 deletions quest/m1/audio-jitter-target/opus-stalled-share.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
# [S] Chromium Opus rows stay under the stalled-share budget

## Goal

Nightly's Audio quality browser rows for Chromium Opus pass their
`stalled_share` budget (0.01) every night. Today a different row fails each
night at 0.014: `opus-48000-bursty` on 2026-11-08 and `opus-48000-wide` on
2026-11-07.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete from the triage of Nightly
runs [42377867523](https://github.com/moq-dev/moq/actions/runs/42377867523)
and [42487375545](https://github.com/moq-dev/moq/actions/runs/42487375545)).
✅ A child of this epic, m1 (rejected: leaving it to the watch quest's
grading).

[moq-dev/moq#8058](https://github.com/moq-dev/moq/pull/8058), which
time-stretches the browser's audio ring, merged on 2026-11-09 and may already
remove these stalls. Grade the scheduled nightlies after it (never dispatch
one). If the rows pass, delete this quest. Otherwise find what stalls Opus
that the stretch does not cover and fix it there; do not loosen the budget to
pass.

Public API: none expected. Wire: none.
12 changes: 9 additions & 3 deletions quest/m2/api-shape/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -29,17 +29,22 @@ Added 2026-11-07 (maintainer, /quest-plan from the 2026-11-07 API sweep of open
what later quests grow (`#[non_exhaustive]` on structs and enums m3 to m5
add to), shapes those quests would otherwise break, and this epic's own
work: before the cut, run a "public item used only by its definition"
search over the published crates and remove each hit, starting with
search over the published crates and JS packages and delete each hit, starting with
`moq_tokio::Server::serve_both`, `hang::Catalog::to_json_pretty`, and
`moq_audio::Format::{is_planar, is_float}`, the shortlist of
[the dead code audit](/quest/m3/audit-dead-code.md). The audit itself stays
in m3 for private code.

Decided 2026-11-09 (maintainer, on [moq-dev/moq#8302](https://github.com/moq-dev/moq/pull/8302):
"delete anything unused in general"): ✅ unused public surface is deleted,
not made private, and the pre-cut sweep covers the JS packages too
(rejected: a separate audit quest, since this epic already owns the sweep).

Rules every child follows:

- Removal beats addition. A public item with no caller in this repository
or a known downstream application goes private, unless an m1 or m2 quest plans a
caller.
or a known downstream application is deleted, unless an m1 or m2 quest plans a
caller; it goes private only when internal code still uses it.
- No serde aliases, refusals, or shims for a spelling `release` never
shipped.
- Each child lists its public API and wire impact in its PR, and its Impact
Expand All @@ -58,4 +63,5 @@ Rejected in planning:
- [Transports take shared types](/quest/m2/api-shape/transport.md) - moq-uring takes a congestion factory and rustls configs, and the tokio server request stops exposing its QUIC internals
- [Crates re-export no whole crate](/quest/m2/api-shape/reexports.md) - no moq crate re-exports a whole crate, moq's own included, so a dependency bump is no moq break, and moq-mux's container `Guard` gets a name
- [Media crates' small surface fixes](/quest/m2/api-shape/media.md) - exporters, the container consumer, catalog color range, rate grant, moq-stats, JSON, and MSF each have one path and one name
- [JS Key.sign takes exactly Claims](/quest/m2/api-shape/auth-sign.md) - `@moq/auth` signs only `Claims`, with no legacy `put`/`get` input, like Rust
- [Bindings' small surface fixes](/quest/m2/api-shape/bindings.md) - Go's read cancel, JSON producer construction, and the Python wheel's test module
25 changes: 25 additions & 0 deletions quest/m2/api-shape/auth-sign.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
# [XS] JS Key.sign takes exactly Claims

## Goal

`@moq/auth`'s `Key.sign` takes exactly `Claims`, as Rust's
`moq_auth::Key::sign(&Claims)` does: its input no longer accepts the legacy
`put`/`get` role spelling, so each role has one spelling. Tokens it writes
are unchanged: when the legacy form says the same thing, it is still written
that way so older verifiers accept it, and `verify` still reads legacy
tokens.

## Plan

Planned 2026-11-09 (maintainer, /quest-complete follow-up of
[moq-dev/moq#8296](https://github.com/moq-dev/moq/pull/8296), which settles
`Key.sign`'s caps spelling). ✅ Its own small breaking PR in this epic
(rejected: folding it into #8296 mid-iterate; dropping it).

`sign` in `js/auth/src/key.ts` takes `z.input<typeof ClaimsSchema>` plus
the rate fields, so its type still admits whatever the schema parses. Narrow
the parameter to `Claims` (as `verify` returns it), keep the scope check on
the parsed value, and update `js/auth/api/*.api.md`.

Public API: `@moq/auth` `Key.sign` no longer accepts `put`/`get` input.
Migrate: pass `publish`/`subscribe` roles. Wire: none.
1 change: 1 addition & 0 deletions quest/m2/cluster-routing/routes/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -537,6 +537,7 @@ benchmarks.
## Required

- [Route session integration](/quest/m2/cluster-routing/routes/session.md) - one router per origin node drives the origin and every lite-07 announce stream
- [A relay folds received routes incrementally](/quest/m2/cluster-routing/routes/relay-receive.md) - a received route batch costs the claims it changes, not the origin's size
- [Announcement availability](/quest/m2/cluster-routing/routes/availability.md) - announcements report availability, and a let-go is offline rather than END, in Rust and JS
- [js/net route layer](/quest/m2/cluster-routing/routes/js.md) - `js/net` encodes, decodes, and resolves the route layer
- [Route layer acceptance](/quest/m2/cluster-routing/routes/acceptance.md) - forwarding fixtures ported and the model deleted, incident fixtures, end-to-end checks, and benchmarks
Expand Down
Loading
Loading