From ff6aa4f6bc7224854e9de2c79b0c7ef4dcd9730f Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Tue, 3 Nov 2026 21:14:30 -0800 Subject: [PATCH 1/3] chore(quest): claim quest/m0/backport-resume-owed-groups.md Quest-Claim: 3645ee9e61c9e05e From 454df4d08fef8ad487f0e2522aaa3d92397a4b82 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Tue, 3 Nov 2026 21:25:41 -0800 Subject: [PATCH 2/3] docs(quest): record release resume exposure and contract decision Co-Authored-By: GPT-6.1 Sol --- quest/m0/backport-resume-owed-groups.md | 80 +++++++++++++++++-------- 1 file changed, 54 insertions(+), 26 deletions(-) diff --git a/quest/m0/backport-resume-owed-groups.md b/quest/m0/backport-resume-owed-groups.md index 41b2aa16757..27206c002bb 100644 --- a/quest/m0/backport-resume-owed-groups.md +++ b/quest/m0/backport-resume-owed-groups.md @@ -1,35 +1,63 @@ -# [S] Release resumes owed groups or proves it doesn't need to +# [S] Release resumes owed groups ## Goal A `release` relay's resumed read gets every owed group or a drop for it, -never a hang, if `release` has the hang. Otherwise, this quest records the -evidence that `release` isn't exposed and is deleted. +never a hang. Exposure is reproduced; the source fix and a separately +authorized patch release remain. ## Plan Decided 2026-11-03 (maintainer, /quest-plan of the follow-ups from [moq-dev/moq#7708](https://github.com/moq-dev/moq/pull/7708)): milestone m0, -beside the other backports. It started once #7708, the resume fix on main -that this ports, merged. - -- Exposure is unproven. `origin/release` (moq-net 0.3.19 at `acf91a00b9`) - has no `live_floor`, `judge`, or `Misses`. The shared live floor that - #7708 fixes arrived with - [moq-dev/moq#4741](https://github.com/moq-dev/moq/pull/4741), which is not - on `release`. `release` does have `resume.rs`, so - the symptom may still come through another path. - `release` also speaks draft 22 (`rs/moq-net/src/ietf/version.rs`), so - check drafts 14 to 22. -- Reproduce first. Port #7708's regression tests (the cache-loss case and - the bounded resumes) onto `release`'s older cache model and run them - there. If none fails, delete this quest and record the run in the PR. -- If one fails, fix it on `release` with the smallest change that matches - main's contract. Expect a hand-port, not a cherry-pick, since the cache - models differ. Then cut a patch release under the release line's usual - rules. Bump the version by hand at the semver level the change needs. -- Keep it separate from [the joined-head FETCH backport](/quest/m0/backport-fill-head-race.md). - #7708 says the PUBLISH_DONE grace correction tracked there is a separate - fix. - -Public API: none expected. Wire: none. +beside the other backports. The main fix has merged. + +The exposure investigation is recorded in +[moq-dev/moq#7815](https://github.com/moq-dev/moq/pull/7815). +At `release` commit `da09919e75f285ed1fe0b45b35d34022ffbc14f2` +(moq-net 0.3.20), adapted paused-time session probes show: + +- Cache group 0, leave until the publisher reports unused, publish groups + 1 through 3, then resume at group 0 with a budget covering the history. + Every IETF draft from 14 through 22 returns group 3 instead of group 0. + Applying release's local cursor cap `set_groups(0..1)` leaves the read + waiting for group 0. All seven lite versions, including lite-07-wip, + retain group 0 in both cases. +- The separate bounded `1..3` probe behind a fetched copy delivers groups + 1 and 2 on drafts 14 through 22. It does not establish that an initially + bounded IETF request is accepted: release opens the subscription while + querying metadata, before the reader's preferences reach it. +- Release's movable local cap intentionally parks at its boundary. Main's + bounded EOF assertion is not applicable to this API. + +The root differs from main's shared `live_floor`: +`origin::run_front` queries a source before splicing the reader's demand. +The IETF subscriber snapshots the empty demand as a live join and records +SUBSCRIBE_OK's Largest as its start. `resume::Subscriber::poll_activate` +then discards the whole warm segment when that start exceeds its edge, +without judging the resumed reader's budget. + +The remaining work is a hand-port that preserves the published API and +finite-budget behavior. Do not merely remove the warm-cache gate: main's +fix distinguishes unlimited from finite history, while release exposes a +concrete `max_age: Duration` and has no unlimited option. Forwarding the +actual demand before IETF setup is the recommended direction; it needs +coverage of metadata-only lookups, fetched copies, and concurrent readers. +Keep the original cache-loss and bounded-resume matrices for drafts 14 to +22 and every lite version. + +Keep this separate from [the joined-head FETCH backport](/quest/m0/backport-fill-head-race.md) +and [moq-dev/moq#7680](https://github.com/moq-dev/moq/pull/7680). +Its queued-miss grace correction is a different path. + +After a reviewed source backport lands, use the normal patch-release +process only with explicit release authorization. Do not bump versions, +tag, or publish as part of the implementation attempt. Verify the +published source includes the fix before deleting this quest. + +Public API: preserve the published surface. Wire: no format change expected. + +## Waiting + +- Maintainer: choose whether to continue with a demand-setup hand-port (recommended), preserving finite-budget cache behavior, or explicitly permit a change to that behavior. Removing the warm-cache gate alone changes which stale cache finite readers receive; main's unlimited-only exception cannot be copied into release's older API. +- Maintainer: authorize the separate `moq-net` patch release after the source fix is reviewed and landed; retain this requirement until the published source is verified. Release cuts and version changes require their own authorization. From cb01b838632ba24e4bdab434b9d4cf2f39bce4a6 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Tue, 3 Nov 2026 22:45:14 -0800 Subject: [PATCH 3/3] quest(release): defer the resumed-history fix to the next cut The maintainer chose no backport (2026-11-04): release stays exposed to the IETF resumed-history cache loss until the next main to release cut brings moq-dev/moq#7708. Delete the backport quest and note the exposure beside the MKV one in the cut's Plan so the cut checks both. Co-Authored-By: Claude Opus 5.5 --- quest/m0/README.md | 1 - quest/m0/backport-resume-owed-groups.md | 63 ------------------------- quest/m2/release/cut.md | 5 ++ 3 files changed, 5 insertions(+), 64 deletions(-) delete mode 100644 quest/m0/backport-resume-owed-groups.md diff --git a/quest/m0/README.md b/quest/m0/README.md index d9632cea647..b720e0832bc 100644 --- a/quest/m0/README.md +++ b/quest/m0/README.md @@ -29,7 +29,6 @@ edge, which embeds moq-relay; it moved there on 2026-09-28. - [Backport audio groupDuration](/quest/m0/backport-audio-group-duration.md) - `@moq/publish` 0.5.4 on `release` can pack several Opus frames per group, making Cloudflare's relay drop audio less often - [Backport the SETUP-path and iroh-key fixes](/quest/m0/backport-security-fixes.md) - `release` refuses a conflicting SETUP Path and keeps persisted iroh keys private, in one patch release - [Backport the joined-head FETCH race fix](/quest/m0/backport-fill-head-race.md) - a `release` relay stops FETCHing a group head its own SUBSCRIBE is bringing, which left live readers hanging -- [Release resumes owed groups or proves it doesn't need to](/quest/m0/backport-resume-owed-groups.md) - after main's resume fix, reproduce its cases on `release`, then backport or delete - [CI cost and latency](/quest/m0/ci/README.md) - cloud CI runs only in a batched merge gate as one `just ci` pass per batch, expensive checks run nightly and gate `release`, caches move to R2, and runners are chosen last - [More tests under load](/quest/m0/test-flakes-2/README.md) - load-only and queue failures, one quest per flake, fixed at the cause, since each one fails a whole batch - [A waiter's record limit changes no behavior](/quest/m0/kio-waiter-records.md) - two moq-net tests stop failing when kio's waiters remember 16 lists, and the limit becomes a tuning knob worth 3.9% per delivered group diff --git a/quest/m0/backport-resume-owed-groups.md b/quest/m0/backport-resume-owed-groups.md deleted file mode 100644 index 27206c002bb..00000000000 --- a/quest/m0/backport-resume-owed-groups.md +++ /dev/null @@ -1,63 +0,0 @@ -# [S] Release resumes owed groups - -## Goal - -A `release` relay's resumed read gets every owed group or a drop for it, -never a hang. Exposure is reproduced; the source fix and a separately -authorized patch release remain. - -## Plan - -Decided 2026-11-03 (maintainer, /quest-plan of the follow-ups from -[moq-dev/moq#7708](https://github.com/moq-dev/moq/pull/7708)): milestone m0, -beside the other backports. The main fix has merged. - -The exposure investigation is recorded in -[moq-dev/moq#7815](https://github.com/moq-dev/moq/pull/7815). -At `release` commit `da09919e75f285ed1fe0b45b35d34022ffbc14f2` -(moq-net 0.3.20), adapted paused-time session probes show: - -- Cache group 0, leave until the publisher reports unused, publish groups - 1 through 3, then resume at group 0 with a budget covering the history. - Every IETF draft from 14 through 22 returns group 3 instead of group 0. - Applying release's local cursor cap `set_groups(0..1)` leaves the read - waiting for group 0. All seven lite versions, including lite-07-wip, - retain group 0 in both cases. -- The separate bounded `1..3` probe behind a fetched copy delivers groups - 1 and 2 on drafts 14 through 22. It does not establish that an initially - bounded IETF request is accepted: release opens the subscription while - querying metadata, before the reader's preferences reach it. -- Release's movable local cap intentionally parks at its boundary. Main's - bounded EOF assertion is not applicable to this API. - -The root differs from main's shared `live_floor`: -`origin::run_front` queries a source before splicing the reader's demand. -The IETF subscriber snapshots the empty demand as a live join and records -SUBSCRIBE_OK's Largest as its start. `resume::Subscriber::poll_activate` -then discards the whole warm segment when that start exceeds its edge, -without judging the resumed reader's budget. - -The remaining work is a hand-port that preserves the published API and -finite-budget behavior. Do not merely remove the warm-cache gate: main's -fix distinguishes unlimited from finite history, while release exposes a -concrete `max_age: Duration` and has no unlimited option. Forwarding the -actual demand before IETF setup is the recommended direction; it needs -coverage of metadata-only lookups, fetched copies, and concurrent readers. -Keep the original cache-loss and bounded-resume matrices for drafts 14 to -22 and every lite version. - -Keep this separate from [the joined-head FETCH backport](/quest/m0/backport-fill-head-race.md) -and [moq-dev/moq#7680](https://github.com/moq-dev/moq/pull/7680). -Its queued-miss grace correction is a different path. - -After a reviewed source backport lands, use the normal patch-release -process only with explicit release authorization. Do not bump versions, -tag, or publish as part of the implementation attempt. Verify the -published source includes the fix before deleting this quest. - -Public API: preserve the published surface. Wire: no format change expected. - -## Waiting - -- Maintainer: choose whether to continue with a demand-setup hand-port (recommended), preserving finite-budget cache behavior, or explicitly permit a change to that behavior. Removing the warm-cache gate alone changes which stale cache finite readers receive; main's unlimited-only exception cannot be copied into release's older API. -- Maintainer: authorize the separate `moq-net` patch release after the source fix is reviewed and landed; retain this requirement until the published source is verified. Release cuts and version changes require their own authorization. diff --git a/quest/m2/release/cut.md b/quest/m2/release/cut.md index dce91cba871..8da4e5a0fa4 100644 --- a/quest/m2/release/cut.md +++ b/quest/m2/release/cut.md @@ -33,6 +33,11 @@ MKV buffering fix to release, which stays exposed until then. Backport it to (maintainer, 2026-11-03). A blanket size cap would refuse supported inputs and miss the decoder's captured children, as [moq-dev/moq#7718](https://github.com/moq-dev/moq/pull/7718) measured. +It likewise ships [moq-dev/moq#7708](https://github.com/moq-dev/moq/pull/7708)'s +resumed-history fix: release (moq-net 0.3.20) loses a resumed IETF reader's +cached groups on drafts 14 to 22. ✅ No backport (maintainer, 2026-11-04; +[moq-dev/moq#7815](https://github.com/moq-dev/moq/pull/7815)), so check +both fixes reach release. Recurs once per major release (decided 2026-10-19). The PR that finishes the cut shifts the milestones up, as the root [questline](/quest/README.md)