From 208cb21ca78e4d9fe986886914ae49a7cd974b38 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Tue, 3 Nov 2026 20:47:11 -0800 Subject: [PATCH] quest(m1): plan the TS clock-revision burst investigation While #7729 was tested, the two-exporter TS fixture dropped a queued TDT/TOT revision with an unlimited SI max-age. Unconfirmed and maybe fixture-only; the quest proves a burst reaches its stamped slots or fixes the fixture. Co-Authored-By: Claude Opus 5.5 --- quest/m1/ts-reliability/README.md | 1 + .../m1/ts-reliability/clock-revision-burst.md | 57 +++++++++++++++++++ 2 files changed, 58 insertions(+) create mode 100644 quest/m1/ts-reliability/clock-revision-burst.md diff --git a/quest/m1/ts-reliability/README.md b/quest/m1/ts-reliability/README.md index ed84e97e7bb..e73c15c5114 100644 --- a/quest/m1/ts-reliability/README.md +++ b/quest/m1/ts-reliability/README.md @@ -108,6 +108,7 @@ catalog linger (#5305). - [A skipped group flags the next PCR](/quest/m1/ts-reliability/skip-flag.md) - a within-instance skip reaches the wire as a flagged PCR discontinuity - [A republish announces as soon as its tracks are known](/quest/m1/ts-reliability/republish-announce.md) - a stitched export keeps the new instance's first group on a short-GOP feed +- [A TS export never loses a queued clock revision](/quest/m1/ts-reliability/clock-revision-burst.md) - a burst of TDT/TOT revisions all reach their stamped slots, or the drop is shown to be the two-exporter fixture's ## Related diff --git a/quest/m1/ts-reliability/clock-revision-burst.md b/quest/m1/ts-reliability/clock-revision-burst.md new file mode 100644 index 00000000000..322e3556d77 --- /dev/null +++ b/quest/m1/ts-reliability/clock-revision-burst.md @@ -0,0 +1,57 @@ +# [S] A TS export never loses a queued clock revision + +## Goal + +The TS export lays every TDT/TOT revision it has queued on the subscription +at its stamped slot on the media grid, however many arrive before media +reaches the first one, and whatever the SI track's max-age. Done when either +a test proves that a burst of queued revisions all reach their stamped slots, +or the loss is shown to be a fixture artifact and the fixture is fixed. + +## Plan + +Planned 2026-11-04 (maintainer, `/quest-plan`), from testing +[moq-dev/moq#7729](https://github.com/moq-dev/moq/pull/7729). While that PR +was tested, the existing two-exporter fixture in moq-mux's TS export tests +dropped a queued TDT/TOT revision even though the SI track's max-age was +unlimited. So the exporter may lose a burst of queued clock revisions for a +reason that has nothing to do with max-age. **This is unconfirmed, and may be +fixture-only.** #7729's body and review threads record no repro for it. +They cover only the max-age regression it fixed: a zero-age SI track skipped +a superseded revision at once, tested by +`a_superseded_si_revision_stays_readable_within_the_export_budget`. + +Background: + +- [moq-dev/moq#7779](https://github.com/moq-dev/moq/pull/7779) placed clock + revisions on the stamped media grid. The export reads a clock table in + order and holds one revision until media reaches its stamped slot, while + later revisions wait on the subscription. +- [moq-dev/moq#7729](https://github.com/moq-dev/moq/pull/7729) gives the TS + importer's SI tracks their catalog's media max-age, because the export's + SI budget is capped by that age. +- The fixture is `export_twice` in + `rs/moq-mux/src/container/ts/export_test.rs`, used by + `late_join_matches_a_running_exporter`, its `_without_video` variant, and + `a_late_clock_revision_remains_visible_between_exports`. Its clock tracks + are created with `timed()`, which sets no max-age, so they are unlimited + on main. Clock revisions arrive between the two exporters' drains, before + their stamped media position. + +Steps: + +- Reproduce it first. Write a test that queues a burst of several TDT and TOT + revisions, each stamped ahead of the media, on an unlimited SI track. Check + that every revision appears at its slot, in order, in each exporter's + output. +- If a revision is lost, find where: the single held revision being replaced + rather than queued, a later revision read past while one is held, or the + delay budget expiring on a revision behind the held one. Fix the exporter, + and keep the burst test. +- If the exporter is correct, find what the fixture does wrong (for example + the drain order, or a revision written after its slot already passed), fix + `export_twice`, and record the cause here before deleting the quest. +- Don't change snapshot tables: they restate the latest value and may skip + superseded ones. + +Public API: none. Wire: the TS output gains any revision that was dropped.