From 0ef5cdfa78ed88eda3f3b60e03a7f715c2c77cba Mon Sep 17 00:00:00 2001 From: Thomas Gravgaard Hansen Date: Fri, 30 Oct 2026 13:50:37 +0100 Subject: [PATCH 1/5] quest: pin the restart a follower gets when the serving epochless route leaves Co-Authored-By: Claude Opus 5.5 --- quest/m1/README.md | 1 + quest/m1/epochless-leave-restart.md | 48 +++++++++++++++++++++++++++++ 2 files changed, 49 insertions(+) create mode 100644 quest/m1/epochless-leave-restart.md diff --git a/quest/m1/README.md b/quest/m1/README.md index fe863117b99..74bb17c8028 100644 --- a/quest/m1/README.md +++ b/quest/m1/README.md @@ -182,6 +182,7 @@ removed). - [HLS and DASH segment conformance](/quest/m1/hls-conformance.md) - nightly ffprobe checks that advertised timing, sync flags, and bytes match the media served - [Unbounded groups](/quest/m1/unbounded-groups/README.md) - groups and frames lose their size caps; a group keeps a 32 MB tail, and readers behind it replay through paced flights or refresh - [Next Object on a cold Rust copy](/quest/m1/next-object-cold-copy.md) - a join after an upstream group report starts at the first frame received, never skipping +- [A follower is told when the serving epochless route leaves](/quest/m1/epochless-leave-restart.md) - wire-level tests pin the restart a follower gets when the serving epochless route is withdrawn or dies while another remains - [Lite track end at session death](/quest/m1/lite-track-end.md) - Rust and JS abort an ended lite track whose tail never settled, and agree on what declares the end - [Legacy lite close drain](/quest/m1/lite-legacy-close-drain.md) - a lite-01..06 graceful close never cuts a final group the subscriber has not read - [moqsrc max-delay](/quest/m1/gst-max-delay.md) - GStreamer's `moqsrc` takes a `max-delay` property instead of a hard-coded 1 s budget diff --git a/quest/m1/epochless-leave-restart.md b/quest/m1/epochless-leave-restart.md new file mode 100644 index 00000000000..73142d7de11 --- /dev/null +++ b/quest/m1/epochless-leave-restart.md @@ -0,0 +1,48 @@ +# [S] A follower is told when the serving epochless route leaves + +## Goal + +When the route serving an epochless broadcast is withdrawn, or its session +dies, while another route for the path remains, a follower downstream of the +relay is announced a restart: an end then a start on lite-06 and older, and +`ANNOUNCE_RESTART` on lite-07-wip. A re-request then resumes on the survivor. +Integration tests in the merge gate pin this over the wire on every +version that can carry an epochless route. + +## Plan + +[Identity and recovery](/doc/concept/identity.md) promises that without an +epoch, any other route winning the path is a restart, and players restart only +on the announcement, never on a request error. If a follower is not told, a +player's track subscriptions end while the path stays announced, and playback +freezes until something else changes the announcement. The release line +before [moq-dev/moq#5087](https://github.com/moq-dev/moq/pull/5087) behaves +this way: two epochless publishers on one path, and the serving one stops. +Main implements the restart in the origin's announce cursor (the "another +instance" case), but only +`an_equal_cost_peer_replacement_waits_for_the_old_route` asserts it, at the +model level with no hold and no wire. + +Decided in quest-plan (2026-10-30): + +- Regression coverage on main, not a fix or a release backport. Reading the code + shows main already behaves correctly. If a test shows otherwise, fix it in + this PR, at the relay's announcement, never by splicing the survivor into + the front or by having players retry on track errors. +- Rust assertion only. `route_dies_without_an_epoch` in + `rs/moq-net/tests/integration/route_change.rs` already covers the + `Disconnect` and `Unannounce` triggers over lite-04 to 06 under mock time. + It checks that the subscription ends and a re-request reaches the survivor. + Extend it, or a sibling sharing its topology, so a follower on the + subscriber's origin also observes the restart for both triggers. + Cover lite-07-wip with an epochless route too, if the topology can produce + one. Player-level resume is already covered on a graceful leave by + `just test native-play` (lite-06 publishers) and + `just test media --media-cases replacement`. A crash differs only in when + the relay notices, which the `Disconnect` trigger already exercises, so no + new browser or native harness case. +- Assert the event kind the version can carry, not just "something arrived". + The `best_route_wins_and_fails_over` helper accepts Update or Start, which + would hide a regression to an in-place update. + +Public API: none. Wire: none. From 5bf04f06d8d5c1ecaaa29a4ffb16eff8634040a2 Mon Sep 17 00:00:00 2001 From: Thomas Gravgaard Hansen Date: Fri, 30 Oct 2026 13:57:25 +0100 Subject: [PATCH 2/5] quest: observe the restart past a wire hop Co-Authored-By: Claude Opus 5.5 --- quest/m1/epochless-leave-restart.md | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/quest/m1/epochless-leave-restart.md b/quest/m1/epochless-leave-restart.md index 73142d7de11..397dd1ebb11 100644 --- a/quest/m1/epochless-leave-restart.md +++ b/quest/m1/epochless-leave-restart.md @@ -33,8 +33,11 @@ Decided in quest-plan (2026-10-30): `rs/moq-net/tests/integration/route_change.rs` already covers the `Disconnect` and `Unannounce` triggers over lite-04 to 06 under mock time. It checks that the subscription ends and a re-request reaches the survivor. - Extend it, or a sibling sharing its topology, so a follower on the - subscriber's origin also observes the restart for both triggers. + Extend it, or a sibling sharing its topology, so a follower one wire hop + past the relay observes the restart for both triggers. Watching + `announced()` on the relay's own origin only sees its local model event, + which `better_route_restarts` already covers; the follower must see what the + relay sends over the wire. Cover lite-07-wip with an epochless route too, if the topology can produce one. Player-level resume is already covered on a graceful leave by `just test native-play` (lite-06 publishers) and From ad7bf259e33db1028bc0102254390ad43daea648 Mon Sep 17 00:00:00 2001 From: Thomas Gravgaard Hansen Date: Fri, 30 Oct 2026 14:00:07 +0100 Subject: [PATCH 3/5] quest: require epochless lite-07 and moq-transport cases Co-Authored-By: Claude Opus 5.5 --- quest/m1/epochless-leave-restart.md | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) diff --git a/quest/m1/epochless-leave-restart.md b/quest/m1/epochless-leave-restart.md index 397dd1ebb11..0d109633cef 100644 --- a/quest/m1/epochless-leave-restart.md +++ b/quest/m1/epochless-leave-restart.md @@ -4,10 +4,11 @@ When the route serving an epochless broadcast is withdrawn, or its session dies, while another route for the path remains, a follower downstream of the -relay is announced a restart: an end then a start on lite-06 and older, and -`ANNOUNCE_RESTART` on lite-07-wip. A re-request then resumes on the survivor. -Integration tests in the merge gate pin this over the wire on every -version that can carry an epochless route. +relay is announced the restart its version carries, such as an end then a +start on lite-06 and older and `ANNOUNCE_RESTART` on lite-07-wip. A re-request +then resumes on the survivor. Integration tests in the merge gate pin this +over the wire, with an epochless route, on lite-04 to 06, lite-07-wip, and +moq-transport 18 and 22. ## Plan @@ -32,15 +33,20 @@ Decided in quest-plan (2026-10-30): - Rust assertion only. `route_dies_without_an_epoch` in `rs/moq-net/tests/integration/route_change.rs` already covers the `Disconnect` and `Unannounce` triggers over lite-04 to 06 under mock time. - It checks that the subscription ends and a re-request reaches the survivor. + Its `Topology::new` mints an epoch, which only those versions drop, so it + is epochless there alone. It checks that the subscription ends and a + re-request reaches the survivor. Extend it, or a sibling sharing its topology, so a follower one wire hop past the relay observes the restart for both triggers. Watching `announced()` on the relay's own origin only sees its local model event, which `better_route_restarts` already covers; the follower must see what the relay sends over the wire. - Cover lite-07-wip with an epochless route too, if the topology can produce - one. Player-level resume is already covered on a graceful leave by - `just test native-play` (lite-06 publishers) and + Lite-07-wip and moq-transport 18 and 22 are required too, through a + topology that announces without an epoch, as `replacement.rs` already does + for its newcomer cases. Moq-transport routes all become epochless with + [moq-transport carries no epoch](/quest/m1/ietf-epoch-cut.md), so this + path must hold there whichever lands first. Player-level resume is already + covered on a graceful leave by `just test native-play` (lite-06 publishers) and `just test media --media-cases replacement`. A crash differs only in when the relay notices, which the `Disconnect` trigger already exercises, so no new browser or native harness case. From 92316e7aeca2e634cba375011b29bef090e155f3 Mon Sep 17 00:00:00 2001 From: quest fixer Date: Sun, 1 Nov 2026 07:59:08 -0800 Subject: [PATCH 4/5] docs: follow completed epoch quest link on base --- quest/m1/epochless-leave-restart.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/quest/m1/epochless-leave-restart.md b/quest/m1/epochless-leave-restart.md index 0d109633cef..ec6cd1ec429 100644 --- a/quest/m1/epochless-leave-restart.md +++ b/quest/m1/epochless-leave-restart.md @@ -44,7 +44,7 @@ Decided in quest-plan (2026-10-30): Lite-07-wip and moq-transport 18 and 22 are required too, through a topology that announces without an epoch, as `replacement.rs` already does for its newcomer cases. Moq-transport routes all become epochless with - [moq-transport carries no epoch](/quest/m1/ietf-epoch-cut.md), so this + [moq-transport carries no epoch](https://github.com/moq-dev/moq/pull/7239), so this path must hold there whichever lands first. Player-level resume is already covered on a graceful leave by `just test native-play` (lite-06 publishers) and `just test media --media-cases replacement`. A crash differs only in when From 7bd75d704bd2ab4f6a548e5153a2ab6b4cf591d9 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 1 Nov 2026 15:31:53 -0800 Subject: [PATCH 5/5] quest: align epochless-leave-restart with main's epochless coverage moq-dev/moq#7239 merged, so route_dies_without_an_epoch's pinned tests already run moq-transport 18 and 22 epochless through the shared topology. The Plan's "whichever lands first" and "lite-04 to 06 alone" lines were stale. Only lite-07-wip needs the epochless topology. Co-Authored-By: Claude Opus 5.5 --- quest/m1/epochless-leave-restart.md | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-) diff --git a/quest/m1/epochless-leave-restart.md b/quest/m1/epochless-leave-restart.md index ec6cd1ec429..8a6011bd6b7 100644 --- a/quest/m1/epochless-leave-restart.md +++ b/quest/m1/epochless-leave-restart.md @@ -32,20 +32,21 @@ Decided in quest-plan (2026-10-30): the front or by having players retry on track errors. - Rust assertion only. `route_dies_without_an_epoch` in `rs/moq-net/tests/integration/route_change.rs` already covers the - `Disconnect` and `Unannounce` triggers over lite-04 to 06 under mock time. - Its `Topology::new` mints an epoch, which only those versions drop, so it - is epochless there alone. It checks that the subscription ends and a - re-request reaches the survivor. + `Disconnect` and `Unannounce` triggers over moq-transport 18 and 22 and + lite-04 to 06 under mock time (`pinned_tests!`). Its `Topology::new` mints + an epoch, which those versions drop (moq-transport since + [moq-dev/moq#7239](https://github.com/moq-dev/moq/pull/7239)), so it is + epochless there but not on lite-07-wip. It checks that the subscription + ends and a re-request reaches the survivor. Extend it, or a sibling sharing its topology, so a follower one wire hop past the relay observes the restart for both triggers. Watching `announced()` on the relay's own origin only sees its local model event, which `better_route_restarts` already covers; the follower must see what the relay sends over the wire. - Lite-07-wip and moq-transport 18 and 22 are required too, through a - topology that announces without an epoch, as `replacement.rs` already does - for its newcomer cases. Moq-transport routes all become epochless with - [moq-transport carries no epoch](https://github.com/moq-dev/moq/pull/7239), so this - path must hold there whichever lands first. Player-level resume is already + Moq-transport 18 and 22 get the follower assertion from the shared + topology. Lite-07-wip is required too, through a topology that announces + without an epoch, as `replacement.rs`'s `epochless_lite07` already does. + Player-level resume is already covered on a graceful leave by `just test native-play` (lite-06 publishers) and `just test media --media-cases replacement`. A crash differs only in when the relay notices, which the `Disconnect` trigger already exercises, so no