Skip to content

sync: upstream moxygen 2401c3bb77d7 - #629

Merged
omoq-sync-bot[bot] merged 10 commits into
mainfrom
sync/2401c3bb77d7
Nov 17, 2026
Merged

omoq-sync-bot[bot] merged 10 commits into
mainfrom
sync/2401c3bb77d7

Conversation

@omoq-sync-bot

@omoq-sync-bot omoq-sync-bot Bot commented Nov 17, 2026 •

Copy link
Copy Markdown

Automated upstream sync.

  • Upstream commit: 2401c3bb77d7
  • Upstream status: success (green)
  • Merge type: git merge (sync/2401c3bb77d7 → main)

The verify workflow will validate the standalone build. On success, this PR auto-merges.
Devs can push conflict-resolution commits to sync/2401c3bb77d7 if needed.

Created by omoq-upstream-sync workflow.


This change is Reviewable

kvtsoy and others added 10 commits November 16, 2026 11:01
Summary: The switch checker could only watch proxies that accept unauthenticated SUBSCRIBE and FETCH. A proxy that checks viewer tokens refused every viewer. With --auth_token, every SUBSCRIBE and FETCH the checker sends carries the token as its AUTHORIZATION_TOKEN parameter.

Reviewed By: sharmafb

Differential Revision: D131079120

fbshipit-source-id: 17f765c6fc5a8cd308256864c5c103c67fab4b4b
Summary: The switch checker offered standard ALPNs such as moqt-16, and the switch origin accepted only those. moxygen's servers and relays know each draft by its Meta ALPN, such as moqt-16-meta-00. So a checker couldn't reach them over raw QUIC, and a relay couldn't reach the origin. The checker now offers the Meta ALPNs, and the origin accepts both kinds.

Reviewed By: sharmafb

Differential Revision: D131079121

fbshipit-source-id: b8db7e82fe458036a4a122ee52a3e50f154a6635
Summary: A test that injects a fault while the switch checker watches has to know that every viewer has subscribed first, and the checker shows nothing of it before its report. It now writes a line to stderr once its viewers have subscribed, which a test can wait for.

Reviewed By: sharmafb

Differential Revision: D131079123

fbshipit-source-id: a13dd9756307c39791c125d9caf17ce73a5cbcb2
Summary: A sampled MoQ session can retain an unbounded log before viewer authorization. Limit the media-event prefix to 1024 events / 192 KiB, and keep recent control messages in a separate 256-event / 64 KiB ring so ordinary media traffic cannot exhaust end-of-session diagnostics. Count dropped events and release records after emission, also fixing duplicate emission when a peer-closed session is destroyed. File logging retains its existing behavior.

Reviewed By: ShenYi

Differential Revision: D131818954

fbshipit-source-id: d5557c16510cdfc02b659c85f37c70b28d7a95dd
Summary:
A cached replay can outlive the upstream subscription that authorized it. Add an optional `cancelReplayOnUnauthorized` callback, evaluated when upstream sends `UNAUTHORIZED`, to cancel replay and invalidate the track cache without disabling live stream recorders. Normal publication ends retain their existing streaming and history behavior.

___

landed-with-radar-review

Differential Revision: D131820375

fbshipit-source-id: 484835786f8d8a051c5b01345b6e8aa9696e9f21
Summary:
`moq_switch_origin` served WebTransport on `MoQServer`'s default path, which takes the client's `WT_INITIAL_MAX_DATA` as its send window and never applies the client's credit updates. A client that enforces WebTransport data credit therefore gets only the first 64 KiB of stream data on each session. With several tracks sharing one long-lived session, that runs out within a minute, and every track on the session stalls. The origin now serves WebTransport on `HqWtSession`, which applies the client's data and stream credit updates.

___

landed-with-radar-review

Differential Revision: D132354384

fbshipit-source-id: 85c0d8c701279d660fad1357399cd943829004b9
Summary: Count received upstream session GOAWAYs once per proxy registry and request-stream GOAWAYs once per subscription notification. The existing per-track counter still measures how many tracks act on those notifications.

Reviewed By: kvtsoy

Differential Revision: D131381668

fbshipit-source-id: 87afe716b306a5e009093e38a907b1d41c8c6cfe
Summary:
**P1, relay tracing: for design review with the CDN (Traffic Protocols) owners before it lands.** It heads its own stack on master, apart from the EdgeTee and FBMS diffs: those hops and FBMS to player are measurable without it, with relay time inside the player's DL parent hop until this stack lands.

Design: https://artifacts.internalmeta.com/a/969199255910755. A relay needs to see each object once as it arrives from upstream, know when its last byte first goes downstream, and read and rewrite what the upstream says about the track (its SUBSCRIBE_OK track extensions), all without rewriting objects or paying for it on tracks nobody watches. The MoQ LiveTrace relay stages (next diff) are the first user: they learn what to log from track extensions FBMS sets, and replace the parent-source extension with their own, as CDN tiers rewrite DASH's `X-FB-Video-LiveTrace-ParentSource` header.

`MoQProxyOptions::trackObservers` gives each new track an optional `MoQProxyTrackObserver`:
- `onUpstreamExtensions`: the track extensions in the SUBSCRIBE_OK of each upstream the track reads from, its first and each that replaces it after a switch (lost or GOAWAY).
- `downstreamExtensions`: what downstream subscriptions are told, given the first upstream's extensions; unchanged by default. Today every subscriber gets the first upstream's extensions as they are, and that stays so across switches.
- `onObjectData`: every chunk of every upstream object, once, before fan-out, with its group/subgroup/object ids, its extensions on the first chunk, the payload chunk and a completion flag. Covers subgroup objects (whole and streamed), object streams and datagrams; object statuses are not reported.
- `onObjectForwarded`: after the forwarder call that hands an object's last byte downstream, if it wrote it to at least one subscription, with the time of the first write (track clock). At most once per object.

How: `MoQProxyTrack` passes the first upstream's extensions through the observer before giving them to its `MoQForwarder`, and reports each later upstream's when it adopts it. Its recorders report each chunk to a per-track `MoQProxyObjectTap`, and run the forwarder call that completes an object inside `MoQProxyObjectTap::forwardLastByte`. Downstream consumers of an observed track are wrapped in counting filters, which count accepted data writes made during that call. `MoQForwarder` is unchanged.

Zero overhead when unset: a track whose factory returns null (or with no factory) has no tap, its downstream consumers are not wrapped, and the recorders only test a null pointer.

The hook is generic: no Meta-internal dependencies. The new files carry the moxygen Apache header; `proxy/CMakeLists.txt` lists the new source.

Blast radius: `MoQProxy` users (revproxy MoQ relay, `MoQProxyServer`, rt_gateway). Behavior is unchanged unless `trackObservers` is set; nothing sets it in this diff.

Reviewed By: sharmafb

Differential Revision: D131950755

fbshipit-source-id: 6577d9ae13f19e66e36dfab947a8347a3d31376f
…watchman used

Summary:
Nothing builds Watchman with getdeps any more: its GitHub workflows use plain CMake, its Sandcastle job is gone, and the scripts that called `getdeps.py build watchman` were removed earlier in this stack. This does for the watchman manifest what D128476492 did for eden's. It keeps the `[git]` and shipit sections, so `getdeps.py fetch watchman` still makes the GitHub-layout copy to try the open-source build in, and makes the builder `nop`.

Its dependencies, CMake defines and environment section go, and so do the pcre2 and python-setuptools-69 manifests, which only watchman depended on. Watchman's CMake fetches neither: it finds PCRE2 installed, and uses the setuptools of the build's Python.

watchman no longer depends on folly, fizz, fbthrift or wangle in getdeps, so the determinator no longer lists it for changes to them, and `determinator_test.py` drops it from those expectations.

___

landed-with-radar-review

Differential Revision: D132392956

fbshipit-source-id: da88eaf62a6f626164913c66a124a152bba7750b

@gmarzot gmarzot left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Auto-approved by upstream sync workflow.

@coderabbitai

coderabbitai Bot commented Nov 17, 2026

Copy link
Copy Markdown

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

Run configuration

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

@omoq-sync-bot
omoq-sync-bot Bot merged commit af08bf4 into main Nov 17, 2026
3 checks passed
@omoq-sync-bot
omoq-sync-bot Bot deleted the sync/2401c3bb77d7 branch November 17, 2026 13:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants