Skip to content

fix(moq-net): a gapped group is dropped with debug! and returns Ok() #5371

Closed

Description

@kkroo

Summary

In rs/moq-net/src/ietf/subscriber.rs, a group whose object IDs are non-contiguous is dropped silently: the caller is told the stream was handled successfully.

Detail

The catch-all arm logs at debug level and aborts the group producer, but does not return:

Err(err) => {
    tracing::debug!(%err, group = %producer.sequence, "group error");
    let _ = producer.abort(err);
}

Control then falls through to Ok(()).

Its two sibling arms both propagate instead:

  • MessageTooLarge returns Err(err), with a comment saying the refusal is returned so the dispatcher sends STOP_SENDING.
  • MalformedTrack logs at warn level and returns Err(Error::MalformedTrack).

So the non-contiguous-object-ID case is quieter than both: debug rather than warn, and a success return rather than a refusal.

Why it matters

A relay in this state forwards a truncated track with no signal at any normal deployment log level. Measured on an earlier revision: three objects in, zero out, Ok(()), session still open.

Suggested fix

Either raise this log to warn, or return the refusal as the two sibling arms do. We have no preference and are happy to send a PR for whichever behavior you prefer; this is reported rather than patched precisely because the choice belongs upstream.

Not related to #4926's declined proposal

Issue #4926 asked for an object-level API (opaque properties, non-contiguous object IDs, a subgroup layer), which was declined. This report is about how the existing refusal is reported and stands whether or not non-contiguous IDs are ever supported.

Activity

  1. kixelated commented on Oct 13, 2026

    @kixelated
    Collaborator

    Thanks for the report. The gap case is already fixed on main: #5233 made a non-contiguous object ID return Unsupported, which takes the refusing arm and stops the stream, so it no longer reaches the catch-all.

    You were right that the catch-all is too quiet, though. A malformed extension block, a duplicate stream for a live group, and FrameTooLarge still log at debug and return Ok(()). We're also making the intentional subgroup/datagram limits fail loud: a subgroup other than 0, or a datagram past object 0, will abort the track instead of forwarding a truncated group. That work is planned as quest/m1/ietf-ingest-refusal.md in #5380. Once that merges, anyone who wants to donate tokens can run /quest-start quest/m1/ietf-ingest-refusal.

    (Written by Claude Opus 5.5)

  2. added a commit that references this issue on Oct 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions