Skip to content

ietf: answer a FETCH that spans more than one group #8577

Open

Description

@michaelstuurman

A Rust publisher refuses a Standalone FETCH once it spans more than one group. doc/concept/standard.md lists this as "History is served one group at a time". Measured against our relay (moq-relay d66c892, draft 18, a live track with 1 s groups); the refusal is still in rs/moq-net/src/ietf/publisher.rs on main (bec81e4):

  • one cached group: FETCH_OK, all 50 objects;
  • two groups: REQUEST_ERROR 3, "FETCH spans more than one group".

Why we ask

We run a relay built on moq-relay as an origin, and want a moqx relay as edge in front of it. From its cache code, moqx can forward a client's range upstream on a miss, or coalesce the missing groups into one range. Time-shifted playback through such an edge would then get NOT_SUPPORTED from us. We have not run this yet.

moq-net's own subscriber also sends one range per run of missing groups, so two moq-net relays in a chain meet the same refusal.

draft-ietf-moq-transport-18 §10.12.1 defines a Standalone Fetch by a Start and End Location, so a multi-group answer is the general case.

What it would take

  • The range itself: walk the groups from start to end through the existing per-group path (fetch_group), in order, on the one fetch stream. Fetch the next group only when the stream has room, and stop when the requester cancels.
  • Missing versus unknown: §10.12.3 treats a gap in a finished stream as "does not exist", and tells a relay not to guess when it does not know: confirm upstream, or reset with UNKNOWN_OBJECT_STATUS. A range needs each fill to say which it was: a group that never existed or has expired can be skipped, one the source cannot reach right now cannot. In the same measurement a single group a few seconds old came back as REQUEST_ERROR 3 "unsupported", because the group had left the relay's cache and the publisher behind it (moqxr) declines a Standalone FETCH. A range would have to decide what that means for the groups around it.
  • End Location: a range past the largest object is cut short in FETCH_OK, and a start past it is INVALID_RANGE.

Limits (groups or bytes per FETCH, concurrent fills) and the cache budget for old groups could stay with the relay.

Question

Would you accept this in moq-net? We can write it, but would like your view on the shape first, since it reverses a documented choice. The second point changes what a track::Requests handler can answer, so that is the part we would most like your opinion on.

(Written with Claude Fable 5.2)

Activity

  1. coderabbitai commented on Nov 12, 2026

    @coderabbitai
    Contributor
    🔗 Related PRs

    #6724 - feat(net): fetch missing runs as one FETCH, and serve multi-group FETCH [merged]


    📝 Issue Planner

    Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.

    • Create Plan

    🧪 Issue enrichment is currently in open beta.

    You can configure auto-planning by selecting labels in the issue_enrichment configuration.

    To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

    issue_enrichment:
      auto_enrich:
        enabled: false

    💬 Have feedback or questions? Drop into our discord!

  2. added
    questBeing tracked/planned in a quest. See `quest/`
    on Nov 12, 2026
  3. kixelated commented on Nov 12, 2026

    @kixelated
    Collaborator

    Sorry there's just so many issues with FETCH that we had to rip it out. I think maaaybe we could do it again but we would heavily rely on the previous group ID parameter and probably make it required. The fact that MoQ can intentionally cause gaps is frustrating from a CDN point of view because our SUBSCRIBE cache is useless for FETCH and we need to go upstream for every fucking tiny hole.

  4. kixelated commented on Nov 12, 2026

    @kixelated
    Collaborator

    Thanks for the detailed write-up. We'll accept a narrow version, tracked in quest/m4/ietf-range-fetch.md (planned in #8582).

    Shape: a Rust publisher answers a multi-group Standalone FETCH (or draft 20+ fill) only when every group in the range is either cached or proven never produced: produced locally, marked by a moq-lite SKIP or a budget-independent drop, or covered by a received Prior Group ID Gap. Proven holes go out as End of Non-Existent Range (0x8C). A range with any unknown group is refused whole, as today. We won't reset with UNKNOWN_OBJECT_STATUS partway, and the relay won't confirm upstream group by group. The quest waits on our Prior Group ID Gap ingest (quest/m2/time-addressing/ietf.md). This narrows #7204 and our moq-transport scope freeze, which is recorded in the quest.

    On your track::Requests question: the handler shape doesn't change. A Fill already separates gone (drop_groups) from unknown (refuse_groups). One limit to expect with a moqx edge: a group dropped or expired upstream over moq-transport has no marker, so ranges over such gaps stay refused.

    Anyone who wants to donate tokens (you offered to write it) can run /quest-start quest/m4/ietf-range-fetch. This issue closes when that PR lands.

    (Written by Opus 5.5)

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

    questBeing tracked/planned in a quest. See `quest/`

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions