MOQ

IETF Media over QUIC (MOQ) Working Group Interim Meeting Date: April 27, 2026

Summary

The MOQ working group held a virtual interim meeting to discuss pending issues and pull requests for draft-ietf-moq-transport. Key discussions focused on identifying the start of subgroups, managing control message ordering via request IDs, and refining delivery timeout semantics. The group reached a tentative consensus to remove the required_request_id mechanism for the upcoming Draft 18 and approved a new subgroup-based delivery timeout.

Key Discussion Points

1. Subgroup IDs and Identifying Subgroup Starts

Presentation: MOQT PRs and Issues Discussion:

Ian Swett presented a proposal to constrain Subgroup IDs such that the Subgroup ID must equal the Object ID of the first object in that subgroup. This is intended to allow receivers to identify if they have received the beginning of a subgroup, which is necessary for features like "rewind" or efficient caching.
Mo Zanaty argued that while the problem is real, removing application semantic control over Subgroup IDs is undesirable, particularly for video layering. He suggested using explicit start/end markers in the subgroup header instead.
Victor Vasiliev noted that Subgroup IDs currently affect prioritization; changing them to match Object IDs would impact the scheduling algorithm.
Cullen Jennings and Magnus Westerlund expressed concerns regarding temporal ordering in layered codecs (e.g., SVC), where the encoding or transmission order might not align with a strict Object ID-based Subgroup ID.
Alan Frindell and Suhas Nandakumar concluded that the proposal should be reconsidered, likely moving toward an explicit marker or header extension to signal the start of a subgroup.

2. Required Request ID and Control Stream Ordering

Presentation: Required Request ID Discussion:

Alan Frindell explained that the required_request_id introduced in Draft 17 (to manage ordering across multiple BIDI control streams) creates unbounded state for the receiver.
Martin Duke highlighted that many use cases (like "make-before-break" or "atomic switch") are not actually solved by this mechanism because it only guarantees receipt order, not processing order or data-plane synchronization.
Victor Vasiliev proposed a more generic "wait for X" signal on streams as an alternative, though he acknowledged the reliability and state issues persist.
Mo Zanaty and Ian Swett questioned the necessity of the feature, suggesting that if applications require strict ordering, they should wait for an acknowledgment before issuing the next dependent command.
Suhas Nandakumar voiced support for keeping the mechanism but acknowledged the implementation difficulties.
Colin Jennings suggested exploring "atomic" operations or multi-request streams as an alternative to cross-stream synchronization.

3. Delivery Timeouts

Presentation: MOQT PRs and Issues (Delivery Timeout) Discussion:

Victor Vasiliev proposed splitting delivery timeouts into two types: delivery_timeout_object (current) and delivery_timeout_subgroup.
The subgroup timeout starts when a subgroup FIN is queued. This prevents a single late object from triggering a full stream reset prematurely, which is particularly beneficial for base-layer video stability.
Alan Frindell and Mo Zanaty supported the change, noting it is easy to implement and provides better flexibility for enhancement vs. base layers.

4. Metadata and GraphQL Subscriptions

Discussion:

Alan Frindell discussed difficulties implementing GraphQL over MOQT due to Draft 17's strict parameter negotiation and the 4K limit on track names. Intermediaries often need to inject metadata (e.g., Client IP, VIA headers).
Victor Vasiliev and Cullen Jennings cautioned against allowing arbitrary subscriber-to-publisher metadata flows, as it could break the aggregation and fan-out properties that define MOQ's efficiency.
The group discussed reverting to a TLV-based parameter system with mandatory/optional ranges, but no consensus was reached.

Decisions and Action Items

Decision: Remove required_request_id from draft-ietf-moq-transport for Draft 18. A poll showed a strong preference to remove the complexity for now and revisit in London if concrete use cases remain unsolved.
Decision: Merge the delivery_timeout_subgroup proposal into draft-ietf-moq-transport.
Action Item: Cullen Jennings to document specific use cases (like "make-before-break") that may require control plane synchronization to determine if a future replacement for required_request_id is necessary.
Action Item: Alan Frindell and the editors to ensure "joining fetch" still functions by leveraging existing subscribe_id correlations without relying on required_request_id.

Next Steps

Draft 18 will be prepared with the removal of required_request_id and the addition of subgroup timeouts.
The next virtual interim (May 11, 2026) will focus on Mo Zanaty’s filter proposal.
The working group will continue discussing metadata and subscriber-sent parameters on the mailing list and GitHub.

Automatic IETF Minutes

MOQ