Proposals for MoQ Transport draft-23
MOQOM implements Media over QUIC Transport (draft-ietf-moq-transport) twice, in a Rust relay
and in a Swift client. These are changes we intend to propose for the next revision, based on
building both.
Status: the first proposal, a PUBLISH_DONE code for an expired token, was filed on 7 October 2026 as issue #1969 and pull request #1970. The rest aren’t submitted yet. Nothing here is in any draft until the working group adopts it.
How this works
Section titled “How this works”draft-ietf-moq-transport-23 is written by the draft’s editors and published by the IETF MoQ
working group. Anyone can contribute to it in three ways:
- File an issue or pull request on the working group repository, github.com/moq-wg/moq-transport.
- Raise it on the MoQ mailing list or at a working group meeting.
- Post an individual Internet-Draft for features that belong in an extension rather than in the core protocol.
The working group decides what goes into -23 or later. We file each change below as its own issue, so it can be accepted or rejected separately.
1. A PUBLISH_DONE code for an expired token
Section titled “1. A PUBLISH_DONE code for an expired token”Problem. In draft-22, EXPIRED_AUTH_TOKEN exists as a session termination code, a request
error code and a stream reset code, but not as a Publish Done code (§12.4). When the token that
authorised an established subscription expires, a publisher can only end the subscription with
UNAUTHORIZED (0x1) or INTERNAL_ERROR (0x0). The subscriber can’t tell whether it should
refresh its token or give up.
Proposal.
- Add
EXPIRED_AUTH_TOKENto the PUBLISH_DONE Codes registry as 0x7. That value is unused there, and it’s the valueEXPIRED_AUTH_TOKENalready has as a Stream Reset code. - In PUBLISH_DONE (§9.9): a publisher SHOULD use this code when it ends a subscription because the authorizing token expired. The subscriber can subscribe again with a refreshed token, or avoid the interruption by sending a REQUEST_UPDATE with a new AUTHORIZATION_TOKEN before expiry.
- The code refers to the subscriber’s own token. A relay whose upstream credentials lapse doesn’t pass it on to downstream subscribers, who can’t fix that.
This gives subscriptions the recovery path that PUBLISH_NAMESPACE already has (§4.1).
Size. One registry entry, one code definition and a short paragraph. It goes to the working group as an issue and a pull request against the core draft, kept separate from proposal 4 so each can be accepted on its own.
2. Distinguishing Property types from Message Parameter types
Section titled “2. Distinguishing Property types from Message Parameter types”Problem. Track Properties and Message Parameters are separate registries, and both use small
numbers. Some values look related but aren’t: DEFAULT PUBLISHER PRIORITY is Property Type
0x0E (§10.4), while SUBSCRIBER_PRIORITY is Message Parameter 0x20 (§9.20.7). One of our two
implementations sent the property value as the parameter. Under draft-22 the receiver closes the
session with PROTOCOL_VIOLATION (§9.20). That fails loudly, but the error doesn’t point at the
cause, and a value that a later draft defines as a parameter would be misread instead.
Proposal. This is an editorial change:
- In §9.20.7 and §10.4, state which registry each priority value belongs to.
- Add a short note that Property types and Message Parameter types are separate namespaces, and that the same number can mean different things in each.
Size. Editorial.
3. Shared test vectors
Section titled “3. Shared test vectors”Problem. Implementations find wire-format bugs late, usually at interop events. Draft-22 changes the encoding of several messages (LOCATION_FILTER, FILL_PARAMETERS and the Type Flags bitfields), which makes this more likely.
Proposal. We contribute byte-exact encode and decode vectors for every draft-22 control message and parameter to the working group’s interop effort. Our two codecs were written separately and are checked against each other.
Size. No change to the draft text. This would live in the interop repository.
4. Loop protection between relays (separate extension)
Section titled “4. Loop protection between relays (separate extension)”Problem. §4.1 says PUBLISH_NAMESPACE “does not protect against loops”, and draft-22 places relay-to-relay interconnection out of scope (§7). Relays that forward requests to a configured upstream, as content delivery networks pull from an origin, have no in-protocol way to stop a request from circling a misconfigured cycle.
The working group has said it sees loop prevention as an extension
(moq-wg/moq-transport#1798). One
extension already exists: draft-lcurley-moq-cluster. It prevents loops in namespace
advertisements by carrying the path each advertisement took. It doesn’t cover requests that a
relay forwards without an advertisement, which is the gap this proposal fills.
Proposal. An individual Internet-Draft, draft-elliottwave-moq-hop-limit
(summary), designed to work alongside the cluster extension:
- a
HOP_LIMITMessage Parameter on SUBSCRIBE, FETCH, TRACK_STATUS, SUBSCRIBE_NAMESPACE and SUBSCRIBE_TRACKS, which each relay clamps to its own maximum and decrements when it forwards; - a
HOP_LIMIT_EXCEEDEDREQUEST_ERROR code for when the limit reaches zero; - a rule for aggregated requests, so two relays that each attach a request to the other’s pending one don’t wait on each other until a timeout;
- negotiation by a zero-length Setup Option, as §6.3.2 describes.
Status. Written and builds with the IETF tools. Not yet posted.
5. Editorial fixes found while implementing
Section titled “5. Editorial fixes found while implementing”- RENDEZVOUS_TIMEOUT (§9.20.6) gives a unit (milliseconds) but not an encoding. We assume a varint.
- §3.4 refers to a “zero-length” LOCATION_FILTER inside FILL_PARAMETERS. LOCATION_FILTER has no length field of its own, so this probably means Location Filter Type 0x00. The text should say so.
Size. Editorial.
6. Implementation report
Section titled “6. Implementation report”Once both codecs speak draft-22, we’ll send the working group an implementation report. It will say which features we implemented and which we didn’t, what was unclear, and our interop results against other implementations.
Not proposals
Section titled “Not proposals”These are things we’re doing to match the drafts, not changes we’re asking for:
- Moving from draft-19 to draft-22. This covers fill fetch streams, LOCATION_FILTER, PUBLISH_STATE_NOTIFY and INCLUDE_PROPERTIES.
- Aligning our media metadata with the MoQ media format drafts (LOC) rather than defining our own.
- Encoding tokens as Common Access Tokens, as C4M specifies (
draft-ietf-moq-c4m).