MOQ Virtual Interim 2025-05-21

Bluesheet

1. Martin Duke (Google)
2. Magnus Westerlund (Ericsson)
3. Alan Frindell (Meta)
4. Gwendal Simon (Synamedia)
5. Mathis Engelbart (TUM)
6. Mo Zanaty (Cisco)
7. Cullen Jennings (Cisco)
8. Suhas Nandakumar (Cisco)
9. Victor Vasiliev (Google)
10. Will Law (Akamai)
11. Chris Lemmons (Comcast)
12. Michal Hosna (CDN77)

Agenda

• WG Chair Administrivia
• #335 Unique track token

• Editors Time: PR Review

  • PRs affecting Object Model and API

    • #995 Add a PUBLISH message for publisher initiated
      subscriptions
    • #1015 Make TRACK_STATUS mirror SUBSCRIBE

  • Need Review

    • #825 Add FETCH_ERROR for Unknown Status in Range
    • #937 Clarify how Latest Object works when the Largest Object
      is incomplete
    • #938 Explain malformed tracks and what to do with them
    • #949 Fetch Object serialization optimization
    • #1008 Clarify FTN construction
    • #1011 Allow End of Group to be signaled in a non-zero length
      object
    • #1012 Make expires a parameter
    • #1013 Allow AUTHORIZATION TOKEN parameter in CLIENT_SETUP
      and other Auth fixes
    • #1016 New Varint Encoding
    • #1017 Relays MUST have upstream subscription to send
      SUBSCRIBE_OK
    • #1025 add marker for end to end header authentication

Minutes

Note taker: Mathis Engelbart

WG Chair Administrivia

Virtual interop before IETF 123

• Date will be coordinated offline with implementers
• Interop target is draft 11
• Editor's may still publish new drafts which will not (yet) be
  interop target

Unique Track Token (#335)

• Victor presents slides

• Clarifications:

  • Alan: Assumption is that restarting publishers get state out of
    band
  • Martin: VoD content can use a static value of 0
  • Alan: Cache key changes because track/object/group is no longer
    unique
  • Cullen: Group IDs between epochs are no longe comparable so
    jitterbuffer needs to work around this. Do epochs need to be
    distributed to all relays (after restart, clients migth connect
    to different relays)?
  • Victor: Relays can reject everything below largest epoch
  • Cullen: Would new epochs kill last objects on old epochs?
  • Victor: Yes would recommend that. Epoch changes are infrequent
    and means previous client was gone for more than 100ms.
  • Will: This proposal provides one solutions, but leaving it to
    application would allow multiple different solutions fitting
    different use cases?
  • Victor: Always setting epoch to 0 would be the same as MoQT
    today
  • Martin ?
  • Mo: Don't think it's worth the extra overhead for relays
  • Magnus: Could the proposal be added later as extension?
  • Alan: Ingest relay could accept with epoch and forward without
    (?)
  • Suhas: Changing cache key adds more complexity

• Conclusion:

  • Victor: Could be added as extension later but would mean
    applications could not (yet) depend on it
  • Magnus: Many open questions, maybe more deployment experience in
    the future would help
  • Victor: Want to know as subscriber when publisher crashes to
    reset state
  • Cullen: ?
  • Martin: Do we think this is a problem that needs to be solved
    and is this the right solution?
  • Cullen: Would a session ID header extension solve the problem?

• Victor will write a PR

• Further discussion in the issue and in interim meeting in two weeks
• Open question: Do we need to change the object model to solve this
  issue?

Admin reminder

• Deadline for commenting on next interim meeting planning is tomorrow

Publish (#995)

• Please review the PR
• Matching of namespaces changes from exact to prefix matching which
  might be separated into a separate PR

Track Status (#1015)

• Alan presents slides
• Will: TRACK_STATUS_OK from cache is not ok, need to get it from
  upstream
• Cullen: Should remove track alias
• Alan: Track status doesn't consume track aliases
• Suhas: If everything is the same as subscribe ok, maybe it should be
  a flag on subscribe.