(Urgent) Room_finished webhook never delivered + session never finalized for ~12 rooms (project taptalentmain-prtncqhm)

Agents
1

Hi LiveKit team,

We’re seeing rooms that are successfully deleted but never finalized: no room_finished webhook is emitted, and the session stays open indefinitely in Cloud dashboard analytics. We’ve hit this on ~12 interviews recently. One fully documented example below.

Project: taptalentmain-prtncqhm (dashboard project “TapTalent_Main”)

Example room

  • Room name: 0238f75f-b0cb-42dc-b756-200a18b17fe4_1788095328218
  • Room SID: RM_cBBpLC2tgKoD
  • Session ID (dashboard): RM_cBBpLC2tgKoD
  • Created: 2026-08-30 13:08:48 UTC · Deleted: 2026-08-30 ~13:52:25 UTC (server-side DeleteRoom issued by our agent via delete_room_on_close, while 3 participants were connected)

What we observed

  1. All three participant_left webhooks (agent, candidate, avatar participant) were delivered to our endpoint at 13:52:25–26 UTC, each with disconnectReason: ROOM_DELETED — so the deletion itself completed and webhook delivery to our endpoint was working at that exact moment.
  2. No room_finished was ever delivered — verified in our webhook receiver’s logs, which record every incoming webhook before any filtering. Checked continuously for 6+ hours after deletion.
  3. ListParticipants/ListRooms confirm the room no longer exists on the SFU.
  4. The Cloud dashboard session for this room never finalized: END TIME is empty and DURATION kept counting for 2+ hours after deletion (141 min at last check), while “total WebRTC participant minutes” correctly stopped at 87 and the participant timeline sits at zero.

Controls we ran (why we believe this is on the LiveKit side)

  • The same webhook endpoint received room_finished for 20 other rooms the same day (14:00–20:00 IST window), so our receiver, URL, and signature validation are fine.
  • Several of those successfully-finalized rooms were served by the same self-hosted agent workers, same code path, same delete_room_on_close flow — so it isn’t correlated with our agent type or how the room is closed.
  • The failure is per-room: for affected rooms, participant_left arrives but room_finished never does, and in every case the dashboard session is also stuck open. The missing webhook and the unfinalised session appear to be the same internal event.

Other affected rooms (same signature — participant_left delivered, no room_finished, session stuck open):

  • RM_cBBpLC2tgKoD
  • RM_GhcPXihVe2XK
  • RM_vbKJ2jBye3JV
  • RM_iWBWrqkzHZGp
  • RM_cBBpLC2tgKoD
  • RM_J5XAuGQDvkfk
  • RM_DUJgmqNmbp2C

Impact: our post-interview pipeline (duration recording, scoring trigger) keys on room_finished, so each occurrence requires manual intervention. We wanted to report it since the stuck sessions suggest an internal finalization failure rather than webhook delivery loss.

Questions

  1. Can you confirm whether room_finished was generated internally for RM_cBBpLC2tgKoD and if so where delivery failed?
  2. Is there a known issue finalizing rooms deleted server-side while participants are still connected?
  3. Can the stuck session records be closed retroactively?

Happy to provide webhook receiver logs, exact delivery timestamps for the other 7 rooms, or anything else useful.

Thanks,
Vishal Maurya

2

Hi Vishal, thank you for raising. This is now with the engineering team. I see you also raised a support ticket for this so you may hear back through that ticket.

3

When I suppose to get the reply back

4

Thanks for reporting, Vishal. It looks like those events were dropped for your account. We are looking into it. Thanks for reporting.