Many Agent Insights sessions remain ACTIVE after `room_finished` and RoomService confirms the room does not exist

Hi LiveKit team,

We are seeing a significant number of sessions remain marked ACTIVE in the LiveKit Cloud dashboard for many hours after their rooms have closed.

This is not an occasional dashboard inaccuracy or a short synchronization delay. It is a recurring observability gap affecting multiple sessions and making the dashboard unreliable as a source of session state.

Here is one verified example:

  • Room name: livekit-web-20260729T0715-6Xb97c
  • Room SID: RM_exXEuh2zWRhT
  • Agent job ID: AJ_oBasNYWr3Akg
  • Room created: 2026-07-29T07:15:04.523Z
  • Room finished: approximately 2026-07-29T07:16:32Z
  • room_finished event ID: EV_kSQrpaGy52f7

For this session:

  • All participants disconnected.
  • The agent disconnected with ROOM_DELETED.
  • LiveKit emitted room_finished.
  • RoomService subsequently returned no room for this name.
  • The agent dispatch API returned requested room does not exist.
  • The LiveKit Cloud session nevertheless remained marked ACTIVE many hours later.

There is no remaining room or participant that can be cleaned up from our side. LiveKit’s room lifecycle completed correctly, but the corresponding observability record was not finalized.

This prevents us from reliably distinguishing genuine active-room leaks from stale dashboard records and makes the dashboard unreliable for operational monitoring.

We can provide additional affected room SIDs, screenshots, or any other information required. Could you please investigate and explain why these sessions remain marked ACTIVE after LiveKit has emitted room_finished and RoomService confirms that the room no longer exists?

Given the number of affected sessions, this appears to be a systematic observability gap.

Thanks.

Looking at the active sessions in your project, I notice that they are all duplicated. I.e. each session has more than one row associated with it, only one of which shows as active.

This is a bug.
I checked internally and it looks like some other projects are also affected, and a fix is currently rolling out.

Please be aware that this will NOT have an impact on your billing.

Apologies for the inconvenience.