Room.sid hangs indefinitely when room_sid_changed FFI event is not delivered

[damonvjanis]

Summary

Calling await room.sid from inside an agent's JobContext entrypoint occasionally hangs forever, even though the LiveKit server has assigned a SID to the room (visible in Cloud Analytics). Other FFI-driven events for the same Room instance (participant_connected, track_published, connection_quality_changed, SIP hangup → participant_disconnected) continue to flow normally throughout the hang.

The caller experiences silent dial-tone — the agent never publishes audio — until they give up and hang up.

Environment

• livekit==1.1.5
• livekit-agents==1.5.6
• LiveKit Cloud, SIP-inbound voice agents, ~10 concurrent sessions per worker pod

Mechanism

Room._first_sid_future is created in Room.__init__ and is only resolved by the room_sid_changed FFI event handler:

```python

rtc/room.py:189

@Property
async def sid(self) -> str:
if self._info.sid:
return self._info.sid
return await self._first_sid_future

rtc/room.py:770

elif which == "room_sid_changed":
if not self._info.sid:
self._first_sid_future.set_result(event.room_sid_changed.sid)
self._info.sid = event.room_sid_changed.sid
```

In the failure case, _first_sid_future is never set. No exception, no log — the awaiter just suspends forever. Meanwhile the same Room instance continues to receive and dispatch all other FFI events.

Recurrence

• 21 confirmed occurrences over 30 days in one production deployment (~0.7/day)
• Rate is steady — not correlated with any deploy or service version
• Affects multiple pods, multiple service versions, multiple SIP callers
• Each occurrence costs one customer call (caller hangs up at ~60s SIP ringback timeout)

Evidence the SID exists server-side

For affected sessions, the LiveKit Cloud Analytics REST API (`GET /api/project//sessions/`) returns the room with the expected SID. The server assigned it; only the SDK's local future never resolves.

Suspected cause

Race or lost event in the FFI event queue / event-loop bridge specific to `room_sid_changed`. We haven't reproduced it in development yet — the rate is too low.

Workaround we're applying

```python
try:
sid = await asyncio.wait_for(room.sid, timeout=5.0)
except asyncio.TimeoutError:
# log + fall back; abort the job so the worker frees up
...
```

Converts the silent hang into a recoverable error.

What would help from upstream

• Confirmation of whether this is a known/possible failure mode
• An internal timeout on `_first_sid_future` after `connect()` succeeds, raising `RuntimeError` rather than hanging — so callers don't have to arm their own timeout
• Optionally, a way to detect "room connected but SID unresolved" without polling

Happy to share anonymized trace fingerprints if useful.