fix: wait_for_participant waits until participant is fully active

[davidzhao]

requires livekit v1.1.4

[chenghao-mou]

should we update the pyproject and uv lock files?

[davidzhao]

should we update the pyproject and uv lock files?

yup, waiting for python sdk release before doing so..

[devin-ai-integration]

Devin Review found 1 new potential issue.

View 6 additional findings in Devin Review.

[IngLP]

hi @davidzhao , this breaks my implementation! I was using wait_for_participant to setup the agent according to participant metadata, and this used to work because I had time to do so, while the telephone was ringing. Now I am losing that precious time to do the setup.

[davidzhao]

@IngLP can you clarify what actually happens in your use case? this should introduce less than 500ms of latency to wait_for_participant in the typical case.

[IngLP]

@davidzhao maybe that's true for incoming calls, but for outgoing calls, it's not. I had plenty of time while the other extension was ringing, now wait_for_participant only unlocks when the callee has actually answered.

[davidzhao]

@IngLP can you share a code snippet of what you are doing? and describe the current behavior that you are observing?

[IngLP]

Before:

• Outgoing call launched by backend with CreateSIPParticipant, agent config written in sip participant metadata
• The call starts and it's ringing
• Agent starts, waits for participant, reads metadata, setups agent and tools
• Call is still ringing
• Contact answers and everything is ready

Now:

• Contact answers and everything has still to be set up, including recording, which cuts part of the call.

[davidzhao]

@IngLP please share example code snippet.. that'll make it easier for us to see what's going on. if you are calling specific APIs, include the actual calls with the parameters

[IngLP]

@davidzhao it's just like this:

@server.rtc_session()
async def entrypoint(ctx: JobContext):
    [other setup code]

    logger.info(f"connecting to room {ctx.room.name}")
    await ctx.connect(auto_subscribe=AutoSubscribe.AUDIO_ONLY)

    try:
        participant = await ctx.wait_for_participant()
    except RuntimeError as exc:
        if _is_expected_participant_wait_disconnect(exc):
            logger.info(
                "Room disconnected before participant joined; ending job without agent startup"
            )
            return
        raise
    
    [setup agent according to participant metadata, other code]

    agent_session = AgentSessionExt(
        **components,
    )
    await agent_session.start_session(agent, participant, room=ctx.room)

[davidzhao]

@IngLP where do you perform the outbound dial?

[IngLP]

@davidzhao in the backend, a separate deployment from the agents runner.

[decafdennis]

This change also hangs the wait_for_participant call in console mode, because create_mock_room does not mock out the participant state and so the participant never becomes active.