Please see this call ID: SCL_r4KZHxriQUAX. The RM_SpePpXW4Rem9 was created and the agent and participant joined. However, the 200OK was never returned to Telnyx. Both the LIveKit and the Telnyx logs shows this on the SIP captures. The room stayed online for five minutes and the agent debug shows that a 200OK was sent.
Looks like that 200 OK was in response to the CANCEL that LiveKit received from Telnyx.
Looking at the session events for RM_SpePpXW4Rem9, I see the SIP participant left after 16 seconds, then 2 seconds later the agent joined. That’s why the PCAP shows the call as 180 ringing for 16 seconds before the caller hung up.
I’m not sure why the agent took so long to join. I don’t see you hitting any of your build plan limits, perhaps a long Deployment management | LiveKit Documentation, but that is just a guess.
Thanks Darryn. The user left the room because the agent did not greet them (i.e., they did not hear the audio from the agent). If you put the times together, you’ll see the agent and user were in the room and that the agent was active. So because the agent was malfunctioning, that’s why the user hung up. THe agent log dashboard shows that a 200OK was transmitted (supposedly by the agent joining the room). Was that the 200OK from the webRTC SIP or what was supposed to be the SIP 200OK on the trunk?
If you look at the session events at Sign in | LiveKit Cloud it shows the SIP participant leaving at 18:56:18 (UK time, that’s just how the dashboard presents the data to me) and the agent joining 2 seconds later.
What would cause the delay in the Agent joining the room? That’s actually what the problem was.
What would cause the delay in the Agent joining the room?
I’m, not sure exactly. Typically this is either due to cold starts, or the account has reached its agent concurrency limit. It might also be worth checking the agent logs, Where to find your Agent Logs | LiveKit, but take note of the caveat towards the bottom of that blog about “logs related to agent dispatch or job assignment” not being available in Agent Observability, so this would depend on you capturing logs using an external monitoring service.
I swear when I looked yesterday your project was on the build plan, but now it looks like it is on ship; if you recently upgraded, I would not expect to see a repeat of this issue.
Thanks Darryn. For reference, yes, we upgraded. I’ll be on the lookout for a recurrence of this issue.