Setup: voice agents on livekit-agents 1.4.6 / livekit 1.1.2 / livekit-api 1.0.7 (Python 3.11), inbound SIP telephony, LiveKit Cloud (project p_3tqm7ro6kbs). Each call = one agent job; ctx.connect() is invoked at job entry. Background: we previously reported subscriber PeerConnection failures in liv…

Participant Migration and State Mismatch issues in certain SIP calls opened 08:42PM - 03 Feb 26 UTC (UTC) closed 07:55PM - 13 Feb 26 UTC (UTC) [image] zaheerabbas-prod…

[image] Zaheer Abbas: livekit 1.1.2 I think the reconnect hardening is not in 1.1.2 but in 1.1.8. Along with this PR in 1.6.x fix quick reconnect participant keyerror by tinalenguyen · Pull Request #5979 · livekit/agents · GitHub . Right?

The above PR you have linked seems a diff issue Participant Migration and State Mismatch issues in certain SIP calls · Issue #4705 · livekit/agents · GitHub → check this issue where I had raised the same state mismatch and Subscriber pc state failed. It was mentioned here Participant Migration and …

I think your mapping is a little off. Agents 1.4.2 and 1.4.6 both pin livekit==1.1.2 in their pyproject.toml, and current livekit-agents 1.6.0 pins `livekit==1.1.9. Being on agents 1.4.6 leaves you on the same rtc 1.1.2 where I believe the bug lives. It does not seem to be fixed in 1.4.6 as you hav…

Sorry I mentioned 1.2.0 in my earlier message Check this GH message - Participant Migration and State Mismatch issues in certain SIP calls · Issue #4705 · livekit/agents · GitHub which was made in February And this PR linked to this issue - fix full_reconnect downgrade & don't ignore Leave message…

Ok, maybe agents team will weigh in on that comment you made in the PR. I don’t think it is fixed in 1.4.6 but maybe I am wrong. If it is still broken in 1.6.0 it won’t be fixed until at least 1.6.1. I will highlight this thread to the team. If you have a code example that reproduces the issue and…

The problem is this is NOT replicable and occurs intermittently only during production SIP calls. I myself haven’t been able to replicate with multiple hardened network conditions and seems to be an issue on LiveKit Cloud server side

I will ask if we can add some extra logging in 1.6.1 and see if we can get something that will help you find it. From server logs it looks like agent applying the migration Answer too late/stale the same event the SDK logs as signal_event taking too much time.

Thanks for the update @CWilson - any eta on when this will be released and what version of livekit/agents and rtc sdks we will need to upgrade? Any way to verify this fixes things before we do the upgrade? We were told the same last time that upgrading to 1.4.x would solve the issue :folded_hands:

Server-initiated migration fails to resume on agents 1.4.6 — subscriber + publisher PC fail, no recovery, process killed (expected fixed in >1.4.2 per agents #4705)

Getting Started

CWilson June 17, 2026, 11:49am 10

The team dug deeper into the server-side logs for the session referenced above. We believe this PR will address the issue. It appears the failure is if the resume is somehow unsuccessful. This PR addresses that.

Once this is released, I hope you can let us know if the issue is resolved for you.

Signal connection times out on the "v0 path" at agent join, forcing a fallback that adds 0.5–5s of call-setup latency

Topic		Replies	Views
Publisher pc state failed and the agent resumed frequently at around the 8-minute mark Agents webrtc , livekit-cloud	15	106	May 14, 2026
Signal connection times out on the "v0 path" at agent join, forcing a fallback that adds 0.5–5s of call-setup latency Getting Started	46	146	June 23, 2026
Issue: [Python SDK] Network switch causes "stuck-in-reconnecting" and "ghost room" states in livekit-agents Agents agent-sdk-python , livekit-cloud	3	25	May 19, 2026
Livekit Outage - Production Getting Started livekit-cloud	11	140	June 2, 2026
Agent Session Connection Timeout without Redispatch 4/3 3 PM PST Agents agent-development , agent-deployment , livekit-cloud	0	26	April 4, 2026

Server-initiated migration fails to resume on agents 1.4.6 — subscriber + publisher PC fail, no recovery, process killed (expected fixed in >1.4.2 per agents #4705)

Related topics