WebRTC softphone drops iceConnectionState to failed during ACMA recording prompt on Sydney edge

Voice & Telephony
1

Pushed the updated edge configuration through Terraform v1.5.7 last night against mypurecloud.com.au. WebRTC softphone works fine for internal transfers, but the media path breaks exactly when the ACMA recording announcement triggers. Agents in Brisbane and Melbourne report the audio cutting out. Latency to the Sydney edge sits around 35ms normally. Browser console shows RTCPeerConnection flipping to failed right at the 0:04 mark. The token refresh endpoint /api/v2/telephony/providers/edge/users/{id}/webrtc/token gives a 408 when softphone attempts renewal during announcement playback.

ACMA compliance rules force that recording prompt to inject before any PSTN handoff to +61 numbers. Architect flow handles the recording action, but the WebRTC stream cuts out completely. Checked the call flow trace, media server in Sydney is throwing 488 Not Acceptable Here on the SDP answer. Codec negotiation seems dropping the opus stream when recording prompt adds extra DSP processing.

Tried adjusting the STUN/TURN timeouts in softphone settings, doing jack all. Can’t get the negotiation to hold. The issue only happens on Sydney edge, not when routing through Virginia for test calls. Number porting registry shows the +61 prefixes are active, so it’s not a routing mismatch. Just the WebRTC media path collapsing when compliance audio kicks in. Mic stays hot but the audio channel dies.

RTCPeerConnection.oniceconnectionstatechange: failed
SDP negotiation error: 488 Not Acceptable Here
Edge: Sydney (AU)
Client: Chrome 124.0.6367.91
WebRTC SDK: Genesys Cloud Softphone v2.1.4

2

Yeah, that ACMA prompt triggers a mid-call re-negotiation that often trips up the SDP exchange if the TURN candidates aren’t fully resolved. It’s not usually a Terraform issue, but rather how the client handles the ICE restart.

Check your genesys_cloud_user_settings block in your TF state. You might need to force the WebRTC configuration to prefer TURN over STUN during media changes.

resource "genesys_cloud_user_settings" "agent" {
 user_id = var.agent_user_id
 
 settings {
 key = "webrtc"
 value = jsonencode({
 iceTransportPolicy = "relay" # Forces TURN, bypassing STUN failures
 turnServers = [
 {
 urls = ["turn:turn.pure.cloud:443?transport=tcp"]
 username = "generated_user"
 credential = "generated_cred"
 }
 ]
 })
 }
}

Running plan after this usually shows a drift if the manual UI settings overrode your code. The relay policy is stricter but saves you from those edge-specific NAT hairpins in Sydney.