Running Edge 2024.5.2 on Dell R750 nodes in our Tokyo facility. Dual NIC setup handles the primary WAN link and local LAN routing. Network team pushed a firmware update last week that caused the primary uplink to flap for about forty seconds. Edge automatically triggered survivability mode. Local SIP phones stayed registered and voice quality remained solid. The problem shows up with the cloud AI bot traffic. Architect flow routes the bot media through the Edge for local termination. Audio drops completely once failover hits. Console logs show repeated SIP/2.0 488 Not Acceptable Here responses on the bot’s RTP stream. Codec negotiation forces a fallback to PCMU, but the bot engine expects Opus 48kHz. CPU spikes to 92% on the secondary core handling the media bridging. BIOS settings are locked to default power performance mode, so thermal throttling isn’t the culprit.
Tried adjusting the local media buffer in the appliance settings, but that just increases latency without fixing the packet loss. Packet captures on the local 10G switch show the RTCP SR packets getting dropped right after the interface flips to the backup VLAN. Bot traffic doesn’t survive the media path rewrite. The survivability config API throws a 400 Validation failed when trying to force Opus through the local media relay. Looks like the Edge media engine drops any non-standard payload types during failover. We’re stuck routing bot audio through the cloud STT service, which defeats the whole point of local survivability. Mic stays hot on the fallback path, but audio is just dead silence. The last resort is spinning up a separate SIP trunk just for the bot traffic, but that adds another failure point to the rack.