LiveKit: 2.5% of Inbound UDP Calls Dropping
Timeline of updates
- resolved
Our SIP load balancer runs inside of Kubernetes, which internally routes packets from the virtual pod IP to the host IP. We believe a linux kernel bug related to conntrack is causing certain packets (including our 200 OK responses) to be lost in that process. This bug impacts about 2.5% of inbound calls over UDP. Since we sent the 200, but it doesn't make it to the trunking provider, our system believes the call is connected when it isn't. Since TCP retries lost packets, it is not impacted by this bug. We will be rolling out a permanent fix in the next few weeks.
From LiveKit’s published incident history: for incidents before Rewindex began polling this status page, only the final update text is available.
Impact, status and times are the vendor’s own, as last archived. View on status.livekit.io (opens in new tab).