A vulnerability mitigation update required replacing compute cluster nodes, which, when applied, rolled back due to a timeout. This caused some workloads (API, delta, builder) to become temporarily unavailable and triggering some undesired secondary effects, including ungracefully terminating a few long lived instances servicing VPN connections. While the rest of the services came back reasonably quickly within a minute or so, it too around one and a quarter hours to re-establish VPN tunnels.
A scheduled maintenance will be posted later to perform this update during a planned outage window.