Only a small number of Apps were affected. They have not been totally down as the second Node (third, fourth …) was still responding. But indeed, that's not good. One failing Node should not cause downtime for Professional Apps on a production plan, as the load balancer should not use a Node in error state. Problem here was, that the health check of the frontend reported to be OK, while in fact it wasn't. We think that an internal DNS lookup is the related issue. We are looking into that.