A routine infrastructure upgrade briefly restarted a backend server, causing about 2 minutes of downtime.