The workaround we used was to open and close a new connection for every URL.