Recovering a multi-master node after a server failure