Motivation
Currently, we use a synchronised, sequential approach to update the metadata. In some cases, for example, one or two data server restarts cause many shards to trigger a new election, which will introduce too much conflict when acquiring the lock.