In some cases there is no need to have the ZK data persisted to disk. For example if all you are doing is group membership and leadership
election the data is totally ephemeral, storing on disk is unnecessary. We've also seen cases where any non-ephemeral data can be
easily recovered (say configuration data that's generated/read and loaded into zk) and there is less need to worry about recovery of the
data in the case of catastrophic failure (meaning all replicas are lost, remember, recovery is automatic if 2n+1 servers and <= n servers
fail, even if > n fail manual recovery is still possible as long as at least 1 replica, or replica backup can be recovered)
In these cases it makes sense to have a "diskless" zookeeper ensemble. The result should be improved write performance
an less moving parts (no disk to fail!), simplifiying ops in cases where this can be applied.