+1 on the zfs replication. That’s what I use as well with a 3 node cluster. Gives redundancy and high availability without building a ceph cluster (which needs 4 nodes if I remember correctly)