A longhorn volume can only be attached to one node at at time. It can share it with other nodes over nfs. I don't think this is going to scale well.

Just use Postgres with ro replicas.