Implement a distributed database to store the locations of filesystems and their replications. This will probably extend fedfs: https://fedoraproject.org/wiki/Features/FedFS We tried packaging fedfs before and it didn't get a lot of interest, but perhaps it would be more interesting in combination with AFS-like read-only replication such as that based on bug 2005929. We want a krb5-authenticated network protocol to create, nfs-export, snapshot, and replicate volumes, and to update replicated volumes, and maintain in fedfs a catalog of those volumes and any snapshot-of/replica-of relationships. This should make it possible to, for example, update replicas with a single command. That command will take care of transmitting the volume (using send/receive), unmounting the old filesystem, mounting the new one in its place, updating fedfs, and reexporting or restarting any NFS server as necessary.
Also, closely related: 1. Support junction points and a global namespace root so that an NFS client can mount example.com:/ and browse to all the volumes exported by the local organization. 2. Consider implementing a magic /nfs directory like the /afs directory, so that e.g. /afs/example.com/ gives access to example.com's exports. Use the mechanism for DNS discovery described in https://datatracker.ietf.org/doc/html/rfc6641. Those two might eventually need to be separate bugs.