Notice when Jammy is installed using ZFS, separate datasets are created for most sub directories under /var, /srv , /usr and /usr/local. Why would someone want to maintain /var/lib/apt in a snapshot apart from /var/lib/dpkg for example? Are packages with their own dataset, such at NetworkManager and AccountsServer built so that all the other files relating to those services in /etc and /lib/systemd and etc can tolerate separately maintained snapshots in /var/lib? I'd appreciate knowing why the proliferation of all these generates more value than the complexity required. I can appreciate keeping users and /var/log in its own dataset, and /boot for reasons of compatibility, but the rest seems to generate more complexity than value given.
Also, notice no dataset is created for swap, favoring instead a dedicated partition (and making it harder later on to switch to a mirror or etc, with the special setup necessary for swap on zfs). If ZFS is going to be used at all, swap should go there by default. ZFS likes 'whole disks' to the extent possible, also let users benefit from the bitrot protection ZFS offers by locating swap there.
Last, might it be better to leave /boot/efi on ZFS, with /boot/efisda .. /boot/efisdb created on the specific drives formatted to fat then 'synced' when /boot/efi changes?
I appreciate ZFS being an install option at all! The above are 'next things'.
Harry Coin