bootc rollback already covers /usr, so snapshotting the root disk duplicates
it. What snapshots would cover is /var, and they cover it badly: they live on
the same pool as the VM, so fast-vmstore dying takes them with it.
Specific reason not to automate: fast-vmstore is LVM-thin, a snapshot pins
every block it references, and Runbook 18 already records that Data% only
ratchets up. Proxmox schedules backups but not snapshots, so the deletion
half would be ours to write, and that is the half that gets skipped.
By hand before a scary change, yes. As a capability to build, no. Also
records that the irreplaceable data here is small and the media library is
re-acquirable, which are two different problems.
itg into the inventory: the bootstrap chicken-and-egg, the three Fedora
assumptions that are false on Debian (no sudo at all, visudo not on a
non-login PATH, %sudo instead of %wheel and it wants a password), the
subscription nag with the assert that keeps it honest, and an explicit note
on what was left out and why.
Also the two Ansible warnings that had been printing on every fleet run:
pipelining off means -tt means a pty means a stray CR after the JSON, and
why interpreter_python=auto_silent rather than pinning the interpreter in
group_vars/all.
