Staff has to follow the docs, which was designed with other objectives in mind (e.g. minimizing support load). I do not blame them, but unless there's a logical reasoning why exactly that would be a problem, one can't asses how valid it is.
I could also turn it upside down and submit a bug report claiming that the constant /var/lib/pve... writes are killing my SSDs. Sure the answer would be get a proper datacentre kind, it's the easiest answer one can give. It's been +10 years since the sqlite thing was designed, it's not the only way to do it either. It's been catching up with realities by e.g. increasing the max size for its own RAM occupancy. Even running that SQLite with
PRAGMA journal_mode=WAL should not be necessary if it was better designed.So I do not like blanket statements implying that there's only one proper way to do it, sure there's an official way to do it and there's risk-calculated approaches one can take. I was e.g. able to run entire PVE out of RAM-disk booting over iPXE, diskless. It would write-back its state when shutting down. Unclean shutdown? Consider the node dead, will "reinstall" anyhow. Anything bad happened? Nothing. Did I have disaster recovery plan? Yes, backups, like everyone should.
Case in point: I suppose one also does not tell home-labbers to go buy dedicated SLOG devices for their ZFS mirror?
Last edited: