What to back up

--paths is required and has no default: the archive contains exactly the paths you list, separated by ;. Quote the list, because ; separates commands in every shell.

Recommended paths

--paths='/etc/.;/etc/pve/.;/var/lib/pve-cluster/.;/var/spool/cron/crontabs;/root/.ssh'

Path	What it holds	Why
/etc/.	The node’s own configuration: network/interfaces, hosts, hostname, resolv.conf, fstab, kernel/cmdline, modprobe.d/, apt sources, vzdump.conf, corosync/, ZFS and LVM configuration, SSH host keys	Everything you would otherwise rebuild by hand after reinstalling the node
/etc/pve/.	The cluster configuration as readable files: storage.cfg, datacenter.cfg, user.cfg, jobs.cfg, firewall rules, guest configurations, certificates, priv/	To take back a single file: a storage definition, a VM configuration
/var/lib/pve-cluster/.	config.db: the database the files of /etc/pve are stored in	The unit the Proxmox VE documentation restores when a node is lost: see Restore
/var/spool/cron/crontabs	Cron jobs created with crontab -e	They are not under /etc
/root/.ssh	root’s keys and authorized_keys	Access to and from the other nodes and your scripts

Add the folders where you keep your own scripts or files, such as /root/scripts. The cv4pve-admin module starts from a similar list and also includes /var/lib/ceph/.: we have not measured it on a Ceph cluster, so check the size of the archive if you add it.

Why /etc/. misses /etc/pve

/etc/pve is not a normal folder: it is pmxcfs, the Proxmox cluster file system, mounted on top of /etc. The tool runs tar with --one-file-system, which does not enter other file systems mounted inside a path. So /etc/. gives you everything in /etc except /etc/pve.

The content of /etc/pve is saved in two ways, and you want both:

• /var/lib/pve-cluster/.: pmxcfs keeps all of /etc/pve in the SQLite database config.db. This is the file the Proxmox VE documentation moves to a new host to recover a node.
• /etc/pve/. listed explicitly: tar archives a path given on the command line even when it is a separate file system, so you get the same content as plain files you can open and copy back one by one.

The same rule applies to anything else mounted under a path you list: it is left out, and you have to list it on its own.

The cluster database is copied while in use

config.db works with a write-ahead log: the latest changes sit in config.db-wal next to it, not yet in config.db. /var/lib/pve-cluster/. saves both, together with config.db-shm. Keep the three files together: config.db alone is an older state of the configuration. Restore shows how to merge them.

The files are read while pmxcfs is running, as with any file-level copy of a live database. In our tests the copies opened with SQLite passed its integrity check; the readable copy of /etc/pve/. is a second source if one day it does not.

Path syntax

• Use absolute paths. The archive stores them as written: with /etc/. the entries are named /etc/./network/interfaces, with /etc they are /etc/network/interfaces. You need the exact name to extract a single file, so list the archive first.
• A path that does not exist on the node is left out with a warning and the run goes on, so the same list works on nodes with and without, say, /root/scripts. A typo gives the same warning, not an error: read the warnings.
• A path cannot contain a single quote '.