I've already written about some of my backup strategies in Proxmox. Proxmox Backup Server is an option, but it's not always the best option, especially if you're using lxc containers.

LVM and Ceph RBD baked containers have already been covered in another post, but one of the (many) great options, if you use Proxmox, is ZFS. I extensively use ZFS both on FreeBSD and Linux (and always wished that BTRFS could reach the same level of reliability).

When I don't need a networked file system (ceph) or want to use LVM, I tend to install Proxmox VMs and lxc containers on ZFS.. Let's now focus on backing up lxc containers.

Proxmox uses ZFS datasets for lxc containers' storage so you'll find all your files on /poolname/subvol-x-disk-y . We can easily backup as we've done in my previous article, we just need a different method for taking snapshots of all those datasets.

ZFS datasets have a hidden .zfs directory that contains all the snapshots that currently exist of that specific dataset. ls won't show it, but you can cd and it will be working.

Of course we can use native zfs send/receive or a tool like zfs-autobackup, which I use daily for local snapshots and remote replication, but we want to save the files, not the zfs dataset, so we can be able to backup to a different file system. Any file system. So we will be using borg.

Let's suppose our ZFS pool is named "proxzfs". Here is a suggested script. Of course, this is my script, it works for me and I'm not responsible if it doesn't work for you/destroys all your data/eats your server/etc.

#!/bin/bash

/usr/sbin/zfs snapshot -r proxzfs@forborg

REPOSITORY=yourpath/server/whatever:borgrepository/
TAG=mytag
borg create -v --stats --compression lz4 --progress    \
   $REPOSITORY::$TAG'-{now:%Y-%m-%dT%H:%M:%S}'          \
   /proxzfs/*/.zfs/snapshot/forborg/  \
   --exclude '*subvolYouMayWantToExclude-disk-0*'

/usr/sbin/zfs destroy -vrR proxzfs@forborg

borg prune -v $REPOSITORY --stats --prefix $TAG'-' \
   --keep-daily=31 --keep-weekly=4 --keep-monthly=12

This small script will create a @forborg snapshot for any dataset it will find under "proxzfs", then will fire up borg and ask it to traverse the forborg snapshots automatically mounted inside the .zfs directory of any dataset.

After that, it will destroy the 'forborg' snapshots and execute a borg prune_._ That will delete the old backups, according to the policy you have established. This step can be avoided here but I prefer to perform it after a backup so my repository is always consistent with my policy.