Backing up the PVE node - dd, LVM snapshot, or better option?

[bobbybb]

I need to install some software on my PVE node because I cannot use dual GPUs via PCIe passthrough on my VMs.
In case this messed something up how can I backup the root node (just a debian machine, right?) so I can restore it if I mess something up.

I was first thinking of doing dd on the whole drive to an external drive, but that's a lot of data and it would backup the VMs too which I already have backed up with PBS.

Can I take a LVM snapshot of the root node, install my software, and revert back to the snapshot if I mess something up?

What's the best path forward?

 

[ferrets]

I was first thinking of doing dd on the whole drive

How about doing dd with partition?

Or can you run the software in LXC container that you need?

 

[Kingneutron]

https://github.com/kneutron/ansitest/tree/master/proxmox

Look into the bkpcrit script, point it to external disk / NAS, run it nightly in cron. Comments in-script detail where the critical stuff is.

If you're using standard lvm+ext4 root, look into the bkpsys-2fsarchive and RESTORE* scripts. Practice doing a full bare-metal backup of rootfs and restoring PVE into a vm with host-only networking, so you don't get IP conflicts. Easiest way that I know of to access the fsarchiver restore file on the host from in-vm is with sshfs.

https://github.com/nchevsky/systemrescue-zfs/releases

.

If you're using zfs boot/root, you can install to a mirror / zfs RAID1. bkpcrit script still comes in handy here.

 

[gfngfn256]

I was first thinking of doing dd on the whole drive to an external drive, but that's a lot of data

You could use gzip to at least compress the outputted disk image, as I show here.

and it would backup the VMs too

The exact reason, I have PVE host on separate disk/partition from VMs etc.

 

[bobbybb]

ok this makes sense, thanks for the replies everyone. I have pve-root on a partition so I made a dd image of that for now just in case.

Now, I'm curious if a LXC can utilize both GPUs on my host as this would be helpful testing LLM models that need dual GPU access.
I'm familiar with docker, but not so much LXC..

is there a GUI like portainer and a registry of containers like dockerhub?
or if I want to create a Ubuntu LXC to interact with multiple GPUs do I need to setup everything via the command line?

 

[Dark26]

you can use pbs-backup-client with some script, to backup what you want ( file / harddrive/partition. )