Proxmox on Sysadmin Tales

Deduplicated Proxmox backups with Proxmox Backup Server and ZFS

Wed, 22 Apr 2026 23:23:10 +0000

Overview

There are many ways to back up your Proxmox virtual machines.

One of the most popular ways to do so is via Proxmox Backup Server, which gives you many advantages:

Deduplication of common data.
The ability to cherry pick files out of your backups.
Tight integration with the Proxmox ecosystem.

However, how is one to achieve proper 3-2-1 backups using this mechanism?

One way is to run multiple datastores, some local and some using Amazon S3 compatible endpoints.

However, this is a bit much for me, so I came up with a simpler method.

The Details

I run my PBS on a Beelink N100 mini PC, in which I’ve installed a 2TB SATA SSD that I formatted as a single disk zpool.

I then installed Jim Salter’s excellent Sanoid utility, configuring ZFS snapshots on the PBS server.

This is the /etc/sanoid.conf that I’ve defined. You can, of course, set whatever values you like.

[pbs-local/datastore]
        use_template = pbs
        recursive = yes

[template_pbs]
        frequently = 0
        hourly = 36
        daily = 3
        weekly = 1
        monthly = 0
        yearly = 0
        autosnap = yes
        autoprune = yes

Then, once I had my initial snapshots, I went to my TrueNAS backup server and created a new ZFS replication task.

I used this to pull the snapshots for the datastore to the backup server.

The key is to match the expected snapshot naming schema for Sanoid, like so:

This allows me to maintain the deduplicated nature of the Proxmox Backup Server datastore.

You can do the same with any offsite NAS that you may have, as long as it’s running ZFS. This is easily accomplished with something like Tailscale or Netbird.

If I ever have a failure of the Proxmox Backup Server, I can simply set up a new zpool on the repaired system and reverse the direction of the replication.

I also use ZFS Delegation to avoid doing the replication as root, but I’ve covered this topic before.

As always - trust nothing and test those backups!

Issues transferring Windows 11 VM from KVM to Proxmox

Fri, 20 Feb 2026 21:11:38 -0500

I ran into a strange issue when trying to transfer a Windows 11 Pro virtual machine from KVM on my Fedora desktop to Proxmox.

The VM was using a VirtIO block device for it’s storage as I normally do.

I transferred the VM qcow2 file to Proxmox and created a new VM with matching hardware, the imported the disk with:

qm importdisk 300 win11.qcow2 fast

(fast being the name of my SSD zpool.)

However when I went to boot the VM, it refused to boot at all.

Originally, it seemed to be some sort of issue with the EFI bootloader.

I was able to boot into a Windows install disk and recreate the EFI partition with diskpart and bcdboot.

Once I’d done that, the VM attempted to boot Windows. Rejoice… or not.

Nope, now it’s getting an INACCESSIBLE BOOT DEVICE BSOD.

I tried innumerable things without success, the golden ticket wound up being this superuser.com thread.

I had to first:

Change my boot disk to use SATA instead of VirtIO.
Boot into Windows.
Enable Safe Mode bcdedit /set "{current}" safeboot minimal.
Shut down the VM, change the boot disk back to VirtIO.
Let it boot up into Safe Mode.
Revert to normal boot. bcdedit /deletevalue "{current}" safeboot

I still don’t quite understand what the problem was here.

Windows has the concept of “boot start” drivers, yes, where it only loads a subset of the available drivers at first boot, but I had been using the VirtIO drivers since this OS was first installed, so why was this suddenly a problem?

Apparently, rebooting into Safe Mode this way forces Windows to load all available boot start drivers, and because VirtIO worked here, it then remembered that for the next, normal boot.

If anyone has an actual explanation for why this happened I’d appreciate it.

I wasted far too much time on this, I could have wiped and reloaded the VM many times over in the time it took me to fix this, but it bothered me far too much to let it go.

This kind of nonsense makes me glad I’ve switched to using almost exclusively Linux in my personal life.

Passing through Intel Arc A380 to a Proxmox VM

Sun, 17 Aug 2025 18:02:07 -0400

I recently purchased an AsRock Challenger ITX Intel Arc A380 GPU to use for my video transcoding needs.

It was a bit of a headache to get it passed through to a VM in Proxmox, so here’s a log of what worked for me, in hopes that it will help someone else down the road.

My VM is Ubuntu 24.04 LTS, running kernel 6.14 (HWE kernel).

Proxmox is version 8.4.

I was able to successfully pass through the GPU to the VM, but it shows as UNCLAIMED in lshw:

root@gputest:/home/scott# lshw -c display
*-display UNCLAIMED
description: VGA compatible controller
product: DG2 [Arc A380] vendor: Intel Corporation
physical id: 10 bus info: pci@0000:06:10.0
version: 05 width: 32 bits
clock: 33MHz
capabilities: msi pm vga_controller cap_list
configuration: latency=0

I’m not going to go through all the troubleshooting I did, just share what worked.

We need to force the kernel to load the i915 driver.

In addition to this, because I’m using cloud-init and one of Ubuntu’s Cloud Images, I can’t simply specify the kernel parameter in /etc/default/grub. If you’re not using cloud-init, ignore the next step and simply specify this in /etc/default/grub like so:

GRUB_CMDLINE_LINUX_DEFAULT="quiet i915.force_probe=56a5

Then run update-grub and reboot.

Note

These steps may be different if you’re using a distro like Fedora - I believe they now use grubby to accomplish this task.

However, for those of us using cloud-init, you can instead simply specify these parameters in /etc/modprobe.d/

sudo vim /etc/modprobe.d/i915.conf

Note

2025/08/23: This originally was set to load in the Xe driver and disable i915, but I found out later that things like Jellyfin do not have support for Xe, at least not yet. For now, you must force_probe the i915 driver instead.

As of today I have tested this with both Jellyfin and Plex and can confirm the card works as expected.

options i915 force_probe=56a5

Now that that’s done, here are my PCI passthrough settings on the VM:

root@pve01:~# qm config 100
<snip>
cpu: host
hostpci0: 0000:2f:00,pcie=1 <---- This is the GPU itself
hostpci1: 0000:30:00,pcie=1 <---- This is the GPU's audio controller.
<snip>

It does not need to be set as the primary GPU, but you do want your machine type to be q35 and you want to enable PCI express.

You also want to give the VM more than 4GB RAM (so it can initialize the large BAR).

And now on to the thing that took me hours to figure out.

CSM needs to be disabled in BIOS. At least with my motherboard, leaving it on meant that resizable BAR was disabled. Without this, passthrough didn’t work at all.

Other settings you should enable:

Above 4G Decoding (May also be called large memory allocation).
SR-IOV
IOMMU

Most of these can be found either in your PCI subsystem or chipset settings.

Now the card shows normally in lshw.

scott@gputest:~$ sudo lshw -c display
  *-display
       description: VGA compatible controller
       product: DG2 [Arc A380]
       vendor: Intel Corporation
       physical id: 0
       bus info: pci@0000:01:00.0
       version: 05
       width: 64 bits
       clock: 33MHz
       capabilities: pciexpress msi pm vga_controller bus_master cap_list rom
       configuration: driver=xe latency=0
       resources: iomemory:38000-37fff irq:38 memory:c0000000-c0ffffff memory:380000000000-3801ffffffff

I’m not sure that this is 100% needed, but Intel also recommend installing their repo and some of their support packages. It can’t hurt, so why not.

sudo apt-get update
sudo apt-get install -y software-properties-common
sudo add-apt-repository -y ppa:kobuk-team/intel-graphics
sudo apt-get install -y libze-intel-gpu1 libze1 intel-metrics-discovery intel-opencl-icd clinfo intel-gsc
sudo apt-get install -y intel-media-va-driver-non-free libmfx-gen1 libvpl2 libvpl-tools libva-glx2 va-driver-all vainfo

For some reason the intel_gpu_top and clinfo tools don’t seem to recognize the card. Possible that they just haven’t been updated with support for the newer Xe driver over i915.

However, we can confirm that things are working with vainfo and ffmpeg.

I actually took this a step further and built a Docker image with vainfo and ffmpeg (since again, my intention is to use this for Jellyfin in Docker.)

root@f7fe7efcfa51:/# vainfo
Trying display: wayland
error: XDG_RUNTIME_DIR is invalid or not set in the environment.
Trying display: x11
error: can't connect to X server!
Trying display: drm
libva info: VA-API version 1.22.0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_22
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.22 (libva 2.22.0)
vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 24.3.4 ()
vainfo: Supported profile and entrypoints
      VAProfileNone                   : VAEntrypointVideoProc
      VAProfileNone                   : VAEntrypointStats
      VAProfileMPEG2Simple            : VAEntrypointVLD
      VAProfileMPEG2Main              : VAEntrypointVLD
      VAProfileH264Main               : VAEntrypointVLD
      VAProfileH264Main               : VAEntrypointEncSliceLP
      VAProfileH264High               : VAEntrypointVLD
      VAProfileH264High               : VAEntrypointEncSliceLP
      VAProfileJPEGBaseline           : VAEntrypointVLD
      VAProfileJPEGBaseline           : VAEntrypointEncPicture

root@f7fe7efcfa51:/# ffmpeg -hide_banner -encoders | grep qsv
 V..... av1_qsv              AV1 (Intel Quick Sync Video acceleration) (codec av1)
 V..... h264_qsv             H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (Intel Quick Sync Video acceleration) (codec h264)
 V..... hevc_qsv             HEVC (Intel Quick Sync Video acceleration) (codec hevc)
 V..... mjpeg_qsv            MJPEG (Intel Quick Sync Video acceleration) (codec mjpeg)
 V..... mpeg2_qsv            MPEG-2 video (Intel Quick Sync Video acceleration) (codec mpeg2video)
 V..... vp9_qsv              VP9 video (Intel Quick Sync Video acceleration) (codec vp9)

Me IRL this weekend.

Proxmox Setup Notes

Thu, 07 Aug 2025 22:02:02 -0400

Proxmox installation steps

Overview

This post is a reference for me and anyone else that finds it useful.

I’ve never bothered documenting this before, which leads me to missing stuff every time I set up a new box.

This post will continue to be updated if I add anything new.

Installation

The primary thing to get right here is the storage configuration. Even if it’s a single disk I will typically do ZFS.
- Click on Options. Choose whichever ZFS topology you want.
- Click on Advanced Options. Change Compression from “on” to “lz4”. This is the default anyway, but I like to be sure.
- ashift should be set to 12 or 13. I typically do 12, as this targets 4k physical sectors.
- Make sure that you have the correct disks selected. Any disk you do not want to be part of the boot pool should be set to “do not use”.
For the rest of the setup you can pretty much just click Next Next Next. I do recommend setting the correct time zone and a good DNS server.

Post-Install

If you don’t have an Enterprise subscription we need to disable the enterprise repository and enable the one for no-subscription.
- Click on the individual node (not Datacenter) and go to Updates > Repositories.
- Click on Add and select No Subscription from the dropdown.
- Next, select the repositories labeled “Enterprise” and disable them.
Enable Debian Non-Free-Firmware repository.
- This one can’t be done through the GUI. SSH in or use the web shell.
- Edit the sources
```
vim /etc/apt/sources.list.d/pve-no-enterprise.list`
```
- Add the following repo, replacing $VERSION with your release’s Debian base (bookworm, trixie):
```
https://deb.debian.org/debian $VERSION main non-free-firmware
```
- Install the firmware microcode for your hardware vendor
  - Run apt update
  - apt install intel-microcode OR apt install amd64-microcode
At this point, update the system. Reboot if needed.
Next, lets fix the ZFS configuration.
- Click on the Datacenter tab and go do down to Storage.
- Select your zpool (or pools) and click Edit. You will have to do this for every pool individually.
- Enable thin provisioning.
- Change the Block Size to 64k or 128k. Either is probably fine, but the default 16k is abysmal for VM disks. This setting is the zvol equivalent to the recordsize property on datasets. Having this not match your workload will usually result in massive write amplification. This is why ZFS got a reputation for quickly killing SSDs, but it doesn’t have to if you configure your pool correctly.
Click on the individual node again. Go to System > Network and select any bridges that you’ve created. Set them to be VLAN-aware.
Set up the notification system under Datacenter > Notifications. I will typically have it send me only backup failure notices and warnings about things like failed disks.
I will typically set up a VM template at this stage to save myself a bunch of work later. See links below for options.

Other Notes

If you’re trying to set up a new ZFS storage pool and the disks aren’t showing up in the interface - the Proxmox interface will not show you any disks that have partitions on them.

If you’re expecting a drive to show up and it’s not you can go through the following procedure on the command line.

lsblk to find the path of the drive that you need to wipe.
```
lsblk -o NAME,SIZE,MODEL
```
Wipe out all of the partitions.
```
wipefs -a /dev/sdX
```
The ZFS labels (which contain pool metadata) may survive wipefs as it’s stored on the first few sectors of the drive, so lets make sure that any existing labels are gone. If this drive has never been used in a zpool before, you can skip this step.
```
zpool labelclear /dev/sdX
```
Try creating your pool again.

Proxmox on Sysadmin Tales

Deduplicated Proxmox backups with Proxmox Backup Server and ZFS

Overview

The Details

Issues transferring Windows 11 VM from KVM to Proxmox

Passing through Intel Arc A380 to a Proxmox VM

Proxmox Setup Notes

Proxmox installation steps

Overview

Installation

Post-Install

Other Notes

Links