ZFS Storage Pools for Slicer¶

ZFS is an advanced filesystem that was originally developed at Sun Microsystems. The modern equivalent for Linux was ported as the OpenZFS project and has a license that is incompatible with the Linux Kernel, for that reason, it is kept out of tree, and must be installed separately.

Whilst ZFS can run on a loopback device, this is not recommended and may be unstable. Instead, dedicated either a partition or a drive to running ZFS.

Use ZFS for VM storage¶

Before you can start using zfs storage for your VMs, you'll have to run through the installation steps below.

Let's customise the walkthrough example for ZFS.

1) Change the storage type from image to zfs:

config:
  host_groups:
  - name: vm
-   storage: image
+   storage: zfs
-   storage_size: 20G

See the note below on storage_size.

2) Customise the storage_size

The storage_size field is optional for ZFS.

If not specified, the default size of the base snapshot will be used. A custom size can be given, so long as it is equal to or larger than the base snapshot size.

config:
  host_groups:
  - name: vm
    storage: zfs
+   storage_size: 40G

If the base snapshot size is large for any existing VMs, then you can find its lease and remove it before having it re-created with the new settings for the vzol-snapshotter.

$ sudo ctr -n slicer leases ls
ID            CREATED AT           LABELS 
slicer/k3s-1  2025-09-04T13:53:07Z -

Then, find the lease ID for the VM in question, and delete it. The lease ID is the hostname of the VM i.e.k3s-1.

sudo ctr -n slicer leases rm slicer/k3s-1

To delete all leases:

sudo ctr -n slicer leases ls -q | xargs -n1 sudo ctr -n slicer leases rm

Install packages for ZFS¶

sudo apt update
sudo apt install -y zfsutils-linux

The two commands you are likely to need are:

zpool - create and explore pools to be used by ZFS
zfs - manage ZFS filesystems and snapshots

Create a pool for Slicer¶

Let's say that you have a local NVMe i.e. /dev/nvme1n1.

Within that device, you may have /dev/nvme1n1p1 for your operating system, and some free space at /dev/nvme1n1p2.

Use the following command to enroll that free space into ZFS:

sudo zpool create slicer /dev/nvme1n1p2

ZFS has many different options such as checksuming, compression, deduplication, encryption, and the ability to run in different RAID-like modes across multiple drives.

We recommend that you do the simplest thing first, to get it working before tinkering further.

Create a filesystem for Slicer¶

sudo zfs create slicer/snapshots

Install the zvol-snapshotter for containerd¶

The containerd project already has a ZFS snapshotter, however it is unsuitable for use for VMs, therefore we needed to implement our own snapshotter which can present ZFS volumes as block devices to a microVM.

The zvol-snapshotter can be installed using arkade:

arkade system install zvol-snapshotter \
  --dataset slicer/snapshots

You can specify the size of any VM drive that will be created by the snapshotter using the --size flage, e.g. --size=40G. The default size is 20G.

Configure containerd to enable Zvol snapshotter, edit /etc/containerd/config.toml and add:

[proxy_plugins]
  [proxy_plugins.zvol]
    type = "snapshot"
    address = "/run/containerd-zvol-grpc/containerd-zvol-grpc.sock"

Restart containerd:

sudo systemctl restart containerd

Check if the snapshotter is running OK:

sudo journalctl -u zvol-snapshotter -f

Now, test the Zvol snapshotter:

(
sudo -E ctr images pull --snapshotter zvol docker.io/library/hello-world:latest
sudo -E ctr run --snapshotter zvol docker.io/library/hello-world:latest test
)

Adjust the base snapshot size¶

The base snapshot size can be configured when installing or updating the snapshotter with arkade system install by including the --size flag.

You can also manually edit the snapshotter configuration file.

Edit /etc/containerd-zvol-grpc/config.toml and replace the volume size with the desired value, e.g 40G:

root_path="/var/lib/containerd-zvol-grpc"
dataset="your-zpool/snapshots"
-volume_size="30G"
+volume_size="40G"
fs_type="ext4"

Finally reload and restart the service:

sudo systemctl restart zvol-snapshotter

Footnote on ZFS on a loopback file¶

It is absolutely not recommended to use a loopback file for ZFS for Slicer.

$ fallocate -l 250G zfs.img
$ sudo zpool create slicer $HOME/zfs.img

zpool list
NAME     SIZE  ALLOC   FREE  CKPOINT  EXPANDSZ   FRAG    CAP  DEDUP    HEALTH  ALTROOT
slicer   248G   105K   248G        -         -     0%     0%  1.00x    ONLINE  -

$ zfs list
NAME     USED  AVAIL     REFER  MOUNTPOINT
slicer   105K   240G       24K  /slicer