On 5 August 2025, Proxmox Server Solutions released Proxmox VE 9.0. The new major version is based on Debian 13 “Trixie” – whose official release is planned for 9 August, that is, tomorrow – and it brings several features that users coming from VMware have long missed: snapshots on shared LVM storage, and finer control over the placement of high availability (HA) machines.
Below we go through the most important new features, and what to watch out for if you are upgrading a Proxmox VE 8 server or cluster.
What’s under the hood of Proxmox VE 9.0?
| Component | Proxmox VE 8.4 | Proxmox VE 9.0 |
|---|---|---|
| Base system | Debian 12 “Bookworm” | Debian 13 “Trixie” |
| Kernel (default) | 6.8.12 (6.14 optional) | 6.14.8-2 |
| QEMU | 9.2.0 | 10.0.2 |
| LXC | 6.0.0 | 6.0.4 |
| ZFS | 2.2.7 | 2.3.3 |
| Ceph | Squid 19.2.1 / Reef 18.2.4 | Squid 19.2.3 |
You can find the detailed list of changes on the Proxmox Roadmap page.
The most important new features
Snapshots on thick LVM storage
Until now, in Proxmox VE you couldn’t take VM snapshots on shared LVM storage attached over iSCSI or Fibre Channel. For many companies using storage arrays, this was an obstacle to moving away from VMware. In 9.0, with a new storage setting, snapshots work as a “volume chain”: when you take a snapshot, the current disk state stays under the snapshot’s name, and a new volume starts that stores only the changes. For now the feature is a technology preview, so use it with care on production systems.
HA rules: node and resource affinity
Until now, the placement of high availability machines could be controlled with HA groups. 9.0 replaces these with affinity rules, and converts existing groups automatically:
- a node affinity rule says which nodes a resource can run on,
- a resource affinity rule says that certain machines should always run together (for example an application and its database, for low latency), or always on separate nodes (for example two redundant DNS servers or domain controllers).
SDN fabrics
Software-defined networking (SDN) can now also build routed networks (“fabrics”) between nodes, based on OpenFabric and OSPF. This is mainly interesting for larger clusters: for example for a full-mesh Ceph network or underneath EVPN-based virtual networks.
ZFS RAIDZ expansion
With ZFS 2.3, a new disk can now be added to an existing RAIDZ vdev with minimal downtime. Until now this required rebuilding the pool, which was a serious obstacle in a small, single-server environment.
New mobile interface
When opened on a phone, the web interface automatically switches to a new mobile view written in Rust (with the Yew framework). It gives an overview of machines, tasks and storage, and lets you start or stop a virtual machine – useful when all you have to hand at night is your phone.
What is removed or changed
Before you upgrade, go through the known issues and breaking changes in the release notes. The most important ones:
- GlusterFS support has been removed. If you use GlusterFS storage, move the machines off it first.
- cgroup v2 only. The old cgroup v1 environment can no longer be enabled, so containers using a very old systemd (for example CentOS 7 or Ubuntu 16.04) are not supported.
- The
maxfilesbackup setting has been removed; use the retention (prune-backups) settings instead. - Network interface names may change because of the new kernel. If you can only reach a server over the network, have IPMI or a physical console to hand.
- Permissions: the
VM.Monitorprivilege has been removed, and creating a privileged container now requiresSys.Modify.
Upgrading from Proxmox VE 8 to 9
Proxmox has published a detailed guide for the upgrade. The essential steps:
- Backup. Have a recent, tested backup of every VM and container. Also back up the contents of
/etc/pveand the network settings. - Upgrade to the latest 8.4. The
pve-managerpackage must be at least 8.4.1. - Ceph Squid. In a hyperconverged Ceph cluster, Ceph has to be upgraded to 19.2 Squid first; you can’t move on directly from Quincy or Reef.
- Run the checker. The
pve8to9script doesn’t change anything; it only warns about possible problems. Run it before, during and after the upgrade:
pveversion
pve8to9 --full
- Switch the repositories to Trixie. In the Debian and Proxmox repositories, replace
bookwormwithtrixie. Proxmox VE 9 now uses the new deb822-format.sourcesfiles; for example, for the no-subscription repository:
cat > /etc/apt/sources.list.d/proxmox.sources << EOF
Types: deb
URIs: http://download.proxmox.com/debian/pve
Suites: trixie
Components: pve-no-subscription
Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg
EOF
Then delete the old bookworm Proxmox entries so that there are no duplicate repositories.
- Upgrade and reboot:
apt update
apt dist-upgrade
Run the upgrade from a console, from IPMI, or at least from a tmux/screen session, not from the web interface’s console. In a cluster, go node by node, and migrate the important machines off the node being upgraded first.
LVM autoactivation
From 9.0, new logical volumes are created without autoactivation, because on shared LVM storage it could cause errors when creating and migrating machines. For existing volumes, a migration script changes the setting (/usr/share/pve-manager/migrations/pve-lvm-disable-autoactivation); with shared LVM, Proxmox explicitly recommends running it, and pve8to9 also tells you if it is needed.
Is it worth upgrading now?
According to the Proxmox FAQ, Proxmox VE 8 is supported until August 2026, so there is no need to rush. We recommend the following:
- For new installations, it is worth starting with 9.0.
- For a single server, you can safely upgrade after the first point release, with a tested backup to fall back on.
- For clusters and Ceph, try it in a test environment first, and plan a maintenance window.
- With third-party backup software, check whether the vendor already supports the new version.
We wrote about the release of Proxmox VE 8 two years ago, and we summarised why we think it is a good choice for smaller companies in this article. Since Broadcom’s licensing change, many have been moving from VMware to Proxmox; the snapshot and HA features in 9.0 make this even easier.
We help with installing, upgrading and running Proxmox servers and clusters; you can read about this on the Infrastructure and operations page.