Proxmox VE 9.2 Is Out — Here’s Everything That Changed
Dynamic load balancing, WireGuard SDN, Ceph Tentacle 20.2 FastEC, HA maintenance mode, and Linux Kernel 7.0 — released 21 May 2026.
Proxmox VE 9.2 landed on 21 May 2026, and it is a genuinely useful release. Not the kind that just refreshes version numbers — this one addresses real operational pain points that anyone running Proxmox clusters at scale will recognise immediately. Dynamic load balancing, WireGuard baked into the SDN stack, a proper HA maintenance mode, a major Ceph storage upgrade in Tentacle 20.2, and a platform bump to Linux Kernel 7.0 round out a release that makes day-to-day cluster management measurably smarter.
01 / Dynamic Load Balancer — Your Cluster Finally Manages Itself
The headline feature in 9.2 is a long-overdue evolution of the Cluster Resource Scheduler. Previously, Proxmox’s CRS was reactive — it helped decide where to place a new VM at boot time. If a node ran hot hours later because workloads shifted, manual live migration was the only answer.
Version 9.2 makes the CRS dynamic. It now monitors real-time CPU and memory pressure across every node and will automatically migrate HA-managed guests when workload distribution falls out of balance. Configurable thresholds mean you set the rules it plays by — you are not handing the keys to an unchecked algorithm.
02 / WireGuard SDN Fabric — Encrypted Multi-Site Networking
Software-defined networking in Proxmox takes another step in 9.2 with WireGuard as a native fabric protocol option. Integrating it directly into the SDN stack means you can build encrypted overlays between Proxmox nodes across separate sites — data centres, colocation facilities, cloud egress points — without a third-party VPN appliance.
What this unlocks in practice:
- Stretched clusters across geographically separated locations
- Disaster recovery labs connected back to production over encrypted tunnels
- Multi-site VM migration without exposing live traffic on unencrypted networks
For Australian businesses with infrastructure split between Sydney and Melbourne, or a primary data centre and a remote DR site, WireGuard SDN fabric makes multi-site Proxmox genuinely viable without the configuration complexity that traditionally came with it.
03 / Enhanced BGP & EVPN Filtering — Routing Control for Complex Environments
This one is for the network engineers. Proxmox VE 9.2 introduces route maps and prefix lists for BGP and EVPN filtering — the kind of granular routing control that has long been standard in purpose-built network operating systems but was previously absent from the Proxmox SDN stack.
The release also adds route redistribution for OSPF fabrics, additional configuration options for EVPN controllers, and IPv6 underlay support for EVPN. For teams running Kubernetes on Proxmox — an increasingly common configuration — the BGP and EVPN improvements are particularly relevant.
04 / Custom CPU Model Management — Solving the Live Migration Headache
Anyone who has attempted a live migration between Proxmox nodes with mismatched CPU generations knows the scenario: the migration fails because a CPU feature available on the source host does not exist on the destination.
Proxmox VE 9.2 adds a dedicated custom CPU model management interface directly in the Datacenter section of the web console. Administrators can now create, edit, and remove custom CPU profiles through the GUI — the interface displays cluster-wide compatibility so you can see at a glance which nodes support a given profile before assigning it to any VMs.
05 / HA Arm/Disarm — Maintenance Windows Without the Drama
The number of times we have seen HA do exactly the wrong thing during a planned node reboot — fence a perfectly healthy host because the timing lined up with a heartbeat miss — is genuinely embarrassing. The arm/disarm feature in 9.2 is one of those things that should have existed three major versions ago.
PIP Engineering Team
Version 9.2 adds an arm/disarm control directly in the Datacenter HA configuration. Before planned maintenance, you disarm — the entire HA stack suspends cluster-wide. You complete your work. You re-arm, and HA resource states are restored exactly as they were. No manual reconfiguration, no post-maintenance cleanup of HA assignments.
06 / UEFI 2023 Certificate Tools — Getting Ahead of Secure Boot Rotation
Secure Boot certificate management has quietly become a real operational concern as Microsoft transitions from its 2011 UEFI signing certificates to the 2023 certificates. Systems not updated to trust the new certificates will fail to boot from Microsoft-signed bootloaders once the legacy certificates are eventually revoked.
Proxmox VE 9.2 exposes a GUI workflow for enrolling UEFI 2023 certificates across VMs — without requiring you to manually manipulate EFI variable stores. For environments running Windows Server VMs with Secure Boot enabled, this is a practical compliance and continuity tool.
07 / Platform Stack Upgrades — A Full Refresh Underneath
| Component | Version in 9.2 |
|---|---|
| Base OS | Debian 13.5 “Trixie” |
| Linux Kernel | 7.0 (new stable default) |
| QEMU | 11.0 |
| LXC | 7.0 |
| ZFS | 2.4 |
| Ceph (default) | Tentacle 20.2 |
| Ceph (available) | Squid 19.2 |
Linux Kernel 7.0 is the most significant underlying change — improved hardware support for current-generation AMD and Intel platforms, and networking and storage subsystem improvements that amplify the SDN work delivered in the same release. QEMU 11.0, ZFS 2.4, and Ceph Tentacle 20.2 complete a platform refresh that keeps over two million global Proxmox hosts on a well-supported foundation.
08 / Ceph Tentacle 20.2 — A Storage Upgrade Worth Its Own Section
Ceph Tentacle is not a routine dot release. It is a major version step from Squid 19, and several of its changes have direct, tangible impact on anyone using Ceph-backed storage in their Proxmox cluster.
FastEC — Erasure Coding Finally Performs
Erasure coding has always been Ceph’s most space-efficient storage option, but its I/O performance penalty made it unattractive for most block storage workloads. Tentacle introduces FastEC — a new implementation that adds support for partial reads and partial writes, dramatically reducing the read-modify-write overhead that made erasure coding painful on RBD. Erasure-coded pools become viable for VM disk storage in a way they simply were not before.
allow_ec_optimizations. Existing pools are not changed during upgrade. The default erasure coding plugin has also changed from Jerasure (unmaintained) to Intel ISA-L.Instant RBD Live Migration
Images can now be migrated between Ceph clusters without downtime, and can be imported from a wide variety of external sources via NBD stream support. This opens the door to live cluster-to-cluster migrations and DR scenarios that previously required downtime or complex scripted workarounds.
Additional Tentacle Improvements
- BlueStore WAL improvements — reduced I/O latency for write-heavy workloads (databases, backup targets)
- Faster scrubs and bucket listing — all Ceph components moved to a faster OMAP iteration interface
- Native SMB file sharing — Samba-backed SMB shares integrated directly with CephFS, no separate Samba server needed
- Pool availability monitoring — new per-pool data availability tracking via
ceph osd pool availability-status
Should You Upgrade?
Proxmox VE 9.2 upgrades cleanly from 9.1 via the standard apt update path. If you are currently on Proxmox 8.x, a major version upgrade process applies — validate in a non-production environment before proceeding.
For most production clusters, the answer is yes. The dynamic load balancer and HA arm/disarm features alone justify the upgrade for anyone managing multi-node HA clusters. The WireGuard SDN addition makes a compelling case for any business considering multi-site infrastructure expansion. And if your cluster runs Ceph-backed storage, FastEC and the RBD live migration improvements are meaningful changes to how that storage layer performs.
PIP manages and supports Proxmox in Sydney
Whether you need help planning and executing an upgrade to 9.2, want to explore whether Proxmox makes sense as a VMware alternative, or need ongoing support for an existing deployment — the PIP team is ready.