Proxmox 9.2 adds dynamic load balancer & networking
Mon, 25th May 2026 (Today)
Proxmox has released Proxmox Virtual Environment 9.2, adding a dynamic load balancer to its virtualisation platform.
The update also expands software-defined networking features and adds controls for managing custom CPU models across clusters.
The load balancer is one of the main changes in the release. It works with the cluster resource scheduler in a new dynamic mode that takes live node and guest resource use into account when deciding where workloads should run.
In practical terms, the system can migrate guests managed by the high-availability stack when it detects imbalance between cluster nodes. The feature respects user-defined high-availability rules while allowing administrators to adjust how aggressively the balancing mechanism responds.
The addition addresses a familiar issue in clustered virtual infrastructure, where uneven distribution of virtual machines can leave some hosts heavily loaded while others remain underused. By automating part of that process, the release aims to reduce manual intervention in day-to-day cluster management.
Network changes
Version 9.2 also broadens the platform's networking options. Native support for WireGuard and BGP has been added to the software-defined networking stack, alongside route maps and prefix lists for more detailed control of route redistribution in BGP and EVPN environments.
Other networking additions include route redistribution for OSPF fabrics, extra configuration options for EVPN controllers, and IPv6 underlay support for EVPN. Together, these changes extend the platform's reach into more complex network topologies used in larger virtualised environments.
CPU controls
The release also adds a new management interface for custom CPU models. Administrators can create, edit, and remove custom CPU profiles through the web interface in the datacentre section, rather than relying on lower-level configuration work.
The platform now includes an integrated CPU flags selector that shows supported flags across cluster nodes. That visibility is intended to help identify compatibility issues before administrators apply CPU settings to virtual machines that require a specific instruction set or a consistent cluster-wide configuration.
For organisations running mixed hardware estates, CPU compatibility can limit migration and workload placement. A clearer view of supported flags across nodes could help reduce misconfiguration during deployment and maintenance.
Maintenance workflow
Version 9.2 introduces a cluster-wide option to arm and disarm the high-availability manager. The feature is designed for maintenance windows, when administrators may want to suspend automated high-availability actions temporarily.
Disarming the high-availability manager can prevent actions such as node fencing during planned work. Resource states are preserved so that high-availability resources can return to their previous state and placement once maintenance is complete.
The change addresses another operational problem in clustered systems: balancing the need for resilience with the need to carry out maintenance without triggering unnecessary recovery behaviour. A controlled pause in automation can simplify scheduled interventions on infrastructure that would otherwise react to temporary service changes.
Underlying stack
The release is based on Debian 13.5 Trixie and uses Linux kernel 7.0 as the new stable default. It also includes updated versions of QEMU 11.0, LXC 7.0, and ZFS 2.4.
On the storage side, Ceph Tentacle 20.2 is now offered as a stable option alongside Ceph Squid 19.2. These updates place the release on a newer underlying software base for users running virtual machines, containers, and storage services under the same management layer.
Proxmox positions the platform as an open-source virtualisation environment that combines KVM virtualisation, LXC containers, storage, and networking tools in one system. The software is used on more than 2 million hosts globally and is supported by a community of more than 225,000 members.
Support contracts for enterprise users start at EUR 120 per year per CPU. Users can deploy the new version from a full installation image, upgrade through the APT package management system, or install it on top of an existing Debian setup.