Dynamic Load Balancer
in Proxmox VE 9.2

No more manual balancing

Until now, balancing load across a Proxmox cluster was the admin's job — they decided which VM to move when a node got busy. Since 9.2 it can be automatic: once enabled, the load balancer watches the cluster and initiates migrations itself, so no node is overloaded while others have headroom.

Decisions based on real load

In dynamic mode the CRS factors in the average CPU and memory usage of active VMs plus their configured quotas. Static mode uses those configured quotas alone. With the balancer enabled in either mode, each HA manager round (about every 10 seconds) checks whether the cluster imbalance exceeds a threshold.

If the imbalance stays above the threshold for a set number of consecutive rounds (the hold duration), the balancer picks the migration that reduces imbalance the most and performs it only when the gain is at least the ha-auto-rebalance-margin setting (10% by default).

Enabling and tuning

• Where — Datacenter → Options → Cluster Resource Scheduling.
• Scheduler mode — ha=static or ha=dynamic sets how HA picks a node at recovery, and at start only with ha-rebalance-on-start=1 (off by default). Automatic migrations start only with ha-auto-rebalance=1 (off by default) and require one of those two modes.
• Parameters — tune the imbalance threshold and hold duration for your cluster to control sensitivity and migration frequency.

What the balancer respects

• HA-managed VMs only — automatic migrations cover resources managed by the HA stack.
• Always affinity-compliant — migrations never break your node and resource affinity rules. Node affinity only prefers the listed nodes by default and may start the VM elsewhere; strict keeps it on those nodes alone. Resource affinity, including anti-affinity, is strict and will not start the VM when the rule cannot be met.