WZ-IT supports assessment and agreed operations for HA and load distribution through Managed Proxmox. Arrange a discussion.
People searching for “Proxmox DRS” usually mean automatic distribution of running guests. Proxmox uses the Cluster Resource Scheduler (CRS). The dynamic load balancer was introduced with PVE 9.2 on 21 May 2026. As of October 2026.
Contents
- Distinguish HA, CRS and DRS
- The three scheduler modes
- Configure load balancing
- Limits and acceptance
- Operations and related guides
- Sources
Distinguish HA, CRS and DRS
The CRS section in the HA manual addresses placement of HA resources. Recovery after host failure, placement at start and continuous load balancing are different decision triggers. Balanced utilization therefore does not prove that recovery after failure works.
For a VMware replacement, we recommend assessing requirements individually: which guests are HA-managed? Which placement rules are required? Which migrations are permitted? The Proxmox and VMware comparison provides further context. “DRS like VMware” alone is not an acceptance criterion.
The three scheduler modes
The Proxmox developer's explanation distinguishes:
| Mode | Information evaluated |
|---|---|
basic |
Number of active guests |
static-load |
Configured CPU and RAM quotas |
dynamic-load |
Additionally actual CPU and RAM utilization |
The mode affects multiple HA placement decisions. Automatic rebalancing requires static-load or dynamic-load. Dynamic mode also considers node utilization including host services such as Ceph and ZFS. This is not a general optimizer for application response times or storage latency.
Configure load balancing
The documented entry point is Datacenter > Options > Cluster Resource Scheduling. The vendor guide describes scheduling mode and automatic load balancing. Recommended introduction procedure:
- Compare the installed packages and available CRS options with the deployed version. Use at least the PVE 9.2 feature baseline for the dynamic load balancer described here.
- Inventory HA resources, eligible target nodes and affinity rules. Check storage, networks, CPU compatibility and local devices.
- Select
static-loadordynamic-loadaccording to the goal and explicitly enable automatic load balancing. - Assess imbalance threshold, hold duration, rebalancing method and required improvement for the environment. Do not adopt generic values without review.
- Observe migrations and application impact in an approved evaluation scope first, document findings and only then expand the agreed scope.
According to the manual, automatic load balancing issues migrations sequentially and only for HA resources; affinity rules still apply. The auto-rebalance option can exclude individual HA resources. This does not mean automatic migration of every VM in a cluster.
LXC containers cannot be live-migrated. The container documentation describes restart migration instead: stop, move and start again. The application must tolerate that interruption.
Limits and acceptance
“Rebalance on Start” is a separate option. The vendor clarification of 29 July 2026 distinguishes its technology preview status in static-load and dynamic-load from the load balancer. The developer change adding a UI warning confirms that distinction. Do not label all CRS functions as preview or as unconditionally approved.
For acceptance, we recommend separate checks of load distribution, rule compliance, eligible migration and behavior with insufficient capacity. A blocked migration is initially a finding to explain. Inspect rules, device mappings and resource availability instead of removing protective constraints. CPU and RAM sizing remain part of capacity planning.
Operations and related guides
Build an HA cluster covers quorum, fencing and recovery. The cluster update guide describes planned node maintenance. Both complement load balancing and are not replaced by it.
View Managed Proxmox: agree requirements for HA, load distribution and operational responsibility. For a new cluster, see Proxmox setup.
