A Proxmox VE cluster lets you manage several hosts from one web interface, move VMs between them and restart guests automatically if a node fails. This Proxmox cluster setup guide walks through creating a cluster, joining nodes and enabling high availability (HA).
Before you start
- Two or more Proxmox VE nodes on the same major version, freshly installed or with unique VM IDs. See install Proxmox VE.
- Unique hostnames and static IPs that resolve correctly.
- A reliable low-latency network for corosync, ideally a dedicated link.
- For HA: three votes (three nodes, or two nodes plus a QDevice) and shared or replicated storage.
Create the cluster
On the first node: Datacenter → Cluster → Create Cluster, give it a name and select the corosync link (network). Or:
pvecm create mdm-clusterJoin other nodes
- On the first node click Join Information and copy it.
- On the new node open Datacenter → Cluster → Join Cluster, paste the information, enter the first node's root password and choose the link.
CLI alternative on the joining node:
pvecm add 203.0.113.11Check status:
pvecm statusUnderstand quorum
Each node has one vote; the cluster acts only when more than half the votes are present. With two nodes, losing one means no quorum and guests cannot be started or changed. Add a third node or a lightweight QDevice on another machine:
pvecm qdevice setup 203.0.113.50Add shared storage
Live migration without copying disks and HA restarts need storage all nodes can reach:
- Ceph (built in): see Ceph for virtualization.
- NFS or iSCSI from a NAS/SAN.
- ZFS replication: asynchronous copies between nodes every few minutes; failover may lose recent changes.
Migrate VMs
Right-click a VM → Migrate, choose the target node and tick Online for live migration. Local disks can be migrated too, just more slowly. See KVM live migration for the underlying mechanism.
Enable high availability
- Optionally create an HA group under Datacenter → HA → Groups to prefer certain nodes.
- Go to Datacenter → HA → Resources → Add, select the VM and set the requested state to started.
- If a node dies, after fencing (self-reboot by watchdog) the VM restarts on another node.
Troubleshooting
- Join fails: the joining node has existing guests with conflicting IDs, or hostnames do not resolve.
- "cluster not ready - no quorum?": too few nodes online; restore nodes or add a QDevice.
pvecm expected 1is an emergency override only. - Corosync link flapping: network congestion; move corosync to a dedicated, quiet network.
Frequently asked questions
Can I remove a node from a cluster?
Yes, migrate its guests away, power it off and run pvecm delnode nodename from another node. Do not bring it back online unchanged.
Do I need a cluster for one or two hosts?
Not necessarily. Separate hosts with backups are simpler unless you need central management or HA.
Can nodes be in different data centres?
Corosync needs low latency; stretching across sites is not recommended.