High availability From Business
A failed VM host, a failed disk, a failed office PC and a failed storage node are different problems with different answers. This page says what Nodekeeper does for each one today, what it needs to do it, and how long it took in our tests.
On this page: VM hosts · By failure · Prerequisites · Out of scope · Plan
VM HA restarts supported VMs (KVM, PVE) on another host with the same IP and MAC; measured 65 s (PVE) to about 4–8.5 min (KVM) in tests.
| Test (October 2026) | Measured |
|---|---|
| Proxmox VE, confirm mode | VM running on the second host 65 s after a one-click confirmation |
| KVM, three hosts, confirm mode | 3 min 50 s in total, including about 40 s of manual confirmation |
| KVM, two hosts | 8 min 26 s in an earlier run; 4 min 21 s after a detection fix, including 74 s of confirmation |
| Proxmox VE, automatic mode | 5 min 56 s from the last heartbeat |
These are the times of these test configurations, not a guarantee. Your times depend on detection settings, host count, storage and network. Full records: product tour, recovery.
| What failed | What Nodekeeper does | What it does not do |
|---|---|---|
| A KVM or Proxmox VE host running HA VMs | Fence, then cold restart on another host with the same IP and MAC | Live migration — memory contents are lost; data already on disk is kept |
| A disk or a storage host | Distributed storage keeps serving from replicas or erasure-coded chunks; you replace the disk and the cluster rebuilds | Protect against deletion or corruption — those are replicated too |
| One room of a two-room cluster | Stretch mode (two rooms plus an arbiter) keeps the cluster available — tested | Remove the need for the arbiter site |
| The site node that serves Windows office PCs over iSCSI | Nothing automatic yet. Linux office PCs map their disks from distributed storage directly and keep running once booted | Windows PCs lose their system disk until the site node is back or replaced |
| An office PC's own hardware | After the old PC is fenced, the user's cloud PC can be started on another compatible PC | Move open programs or unsaved work from the failed PC |
| The Nodekeeper admin node | The storage data plane and running VMs keep working | Control-plane operations wait until the admin node is back |
HA VM disks on distributed storage managed by Nodekeeper.
Enough free CPU and memory on the remaining hosts to start the VMs of a failed one.
A healthy storage monitor quorum and working fencing; without them the controller does not act.
In the tested setup the VM network runs over the mesh (EVPN), so the same address works on the new host.
VM HA, distributed storage and snapshots are part of Business and are not sold as separate add-ons. Licensing is per node.
The trial covers up to 3 nodes; for a larger HA cluster, talk to us before you build it. Related: Distributed storage · Virtualization · Backups