Two-node cluster quorum behavior
Proxmox uses Corosync for cluster state, which requires a quorum — a majority of nodes must agree for the cluster to function. With two nodes, losing one node loses quorum, and the cluster freezes (no VM starts, no migrations).
Why two nodes is inherently fragile
With 2 nodes, you need both online to have quorum (2/2 = 100%). A network blip between nodes causes both to think the other is down. Neither can establish quorum. Both freeze. This is by design — split-brain (two nodes both thinking they’re primary) is worse than a freeze.
Fix: add a QDevice
A QDevice is a lightweight tiebreaker — a third “vote” without requiring a third full Proxmox node. Run it on any always-on Linux machine (a Raspberry Pi, a small VM, your NAS).
# On the QDevice machine
apt install corosync-qnetd
# On any PVE node in the cluster
pvecm qdevice setup <qdevice-ip>
After setup, the cluster has 3 votes. Losing one PVE node leaves 2/3 — still quorum. The cluster keeps running.
What not to do
Don’t set expected_votes: 1 in corosync.conf. This makes a single node always think it has quorum regardless of the other’s state, which enables split-brain and data corruption on shared storage.
Confirming quorum state
pvecm status
Look for Quorate: Yes and Expected votes matching Total votes.