High availability and storage replication

For hosts that are a Proxmox VE cluster, NexoVirt manages the cluster’s own high availability (HA) and storage replication. HA restarts a guest on another node when its node fails; replication copies the guest’s disks to another node on a schedule, so that restart starts from a recent copy. Both run inside Proxmox VE: NexoVirt only reads and writes their configuration.

A single node has neither. Its HA page and the guest’s High availability section show Needs a cluster and nothing else.

All changes need the hosts.manage permission, ask for a fresh sign-in proof (step-up) and are written to the audit log.

The cluster HA page

Open a node of the cluster and choose Actions > High availability. The page shows:

• Cluster status: cluster name, quorum, the active HA manager, the Proxmox VE version and every node with its local resource manager state.
• HA guests: every guest managed by HA with its requested state, current state, node, group or rule, and the max restart / max relocate values.
• HA groups (Proxmox VE 8) or Node affinity rules (Proxmox VE 9): where HA guests may run.

NexoVirt picks the model from the Proxmox VE version: version 9 and newer use rules, older versions use groups.

Groups (Proxmox VE 8)

A group lists nodes with a priority. Higher priority wins; nodes with the same priority share the guests. Only these nodes (restricted) keeps a guest stopped when none of the listed nodes is online. No failback stops HA from moving a guest back when a higher priority node returns. You pick the group per guest on the guest page. Proxmox VE refuses to delete a group that a guest still uses.

Node affinity rules (Proxmox VE 9)

A rule lists nodes with a priority and the HA guests it applies to, so a rule needs at least one HA guest. A guest can be in only one node affinity rule; choosing another rule on the guest page moves it. Only these nodes makes the rule strict. Resource affinity rules (keep guests together or apart) are shown read-only.

HA for a guest

On the guest page open Settings > High availability.

1. Pick the requested state: Started (HA keeps it running), Stopped or Disabled (HA keeps it stopped and stops it now if it runs), Ignored (HA leaves it alone).
2. Optionally pick a group or node affinity rule. Without one the guest may run on any node.
3. Set Max restart (restarts on the same node) and Max relocate (moves to another node), 0 to 10.
4. Click Add to HA. The confirmation names the guest, its HA id (vm:100 / ct:100) and every value that changes.

Remove from HA keeps the guest’s current power state. On Proxmox VE 9 the guest is also removed from its rule; a rule left without guests is deleted by Proxmox VE.

Fencing

Nodes that run HA guests reboot themselves when they lose quorum, so that a guest never runs twice. Plan for at least three votes (three nodes, or two nodes plus a QDevice) before you turn HA on.

Replication

Replication needs ZFS: every disk of the guest must be on a zfspool storage that exists on both nodes. Disks marked replicate=0, CD-ROM drives and container bind mounts are skipped. If any other disk is on a different storage type, the panel explains which disk blocks it and does not create the job.

In the Replication card:

• Add job: target node, schedule and optional rate limit in MB/s. The schedule uses Proxmox VE calendar events: */15 (every 15 minutes, the default), hourly, daily, 02:30, mon..fri 22:00. The first run copies the full disks.
• Edit: schedule, rate limit, comment, or pause the job.
• Run now: starts the job within a minute.
• Delete: removes the job; Proxmox VE removes the replicated disks on the target node.

The table shows the last and next sync, failed runs and the last error.

How it fits with NexoVirt features

Situation	What happens
No quorum	Reads still work; every change is refused until quorum is back.
Migration safety copy or failed-over copy	Cannot be added to HA or replicated.
Assisted failover	Failover restores a PBS backup on another host, or on another member when only one member of a quorate cluster died. HA guests of a dead member are refused while the rest of the cluster is reachable (“Managed by Proxmox HA: Proxmox restarts it on another member.”), so you never get two copies. The HA card warns when a guest is covered by both.
Migration	Refused for HA guests, across hosts and inside the cluster: move them with HA migrate in Proxmox VE, or remove the guest from HA first.
No quorum alert	The Cluster without quorum alert fires and a banner shows on every panel page.
NexoVirt maintenance mode	Only affects NexoVirt placement. HA keeps restarting and moving guests; use the Proxmox VE node maintenance mode to stop that.
Guest delete	Proxmox VE removes the guest’s HA entry and replication jobs together with the guest.

REST API

The same actions are in the API reference: GET /hosts/{id}/ha, GET|POST|DELETE /hosts/{id}/guests/{type}/{vmid}/ha and /hosts/{id}/guests/{type}/{vmid}/replication (list, create, update, delete, run now).