WZ-IT supports planned update procedures through Proxmox maintenance. Arrange a discussion.
This guide is a recommended runbook for regular package updates in an existing cluster. Update Proxmox covers package sources and installation methods; the major upgrade article covers release changes. As of October 2026.
Contents
- 1. Approve maintenance
- 2. Evacuate one node
- 3. Install packages and reboot
- 4. Accept and return the node
- 5. Approve the next node
- Sources
1. Approve maintenance
Before starting, we recommend a change record with the baseline, node list, responsibilities, backup access and stop criteria. Check whether remaining nodes can accommodate the guests. Assess quorum and storage for the planned outage scenario; the HA cluster guide covers the fundamentals.
The HA documentation distinguishes migration, recovery and shutdown policies. In particular, conditional does not automatically evacuate a node on reboot: HA resources are frozen and then started on the same node. Decide beforehand which guests should migrate and which need a controlled shutdown.
2. Evacuate one node
HA resources have a manual maintenance mode. According to the vendor's Maintenance Mode documentation, the following command submits an asynchronous request; replace NODENAME with the actual node name:
ha-manager crm-command node-maintenance enable NODENAME
ha-manager status
Wait for migrations to finish. The maintenance status alone is insufficient: it may appear before all active services and workers have left the node. Proceed only when no active HA resources or running migration tasks remain.
We recommend handling non-HA guests separately using the inventory. Check local disks, passthrough devices, CPU compatibility and network mappings. Applications requiring interruption need explicit approval; an unavailable live migration must not be assumed successful. The VM migration documentation explains requirements and local resources.
3. Install packages and reboot
Check package sources and read the package plan on the evacuated node. The update documentation specifies the GUI or:
apt update
apt full-upgrade
Stop on unresolved errors. Approve a required reboot only after evacuation is complete and console access is established. Check a newly installed kernel against the running version afterward. Keep manual maintenance mode active until acceptance is complete.
With Ceph, also check its state before and after node maintenance. A separate Ceph major upgrade or blanket use of Ceph flags is outside this runbook; use the respective Ceph upgrade guides.
4. Accept and return the node
Our recommended approval checklist:
| Area | Check before returning the node |
|---|---|
| Packages and boot | No unresolved installation errors, expected kernel running |
| Cluster | Node reachable, quorum present, no unexplained HA errors |
| Data path | Storage, networks and Ceph where applicable checked |
| Guests | Agreed application tests and monitoring checked |
Then explicitly disable manual maintenance mode:
ha-manager crm-command node-maintenance disable NODENAME
ha-manager status
Manual mode survives reboot. Disabling it can trigger return migrations of previously moved resources; wait for and check those movements too. Both behaviors are documented in the HA manual. Do not stop HA services or the watchdog as a shortcut.
5. Approve the next node
Proceed to the next node only with documented acceptance. Interrupt the procedure on quorum loss, unexplained storage faults or failed guest migration. For closure, we recommend comparing package versions across nodes and assigning outstanding tasks to owners.
Discuss Proxmox maintenance: include approvals, backups and acceptance in the agreed operational procedure.
