Proxmox VE 9.2 Snapshot Rollback Test Before Upgrade

Run a Proxmox VE 9.2 snapshot rollback drill on one test VM to verify snapshot, rollback, and config restore paths before upgrade, reducing downtime risk.

Proxmox PulseProxmox Pulse
10 min read
proxmoxsnapshotrollbackqemulvm-thin
Close-up of a small server chassis, drive tray, brass lever, and glass capsule on a dark workbench.

Before you upgrade a Proxmox VE node to 9.2, run a short snapshot drill on one low-risk VM and prove that the snapshot, rollback, and config restore path all still work. The drill takes a few minutes on a test VM, and it gives you a known-good rollback procedure before you touch production workloads.

Key Takeaways

  • Snapshot drill: Prove snapshot creation and rollback on one VM before the upgrade window.
  • Stopped first: Use a stopped snapshot for the first test; live snapshots add guest-state risk.
  • Storage aware: ZFS, LVM-thin, Ceph, and qcow2 support snapshots in different ways.
  • Not a backup: A snapshot lives on the same storage as the VM and does not replace PBS.

Why Test Snapshot Rollback Before Proxmox VE 9.2?

A Proxmox VE upgrade changes packages, kernel modules, QEMU, and sometimes storage behavior. Snapshot rollback exercises several of those paths at once: the storage backend must accept a snapshot, QEMU must accept the restore, and the VM config must come back in a usable state. If you only test after the upgrade, a bad snapshot path becomes an upgrade-day incident.

This drill is also a good companion to the Proxmox VE 9.2 upgrade checklist. The checklist helps you prepare the node; the snapshot drill helps you prepare a recovery action. They solve different problems. A checklist tells you what should work. A rollback drill shows you what happens when you actually press the button.

The concrete outcome is small but valuable: you end the drill with a snapshot name, a restore command, a verified guest state, and a measured time. That is a procedure you can repeat on a production VM before the upgrade, or keep as the first recovery step if the upgraded VM behaves oddly.

What Counts as a Good Test VM?

Pick a VM you can stop for a few minutes. A plain Linux VM is best for the first run. Avoid using a GPU passthrough VM, a clustered HA VM, or the one VM that hosts your DNS server.

  • It has a guest agent installed. If it does not, fix that first with the QEMU Guest Agent check or use SSH for the visible change.
  • It has a small disk, such as a 20 GB to 40 GB virtual disk. Larger disks can make the drill slower without proving more.
  • It has no PCI passthrough and no mounted ISO. These are common reasons a snapshot command fails or behaves differently than expected.
  • It is not managed by HA, or you have placed the HA resource in maintenance mode. Otherwise the cluster may move or restart the VM while you are testing.

Use VM ID 101 as the example. Replace it with your test VM ID.

How to Run the Snapshot Rollback Drill

Step 1: Record the Starting State

Record the starting state so you can tell whether the rollback worked. Note the IP address, running services, and any config you might change later.

qm list
qm status 101

A config excerpt might look like this:

name: drill-01
memory: 2048
cores: 2
net0: virtio=BC:24:11:4A:9F:3C,bridge=vmbr0
scsi0: local-lvm:vm-101-disk-0,size=40G

You do not need to capture every setting. You just need enough context to know whether the VM came back the way you expected.

Step 2: Take a Known-Good Snapshot

For the first drill, stop the VM. A stopped snapshot removes memory state from the test and gives you the cleanest result.

qm stop 101
qm snapshot 101 pre-drill --description "pre-9.2 drill"
qm start 101

If you also want to test live snapshots, take a second snapshot after the VM is running:

qm snapshot 101 pre-drill-live --live 1 --description "pre-9.2 live drill"

Use the stopped snapshot for the main rollback test. Treat the live snapshot as an extra check. Once the stopped snapshot path works, repeat the same rollback test with the live snapshot name if you need to prove live behavior as well.

Step 3: Make a Visible Change

Make two changes: one inside the guest and one in the VM config. The guest change proves the disk state rolls back. The config change proves the VM configuration rolls back.

qm guest exec 101 -- touch /var/tmp/snapshot-drill.txt
qm guest exec 101 -- ls -l /var/tmp/snapshot-drill.txt
qm set 101 --description "changed-after-snapshot"

If the guest agent is missing, use SSH or the console for the file change. The important part is that the change is easy to see.

Step 4: Roll Back and Verify

Stop the VM, restore the snapshot, and start it again. Then verify that the file is gone and the description is back to its old value.

qm stop 101
qm restore 101 pre-drill
qm start 101
qm guest exec 101 -- ls -l /var/tmp/snapshot-drill.txt

The last command should fail because the file no longer exists. That failure is the proof you want. If you also changed the description, check the VM summary or config file to confirm it reverted.

After the test, remove the snapshot when you are sure you do not need it:

qm delete 101 pre-drill --force 1

What Does Rollback Restore?

A snapshot is not only a disk image. A Proxmox VM snapshot stores the disk state, the VM configuration, and, for a live snapshot, the memory state. When you restore, the VM config returns to the version captured by the snapshot. That is why a memory change, network change, or description change made after the snapshot can disappear after rollback.

This matters during upgrade week. You might snapshot a VM, change its boot order for a recovery attempt, then restore it. If the rollback works, the boot order should return to the snapshot value. If it does not, you have a config restore problem, not just a guest filesystem problem.

For the drill, the config file excerpt above is enough. You do not need to test every VM option. Test one visible option and one guest file.

How to Read the Results for Different Storage Types

ZFS

ZFS native snapshots are fast and usually the nicest experience. The snapshot appears as a ZFS snapshot under the VM disk dataset. If your VM disks live on ZFS, run a ZFS pool health check before the upgrade so you are not testing snapshots on a pool that is already degraded.

zfs list -t snapshot -o name,used,refer -s creation

If the pool is healthy and the snapshot command works, ZFS is usually not the part you need to worry about.

LVM-thin

Proxmox uses thin pool snapshots on LVM-thin. The snapshot command is fast, but the thin pool can fill up if snapshots are left around. Watch pool usage during and after the drill.

lvs --noheadings -o lv_name,lv_size,data_percent,metadata_percent

A successful rollback is good. A successful rollback that leaves the thin pool nearly full is still a warning.

Ceph and qcow2

Ceph RBD snapshots are cluster snapshots. They are convenient, but they are not a substitute for backup. A Ceph failure or pool problem can affect both the VM and the snapshot.

qcow2 snapshots use overlay files. They are flexible on file storage, but long snapshot chains can slow I/O and make the storage directory harder to reason about. If you use qcow2, keep the drill simple and do not leave snapshot chains sitting around for weeks.

What Can Go Wrong During the Drill?

The most common failure is a snapshot command that refuses to run. PCI passthrough devices, certain CD-ROM configurations, and unsupported storage types can all stop the test. Start with a simple Linux VM on ZFS or LVM-thin to avoid chasing unrelated problems.

UEFI VMs need one extra check. The EFI disk must be on snapshot-capable storage. If the main disk is on a good storage backend but the EFI disk is not, the snapshot command can fail in a way that surprises you.

The gotcha I keep hitting is the config rollback surprise. After a restore, the VM may come back with older memory, older network settings, or an older description. That is normal, but it can feel wrong if you were expecting the rollback to restore only the guest filesystem.

Live snapshots are also riskier than stopped snapshots. A live snapshot on a running Windows VM can cause time skew or a filesystem check on the next boot. A live snapshot on Linux is often fine, but it is still a runtime state. For the pre-upgrade drill, use the stopped snapshot as the pass/fail test.

HA environments need one more check. If the VM is an HA resource, put the resource in maintenance mode or choose a non-HA VM. Otherwise, the cluster may restart or relocate the VM while you are waiting for the rollback to finish.

How Long Should the Drill Take?

On a 40 GB VM with 8 GB RAM on NVMe-backed ZFS, I have seen a stopped snapshot complete in under 10 seconds and a rollback finish in under a minute. On a 40 GB VM on older HDD-backed LVM-thin storage, the same rollback can take several minutes. The exact number depends on storage, snapshot chain, and whether memory state is included.

Use the time you measure as a budget. If a simple test VM takes more than five minutes to roll back, do not assume production VMs will be faster. Investigate storage, snapshot chains, and disk size before the upgrade window.

When Should You Use a Backup Restore Instead?

A snapshot drill proves a local recovery path. It does not prove your backups are restorable. For that, use a PBS restore test on a spare storage location or a different node. The table below shows where each drill fits.

Drill What it proves Typical time for a 40 GB VM Use before VE 9.2 when
Stopped snapshot rollback Snapshot creation, disk rollback, config rollback 1 to 3 minutes You need a fast confidence check
Live snapshot rollback Runtime snapshot and guest state restore 1 to 5 minutes You need to test running VM snapshots
PBS restore test Backup data can be restored to a usable VM 10 to 30 minutes You need proof independent of the source storage
Host reboot drill Node and VM autostart path 5 to 15 minutes You have already changed host packages

The tradeoff is simple: snapshot rollback is fast, but it shares the VM's storage fate. A PBS restore is slower and requires more planning, but it proves you have an independent copy.

Conclusion

The drill ends with a proven rollback path, a measured time, and a clear understanding of what the snapshot restores. Run it on one test VM before upgrading to Proxmox VE 9.2, then repeat the same steps on a production VM if the test passes. Next, pair the result with a Proxmox VE 9.2 reboot drill so you also know the upgraded node can start the VM cleanly after a host restart.

Share
Proxmox Pulse

Written by

Proxmox Pulse

Sysadmin-driven guides for getting the most out of Proxmox VE in production and homelab environments.

Related Articles

View all →