Virtualization / 2026-11-12
PVE Major Version Upgrade: Decide How You'll Roll Back First
The 7 to 8 upgrade path, the repository changes you must handle in advance, and three ways to roll back if the upgrade fails.
A major PVE upgrade is fundamentally a major Debian upgrade (7 → 8 means bullseye → bookworm). It isn't like a minor version where "one click and you're done" — halfway through, the entire set of repositories and dependencies gets swapped out.
Four Things You Must Do Before Upgrading
- Back up:
vzdump --allto another machine or a NAS, not onto the same disk - Record the version: archive the output of
pveversion -v; if something goes wrong, this is your only baseline for comparison - Check third-party sources: comment out any non-official repositories in
/etc/apt/sources.list.d/first, otherwise you'll get stuck on dependency conflicts - Set aside a window: at least two hours, and don't start at eleven at night
Repository Changes
Going from 7 to 8 means switching sources — every bullseye becomes bookworm:
# /etc/apt/sources.list
deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription
After editing, run apt update first and check for errors. A 404 means the source address is wrong — don't continue.
The Upgrade Process
apt update && apt dist-upgrade -y
Along the way you'll be asked whether to replace config files (/etc/ssh/sshd_config, /etc/network/interfaces, etc.). Read the contents before choosing — pressing Enter blindly will very easily overwrite your custom configuration.
Three Ways to Roll Back
| Method | Prerequisite | Cost |
|---|---|---|
| Snapshot rollback | You took container/VM snapshots before upgrading | Fastest, but it only rolls back the guests, not the host |
| Restore the whole machine | You have a vzdump backup | Half an hour or so |
| Reinstall the host | No backup | Hours, and the guests have to be imported again |
Note: the host itself has no "system snapshot". PVE only snapshots VMs and containers; if the host upgrade fails, all you can do is reinstall plus restore the guests.
Post-Upgrade Verification
pveversion -v # version is right
systemctl --failed # no failed units
ss -lntp | grep -E "8006|3128" # panel and proxy ports are there
Then start any container and confirm storage and networking are both fine. Only when everything passes is the upgrade complete.
One-Line Summary
The difficulty of an upgrade isn't in the commands, it's in whether you have a way back when things break. Do all three — backup, version record, repository cleanup — and a major version upgrade is just ordinary maintenance.
The Real Risks of a Major Version Upgrade
PVE 7→8 is not a "click to upgrade". Three things actually break: the kernel and NIC drivers, storage plugins, VM config compatibility. If either of the first two goes wrong, the machine may come up with no network at all.
So the order has to be: back up first → rehearse next → upgrade production last.
Pre-Upgrade Checklist
# 1. Record the version and repositories
pveversion -v
cat /etc/apt/sources.list
# 2. Back up key configuration (not the VMs, but PVE itself)
tar czf /root/pve-etc-$(date +%F).tgz /etc/pve /etc/network/interfaces /etc/hosts
# 3. Confirm there are no "half-upgraded" packages
apt list --upgradable | head
Item 2 is the one most easily overlooked: your VMs have backups, but the cluster config, storage definitions and VM configs in /etc/pve are not in vzdump.
Two Rollback Paths
| Method | How | Cost |
|---|---|---|
| Snapshot rollback | Take an LVM/ZFS snapshot of the root partition before upgrading | A few minutes, the cleanest option |
| Reinstall + restore | Reinstall 7.x and import the /etc/pve backup | Half an hour or more |
If the root partition is on LVM, use lvcreate -s to make a snapshot first; if the upgrade fails, boot from the snapshot. Without snapshot capability, at minimum keep the /etc/pve tarball on another machine.
Post-Upgrade Verification
pveversion -v # version and kernel
systemctl --failed # any failed units
qm list; pct list # whether all VMs/containers are up
ip a # whether the bridges are OK
The bridge is the key one: if vmbr0 doesn't come up, every VM is "alive but unreachable", and it looks like everything is dead.
A Blunt Truth
The most reliable way to upgrade a standalone PVE is to keep a spare machine. If you don't have one, schedule the upgrade window on a day when you have four uninterrupted hours — don't do it in the middle of the night. If something breaks at midnight, you won't have the energy to finish it the next day.
One-line summary: writing "where things break" and "how to verify" together is more useful than writing only the config — config can be copied, the order of judgment cannot.