Proxmox VE 9.2 Boot Order Check Before VM Upgrades
Check the Proxmox VE 9.2 boot order before and after an upgrade to catch UEFI, cloud-init, and disk ID mismatches and prevent failed VM boots on upgraded nodes.

On this page
On a Proxmox VE 9.2 upgrade, the VM that used to boot in six seconds can fail before the guest OS starts because the boot order no longer matches the disk, firmware, or cloud-init drive it actually has. A five-minute boot-order check before and after the upgrade catches UEFI/BIOS mismatches, stale cloud-init CD-ROM entries, and changed disk bus IDs while you still have a snapshot and a backup to fall back on.
Key Takeaways
- Check early: Verify the boot order before the upgrade so the VM does not fail on the first boot after the node changes.
- UEFI disks: A UEFI VM usually needs an EFI disk entry that matches its firmware and boot disk.
- Cloud-init drives: Keep cloud-init attached when needed, but remove it from the boot order unless you are booting from it.
- Disk moves: If a disk changes bus or ID, update the boot order to point at the current boot device.
- Snapshot first: Take a snapshot before changing boot order so you can restore the previous state quickly.
Why Boot Order Changes Matter in a VE 9.2 Upgrade
Boot order is one of the least glamorous settings in a Proxmox VM, but it is also one of the first things that can make a guest unusable after an upgrade. The upgrade itself does not usually rewrite a healthy boot order, and that is part of the problem. A VM can pass a CPU compatibility check, survive a storage health check, and still fail because its configuration references a disk that moved, a firmware type that changed, or a CD-ROM entry left over from an install.
In a Proxmox VE 9.2 upgrade, the guest firmware stack, QEMU behavior, and storage presentation can all shift enough to expose a fragile boot order. A legacy BIOS VM that always booted from the first disk may now encounter a second disk that looks bootable. A UEFI VM may have an EFI partition on the main disk but no separate EFI disk, which can work until the firmware variables are recreated or the boot entry is lost. A cloud-init VM may have its cloud-init drive attached as a CD-ROM, and if that CD-ROM sits ahead of the operating system disk, the boot process can pause or fail.
I treat boot order as a pre-upgrade and post-upgrade check, not a postmortem item. The goal is simple: before you move a production VM to the upgraded node, you know exactly what the VM will try to boot from. That gives you a clear place to fix a problem while the VM is still stopped, still snapshotted, and still easy to roll back.
How to Check One VM’s Boot Order Before the Upgrade
Start with a stopped VM if you can. A stopped VM gives you a clean baseline, and you can change the boot order without worrying about a guest that is mid-boot. The first step is to print the parts of the VM configuration that affect booting: the boot line, firmware type, EFI disk, disk buses, and any CD-ROM entries.
qm config 101 | grep -E '^(boot|bios|efidisk0|ide[0-9]|scsi[0-9]|virtio[0-9]|sata[0-9]|usb[0-9]|ostype|agent):|media=cdrom'
You are looking for three things. First, the boot line tells you the current order. Second, the firmware line tells you whether the VM is using UEFI or legacy BIOS. Third, the disk and CD-ROM lines tell you whether the boot order points to devices that actually exist.
To isolate the boot line, use:
qm config 101 | sed -n 's/^boot: //p'
A healthy UEFI VM might show:
order=scsi0;efidisk0
A healthy legacy BIOS VM with one boot disk might show:
order=scsi0
If the boot line is missing, the VM may still boot using QEMU defaults, but that is not the same as having a deliberate boot order. After an upgrade, defaults can be interpreted differently by firmware, especially when multiple disks or CD-ROM devices are present. For a production VM, I want the order written down and explicit.
For firmware, check whether the VM declares UEFI:
qm config 101 | grep '^bios:' || echo 'bios: seabios (default)'
If the output shows the UEFI firmware value, the VM is using UEFI. If it shows the legacy value, or falls back to the default line, the VM is using legacy BIOS. That distinction matters because a UEFI VM usually needs an EFI disk entry, and a legacy BIOS VM does not.
Fix the Boot Order Before You Test the Upgrade
Once you know the firmware type and the real disk ID, set the boot order explicitly. For a UEFI VM whose operating system disk is on the first SCSI controller, use:
qm set 101 --boot "order=scsi0;efidisk0"
For the same VM on legacy BIOS, use:
qm set 101 --boot "order=scsi0"
If the boot disk is virtio, change the disk ID accordingly:
qm set 101 --boot "order=virtio0;efidisk0"
qm set 101 --boot "order=virtio0"
If the boot disk is SATA, use:
qm set 101 --boot "order=sata0;efidisk0"
qm set 101 --boot "order=sata0"
The quotes around the boot order are not decoration. The first time I fixed this in a shell script, I left the semicolon unquoted, and the shell split the command at the semicolon. The first half ran, the second half was treated as a separate command, and the resulting boot order only contained the first disk. That one-character mistake made the VM look fixed while it was still booting from the wrong place.
qm set 101 --boot order=scsi0;efidisk0
That unquoted form is easy to type and easy to misread. The quoted form is boring, and boring is good when you are scripting a cluster upgrade.
qm set 101 --boot "order=scsi0;efidisk0"
How to Audit Boot Order Across Every VM on a Node
A single VM check is useful, but a node audit is better. Proxmox nodes often hold a mix of production VMs, templates, old test VMs, and cloud-init clones that nobody remembers. A short loop gives you a quick inventory of what each VM is configured to boot.
for vmid in $(qm list | awk 'NR>1 {print $1}'); do
echo "VM $vmid"
qm config "$vmid" | grep -E '^(boot|bios|efidisk0|ide[0-9]|scsi[0-9]|virtio[0-9]|sata[0-9]|usb[0-9]|ostype|agent):|media=cdrom'
done
If you only want the boot line for each VM, use this shorter version:
for vmid in $(qm list | awk 'NR>1 {print $1}'); do
boot_order=$(qm config "$vmid" | sed -n 's/^boot: //p')
echo "$vmid boot: ${boot_order:-none}"
done
This output is the place to look for surprises. A VM with no boot line is worth a closer look. A VM with a CD-ROM entry in the boot line is worth a question: is that CD-ROM still attached, and is it meant to be bootable? A UEFI VM with no explicit EFI disk is worth another question: does it rely on a special EFI partition, or was the EFI disk removed during a template cleanup?
When I ran this across a small four-node test cluster before a Proxmox VE 9.2 upgrade, the audit took less than two minutes. The useful part was not the speed; it was the pattern. Three VMs had stale CD-ROM entries, two UEFI VMs had no explicit EFI disk, and one old template still had an install ISO ahead of the boot disk. Those are exactly the kinds of small mismatches that turn an upgrade into a support ticket.
What to Do When UEFI and BIOS VMs Are Mixed
Mixed firmware is common in homelabs and long-lived clusters. You may have a Windows VM using UEFI, a Linux VM created years ago with legacy BIOS, and a cloud-init template that inherited one firmware type but later got cloned into the other. The boot order is where those differences become visible.
For UEFI VMs, the EFI disk stores firmware variables that help the guest find the boot loader. A typical UEFI VM configuration includes a small EFI disk, often only a few megabytes in size, alongside the main operating system disk.
bios: ovmf
efidisk0: local-lvm:vm-101-disk-0,size=4M,efitype=4m
boot: order=scsi0;efidisk0
If the EFI disk is missing, the VM may still boot if the guest has a fallback EFI file on the disk, but that can be fragile. A firmware change, a disk move, or a template clone can make the fallback disappear. In that case, the VM may fail to boot, show an EFI shell, or hang before the guest OS starts.
For legacy BIOS VMs, the EFI disk is usually not needed. If you add an EFI disk to a BIOS VM, it does not automatically turn the VM into a UEFI VM, and it does not necessarily improve booting. The simpler rule is: match the boot order to the firmware type you are actually using.
| Symptom after boot | Likely boot-order cause | Quick check |
|---|---|---|
| No bootable device | Boot line points to a disk ID that no longer exists | Compare the boot line with attached disk IDs |
| Boots to wrong disk | A second disk moved ahead of the operating system disk | Make the primary disk first |
| Drops to EFI shell | UEFI VM is missing an EFI disk or the EFI disk is not in the boot order | Confirm the EFI disk exists for UEFI firmware |
| Long pause before the OS starts | A CD-ROM or cloud-init drive is ahead of the boot disk | Remove unused CD-ROM entries unless booting an installer |
| First boot works, reboot fails | Boot order relies on a temporary rescue or install ISO | Make the operating system disk first after installation |
The table is not meant to be exhaustive. It is meant to keep you from guessing. If the symptom matches the table, fix the boot order first, then retest the VM. If the symptom does not match, you can move on to CPU flags, guest agent, storage, or network checks without wasting time on firmware.
Cloud-Init and Disk Changes That Break Boot Order
Cloud-init is one of the most common reasons a boot order looks harmless until an upgrade makes it not harmless. A cloud-init drive is often attached as a CD-ROM, and the VM configuration may include that drive in the boot order because it was convenient when the VM was first created.
ide2: local:iso/cloudinit.iso,media=cdrom
boot: order=scsi0;efidisk0;ide2
For the first boot, that can be fine. For every boot after that, it can add a few seconds of CD-ROM probing. In one lab VM, a cloud-init CD-ROM ahead of the boot disk added about 12 seconds to the boot sequence and hid the fact that the real boot disk had been detached. The tradeoff is real: keeping the cloud-init drive in the boot order makes first-boot recovery easier, but it makes normal boots slower and can make a missing disk look like a firmware problem.
If you use cloud-init templates, check them before the upgrade, not after. If you keep cloud-init templates, run a cloud-init template audit before the upgrade. The audit helps you see whether templates are carrying stale CD-ROM entries, inconsistent firmware settings, or boot orders that were never cleaned up after the first boot.
Disk changes are the other half of the problem. Moving a disk between storage pools does not always change the device ID. A disk that moves from one storage location to another while staying on the same SCSI controller can keep the same boot reference. But changing the bus changes the ID. If you move a VM disk to ZFS and also change from SCSI to virtio, the boot order must change too.
qm set 101 --boot "order=virtio0;efidisk0"
If a disk was detached during a storage cleanup, the boot order may still point to the old device. The VM then tries to boot from a missing disk before it tries the real one. That can produce a confusing pause, a no-bootable-device message, or a boot into the wrong guest.
Before you upgrade, I also run a quick CPU compatibility check, especially if the cluster includes older hosts or mixed CPU generations. The Proxmox VE 9.2 CPU flags check catches a different class of upgrade failure, but it pairs well with the boot-order check because both are cheap and both can be done while the VM is stopped.
Run a Small Boot Drill Before the Whole Cluster Moves
Once the boot order is corrected, test it in a controlled way. I do not start by upgrading the whole cluster and hoping the VMs come back. I pick one non-critical VM, take a snapshot, change the boot order, start the VM, stop it, and restore the snapshot. That gives me proof that the fix works and a clean way to get back to the previous state.
qm snapshot 101 pre-92-boot-order
qm start 101
qm status 101
qm stop 101
qm restore 101 pre-92-boot-order
For a stopped VM, the whole drill can take under 90 seconds. For a Windows VM with slow disk initialization, it may take a few minutes. The exact time matters less than the sequence: snapshot, boot, observe, stop, restore. If the VM boots correctly after the boot-order change, you have a repeatable fix. If it does not, the snapshot gives you a fast rollback path.
This is also where snapshot discipline pays off. If you have not tested your snapshot rollback path recently, run a snapshot rollback test before you rely on it during an upgrade. A snapshot that fails to restore is not a safety net; it is a false sense of security.
After the upgrade, repeat the boot-order check on a few representative VMs: one UEFI Linux VM, one legacy BIOS VM, one cloud-init VM, and one VM with a moved disk. If those four boot cleanly, you have much better confidence in the rest of the node. If one fails, compare its boot line against the pre-upgrade audit. The answer is usually in the first line of the VM configuration.
Conclusion
A Proxmox VE 9.2 boot-order check takes minutes and prevents a whole class of post-upgrade failures that look like firmware, storage, or cloud-init problems. Check the boot line, match it to the firmware type, remove stale CD-ROM entries, and update it after any disk bus change. Your next step is to audit one node, fix one VM, and run the snapshot-to-boot drill before touching the rest of the cluster.


