Proxmox VE 9.2 Live Migration Drill After Upgrading

Run a Proxmox VE 9.2 live migration drill after upgrading to catch CPU, SDN, and storage issues early, then prove VMs can move safely before production returns.

Proxmox PulseProxmox Pulse
12 min read
A gloved hand lifts a small silver server blade from a dark rack with blue light trails.

After you upgrade a cluster to Proxmox VE 9.2, a short live migration drill is the fastest way to prove that virtual machines can still move between nodes. It exposes CPU model mismatches, SDN VLAN problems, and storage path errors while the workloads are still quiet. The result is a small set of fixes you can make before production traffic returns.

Key Takeaways

  • Pick small: Use a 4 to 8 GiB test VM with a 40 to 100 GB disk so the drill stays under a few minutes on shared storage.
  • Check CPU: A host CPU type is fast but can block migration when nodes are not identical.
  • Verify SDN: The bridge and VLAN tag must be active on the target node, not only in the shared configuration.
  • Watch storage: Shared storage moves state quickly; local storage can turn a live migration into a long disk copy.
  • Prove the guest: Confirm the guest agent, IP address, and network interfaces on the target node before calling the drill complete.

Why Run a Migration Drill After Proxmox VE 9.2?

A package upgrade can change more than the version number. QEMU, KVM, storage helpers, network scripts, and SDN behavior can all shift between releases. If you have not finished the Proxmox VE 9.1 to 9.2 upgrade checklist, pause and finish that first.

The reboot drill proves the node comes back. The migration drill proves the virtual machine can leave it. Those are different problems. A node can boot, join the cluster, and still fail to receive a running guest because of CPU compatibility, a missing VLAN bridge, or a storage path that only exists on the source node.

A good drill has a pass or fail answer for each layer. You are not trying to migrate every workload. You are trying to prove the path with a small, disposable VM.

How to Pick a Safe Test VM

Choose a virtual machine that is disposable. Do not use the one that runs your backup server, your monitoring stack, or the only copy of a service you need right now.

A good test VM has these traits:

  • Linux guest, preferably minimal
  • 4 to 8 GiB of memory
  • 40 to 100 GB disk
  • QEMU guest agent installed
  • No GPU passthrough
  • No unusual PCI devices
  • No huge memory balloon or hotplug settings

If you use snapshots heavily, pair this drill with a snapshot rollback test. A VM that can roll back cleanly is a much better migration candidate than one with a long snapshot chain.

Start by listing the virtual machines and checking the one you selected.

qm list
qm status 101
qm config 101

If the test VM is not running, start it before the drill.

qm start 101

Use a VM ID that exists in your cluster. The examples below use VM ID 101, node pve1, and target node pve2. Replace those values with your real names.

How to Check CPU Model Before Migration

CPU compatibility is the first place live migration can fail after an upgrade. Proxmox VE stores the CPU type in the virtual machine configuration. The shared cluster configuration makes that value visible everywhere, but the target node still has to support the CPU model presented to the guest.

A typical configuration file may look like this:

# /etc/pve/qemu-server/101.conf
cpu: host
memory: 8192
net0: virtio=BC:24:11:8A:2B:3C,bridge=vmbr1,tag=101
scsi0: local-lvm:vm-101-disk-0,size=64G

Check the CPU line and the generated QEMU command.

qm config 101 | grep '^cpu'
qm showcmd 101 | grep -E 'cpu|machine'

If the CPU line is missing, Proxmox VE uses its default CPU type for that guest. That may be fine, but you should still know what the target node will receive.

Compare the physical CPUs on the source and target nodes. Run the following on each node.

grep -m1 "model name" /proc/cpuinfo
grep -m1 flags /proc/cpuinfo | tr ' ' '\n' | sort > /tmp/cpu-flags.txt

A matching CPU model name is not enough. Microcode, generation, and feature flags can differ. In one 9.2 lab, a virtual machine using the host CPU type migrated from a newer Xeon host to an older Xeon host and the guest could not start cleanly on the target. The fix was not a cluster-wide flag. It was changing that virtual machine to a common compatibility CPU type.

If the nodes are not identical, use a conservative CPU type for the test.

qm set 101 --cpu qemu64

The tradeoff is straightforward. The host CPU type gives the guest the best access to modern instructions, but it reduces migration flexibility. A common compatibility type is slower in theory, but it is predictable. For a migration drill, predictability wins.

How to Test SDN and VLAN Paths

SDN problems are sneaky after an upgrade because the shared cluster configuration can look correct while the target node is missing the local plumbing. The virtual machine may migrate, show a link inside the guest, and still fail to get a DHCP lease.

Check the network line in the virtual machine configuration.

qm config 101 | grep '^net'

Then verify the bridge and VLAN state on the target node. If your virtual machine uses a different bridge, substitute the real bridge name.

ip -d link show vmbr1
bridge vlan show dev vmbr1

The first command confirms the bridge exists and is up. The second command shows whether the VLAN is actually present on the bridge. If the bridge is not VLAN-aware, or if the VLAN interface was not created after the upgrade, the guest can appear connected while the path is broken.

One gotcha I have seen after network-related upgrades is a target node that kept an old bridge definition while the shared SDN configuration expected a VLAN-aware bridge. The source node worked because its local bridge was already correct. The target node failed quietly. The guest migrated, the firewall rules looked fine, and the VM still could not reach the gateway.

For the drill, do not stop at the Proxmox GUI. Confirm the bridge and VLAN from the shell on the target node.

How to Test Storage Without Moving the Wrong Disk

Storage is the second big failure point. A live migration can be fast if the disk is on shared storage. It can be slow if the disk is local and must be copied online. It can fail completely if the target node cannot see the storage ID used by the virtual machine.

Check the disk paths first.

qm config 101 | grep -E '^(scsi|virtio|sata|ide)'

Then confirm the target node can see the storage ID. If your storage is not local-lvm, substitute the real storage ID.

pvesh get /nodes/pve2/storage/local-lvm --output-format json

The storage must be active and available on the target node. A storage path that only exists on the source node will not make the drill useful.

Storage situation What the migration does Drill expectation
Shared block storage, such as Ceph or iSCSI Copies memory and device state Seconds to tens of seconds on a healthy 10 GbE link
Shared file storage, such as NFS or SMB Copies state and may wait on I/O Seconds to minutes, depending on disk load
Local disk on the source node Copies the disk online while the guest runs Minutes to tens of minutes
Local disk with snapshots Copies a more complex disk chain Longer and less predictable

For a first drill, shared storage is the better test. It isolates CPU and SDN problems from disk-copy time. Once those pass, you can repeat the drill with a local-disk VM to understand your real copy speed.

I have seen a 100 GB local disk migration take just over 14 minutes on a 1 GbE link. That is acceptable for a planned maintenance window, but it is not a good first drill. If your test VM has a large local disk, use a smaller VM first or move the disk to shared storage before testing.

How to Run the Short Live Migration Drill

Before moving anything, save the current configuration so you can compare it later.

qm config 101 > /tmp/101-before-migration.conf

Confirm the virtual machine is running.

qm status 101

Run the live migration from the source node to the target node.

qm migrate 101 pve2 --online

Watch the command output. A healthy shared-storage migration should finish quickly. On a healthy 10 GbE cluster with shared storage, a small 8 GiB VM can move in under 15 seconds. If the command hangs, do not assume it is still healthy. Check the target node status, the storage path, and the guest console.

After the command finishes, confirm the virtual machine is on the target node.

qm list | grep 101

Then save the configuration again and compare.

qm config 101 > /tmp/101-after-migration.conf
diff /tmp/101-before-migration.conf /tmp/101-after-migration.conf

A clean migration should not surprise you with unexpected configuration changes. If the diff shows something you did not expect, investigate before moving on.

How to Verify the Guest on the Target Node

Migration success is not just the Proxmox command returning cleanly. The guest has to be alive, reachable, and usable on the target node.

Use the QEMU guest agent to confirm the guest is responsive.

qm agent 101 ping
qm agent 101 get-host-info
qm agent 101 get-ip-addresses
qm agent 101 get-network-interfaces

If the guest uses DHCP, give it a few seconds before pinging it. The network stack may need a moment after the migration.

ping -c 3 10.0.10.101

Replace the IP address with the guest's real address. For Windows guests, expect a slightly slower network re-registration. If the guest agent is not installed, do not skip this check. The QEMU guest agent check is a good companion drill because backups, shutdowns, and IP reporting all depend on it.

The pass signal is boring: the guest answers, the IP is correct, the network interfaces look normal, and the service you care about still responds.

How to Migrate Back and Repeat

Do not leave the test VM on the target node after the first successful move. Migrate it back to the original node.

qm migrate 101 pve1 --online

Then verify it again.

qm status 101
qm list | grep 101
qm agent 101 ping

Migrating back proves the return path is also healthy. That matters because SDN, storage, and CPU issues are often asymmetric. One node may have a stale bridge, a missing VLAN, or a storage mount that only works under certain conditions.

Repeat the drill with at least one more virtual machine:

  • One Linux VM with a simple bridge network
  • One SDN or VLAN-tagged VM
  • One VM with a larger disk, if you use local storage

If you have GPU passthrough, do not make that your first migration test. GPU passthrough has its own upgrade checks, and a failed migration can leave the guest in a confusing state.

What to Do When the Migration Fails

When the drill fails, do not panic and do not force the migration until you know the layer that broke.

CPU Failure

If the guest cannot start on the target node, check whether the CPU type is too specific to the source node.

qm config 101 | grep '^cpu'
qm showcmd 101 | grep -E 'cpu|machine'

If the virtual machine uses the host CPU type and the nodes are not identical, change it to a common compatibility type.

qm set 101 --cpu qemu64
qm migrate 101 pve2 --online

This is not always the final production setting. It is the diagnostic step that tells you whether CPU compatibility is the blocker.

SDN Failure

If the guest migrates but loses network access, check the target node bridge and VLAN state.

ip -d link show vmbr1
bridge vlan show dev vmbr1

If the VLAN is missing, fix the SDN or bridge configuration on the target node before migrating production workloads back. Do not assume the shared SDN configuration will repair a broken local bridge by itself.

Storage Failure

If the migration fails immediately or the target node cannot see the disk, check the storage ID.

pvesh get /nodes/pve2/storage/local-lvm --output-format json

A common storage gotcha is a storage ID that is active on one node and disabled or missing on another. That can happen after storage package changes, mount changes, or manual cleanup. The virtual machine will not migrate cleanly to a node that cannot see its disk.

Timing Failure

If the migration takes too long, look at storage type and network speed. A local disk copy over a slow link can make a live migration feel like a restore. That is not always a bug, but it is a bad surprise during a maintenance window.

The honest tradeoff is this: online migration with local storage avoids planned downtime, but it can consume network bandwidth and make the drill less useful for large virtual machines. If your goal is to test CPU and SDN, use a small VM on shared storage. If your goal is to test disk copy speed, use a representative local-disk VM and give it enough time.

Conclusion

A live migration drill after Proxmox VE 9.2 turns a successful package upgrade into a proven cluster operation. Your next step is to pick one small test VM, move it from one node to another and back, and record the CPU, SDN, and storage checks that pass. Once that drill is boring, extend it to one SDN-tagged VM and one representative storage setup before production workloads return.

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 →