Proxmox VE 9.2 Upgrade: SDN VLAN Link Failure Test

Proxmox VE 9.2 upgrade: run a controlled link failure to verify SDN VLANs, Linux bridge recovery, and firewall rules, avoiding surprise network outages.

Proxmox PulseProxmox Pulse
11 min read
Server rack and network switch with fiber cables and a prism in a dark data center aisle.

A controlled link failure before upgrading to Proxmox VE 9.2 is a small, boring test that can save you from a very loud upgrade surprise. It shows whether your SDN VLANs, VLAN-aware Linux bridge, Proxmox firewall rules, and VM NICs actually recover after a path disappears, instead of only looking correct in the GUI. If you are still deciding what else to check, the Proxmox VE 9.2 upgrade checklist for homelabs covers the host side, while this test covers the network path.

Key Takeaways

  • Baseline first: Capture VLAN tags, bridge state, firewall rules, and packet loss before VE 9.2 touches the node.
  • One path down: Fail a secondary NIC or bond slave, not the management path, to test VLAN recovery without locking yourself out.
  • Logs matter: A clean GUI after the test is not proof; ping loss, bridge VLAN tables, and firewall logs are proof.
  • Firewall is part of the test: SSH hardening, fail2ban, and Proxmox firewall rules must survive both the outage and the upgrade.
  • Repeat after VE 9.2: The same 30-second failure should produce the same recovery window, or you have a regression to investigate.

On a lab node running Proxmox VE 9.1.4 before the VE 9.2 upgrade, I had seen an SDN VNet look correct in the GUI while a test VM still behaved like it was on the wrong segment. The GUI had the VNet, the bridge was up, and the VM had a tag, but the real failure path was still unclear.

The first check is simple. Confirm the installed version so you know what baseline you are testing.

pveversion

Then inspect the bridge and VLAN state from the command line.

ip -d link show vmbr0
bridge vlan show

That gives you the raw state underneath the GUI. A link failure adds the missing dimension: what happens when one path disappears?

This matters for SDN VLANs because several layers have to cooperate:

  • The physical NIC or bond slave must fail over cleanly.
  • The VLAN-aware Linux bridge must keep the correct VLAN mappings.
  • The VM NIC must retain the correct tag and firewall attachment.
  • Proxmox firewall rules must not suddenly drop traffic after recovery.
  • SSH, fail2ban, and management access must remain reachable.

A reboot test can show whether services start correctly. A link failure test shows whether the running network stack survives a path change. That is closer to what happens during a VE 9.2 upgrade when packages restart services, bridges are reloaded, and SDN state is re-read.

What should the test VLAN and VM look like?

Pick a VLAN you can afford to interrupt. In my lab, I used VLAN 40 as a low-risk workload VLAN and kept management on a separate VLAN. That separation is one of the most basic network segmentation best practices: do not test workload networking by risking the path you use to reach the host.

Use one Linux bridge for the SDN VLAN zone, then attach the test VM with an explicit VLAN tag and firewall enabled.

# /etc/pve/qemu-server/101.conf
net0: virtio=BC:24:11:AA:BB:40,bridge=vmbr0,tag=40,firewall=1

Confirm the VM configuration before you start.

qm config 101 | grep net0

The test VM needs three things:

  • A static IP address on the VLAN being tested.
  • A reachable gateway on that VLAN.
  • SSH access from a second node, not only from the host under test.

If the test VM is the only way you can reach the node through that VLAN, you are testing one leg too few.

How to prepare the SDN VLAN and firewall rules?

Preparation is where most of the value is. You want the test to fail in a predictable way before VE 9.2 changes anything.

  1. Confirm the SDN VNet exists and is attached to the expected bridge.
pvesh get /cluster/sdn/vnets --output-format json-pretty
  1. Confirm the VM firewall file exists and contains the expected rules.
cat /etc/pve/firewall/101.fw
  1. Add a narrow allow rule for the test. If the VM is on VLAN 40 and you are testing SSH from the same subnet, a rule like this is enough for the test.
IN ACCEPT -source 10.20.40.0/24 -dest 10.20.40.10 -dport 22 -proto tcp
  1. Check the SSH daemon configuration on the test VM.
sshd -T | grep -E '^port |^permitrootlogin |^passwordauthentication '
  1. If fail2ban is part of your SSH hardening, confirm the service and jail are active.
systemctl status fail2ban --no-pager
fail2ban-client status sshd

A small jail definition is often enough for a homelab or staging VLAN.

[sshd]
enabled = true
port = 2222
maxretry = 3
bantime = 10m

If you changed SSH to a non-standard port, make sure the Proxmox firewall rule and the guest firewall both agree. This is one of those details that looks obvious before an upgrade and becomes expensive after one.

For the host itself, keep management access out of the failure path. If the node has one NIC, use an IPMI, iLO, iDRAC, or serial console. If it has two NICs, put management on one path and the test VLAN on the other.

The safest version of this test uses a secondary interface that carries the SDN VLAN, not the management interface. In a lab node with a management NIC and a second NIC attached to the test switch, the failure can be as simple as bringing the second NIC down for 30 seconds.

ip link set dev eno2 down
sleep 30
ip link set dev eno2 up

If the SDN VLAN rides on a bond, fail one slave instead of the bond itself.

cat /proc/net/bonding/bond0 | grep -E 'Slave Interface|MII Status'

Then bring one slave down and back up.

ip link set dev eno2 down
sleep 30
ip link set dev eno2 up

Open a second terminal and watch link state changes while you run the failure.

ip monitor link

From the test VM, start a ping to its gateway before you fail the link.

ping -c 60 -i 0.5 10.20.40.1 | tee /tmp/vlan40-before-92.log

From another node on the same VLAN, ping the test VM during the same window.

ping -c 60 -i 0.5 10.20.40.10 | tee /tmp/vlan40-cross-node-before-92.log

The timing matters. A 30-second outage with half-second ping intervals produces 60 packets, which is enough to see the failure edge and the recovery edge without making the test feel like a production outage. The first log from my test was only 18 KB, but it clearly showed when the VLAN path dropped and when it came back.

What should you capture during the 30-second outage?

Do not rely on the GUI alone. Capture the state before, during, and after the failure.

Check the bridge VLAN table after recovery.

bridge vlan show

Check the VM state on the Proxmox node.

qm status 101

If the guest has the QEMU Guest Agent installed, confirm the guest still reports its interface state cleanly.

qm guest exec 101 -- ip addr show eth0

If the guest agent is not running, see the QEMU Guest Agent check after VE 9.2 upgrade before you treat missing IP information as a network failure.

Check the Proxmox firewall journal for the test window.

journalctl -u pve-firewall --since "1 minute ago" --no-pager

Check fail2ban as well, especially if the test VM uses SSH hardening.

journalctl -u fail2ban --since "1 minute ago" --no-pager

You are looking for three outcomes:

  • Packet loss should be limited to the outage window and the first few packets after recovery.
  • The bridge VLAN table should still show the correct VLAN associations.
  • Firewall logs should not show a sudden flood of unexpected drops from the test VM or management host.

If the ping log shows 40 lost packets and then recovery, that is not automatically bad. If it shows 60 lost packets and the VM never returns, you have a real problem. If it shows intermittent loss for 90 seconds after the interface comes back, you may have a spanning tree, bonding, or bridge reload issue.

How to compare pre-upgrade and post-upgrade results?

The point of the pre-upgrade test is not to prove the network is perfect today. The point is to know what normal looks like before VE 9.2 changes the system.

After the upgrade, repeat the same test with the same interface, the same VLAN, the same ping interval, and the same 30-second failure window.

Failure method What it proves What it misses
Secondary NIC down Host bridge port recovery and VM path recovery Upstream switch port behavior
Bond slave down Linux bonding failover and rejoin Switch-side redundancy
VM NIC down Guest driver and firewall attachment recovery Physical cable or switch failure
Switch port down Full path including switch and host Host-side administrative control
Full node reboot Service startup and SDN reload Quick runtime path failure

The host-side link failure is the safest first test, but it has an honest tradeoff: it tests the Proxmox node more than the whole network. A switch port failure or a second-switch failure is closer to real production, but it is harder to control and easier to disrupt unexpectedly. For a VE 9.2 upgrade, I would run the host-side test first, then repeat it after the upgrade. If the result changes, only then escalate to switch-side testing.

What gotchas show up in real Proxmox SDN VLAN tests?

One experience-based gotcha: the GUI can show an SDN VLAN as attached while the bridge VLAN table is stale. I have seen cases where the configuration looked correct, but the running bridge did not reflect the expected VLAN until the network configuration was reloaded.

ifreload -a

Do that during a maintenance window. Reloading network configuration can briefly interrupt interfaces, and on a remote node that can mean losing your SSH session at the worst possible moment.

Another gotcha is firewall inheritance. A VM can be attached to the right VLAN and still be unreachable because the Proxmox firewall is enabled at one level but not configured where you think it is. A VM config might have firewall enabled, but the datacenter or node firewall policy may be doing more than you expect.

cat /etc/pve/firewall/101.fw

A third gotcha is SSH hardening colliding with the test. If you move SSH to port 2222, enable fail2ban, and test from a host that has been banned by a previous retry storm, the failure looks like a VLAN problem when it is really an SSH access problem.

A fourth gotcha is treating this test as the only network test. It is not a replacement for a reboot test. If you want to verify SDN startup, VLAN tags, and firewall rules after a full service restart, pair this test with the Proxmox VE 9.2 reboot drill to verify SDN and VLANs.

For network segmentation best practices, keep the test simple:

  • Management VLAN separate from workload VLANs.
  • One SDN VLAN zone per bridge where that fits your design.
  • Explicit VLAN tags on VM NICs for test segments.
  • Proxmox firewall enabled where it can log and block, not enabled everywhere by accident.
  • SSH hardening and fail2ban on hosts and reachable admin VMs.
  • A second access path for the node under test.

Conclusion

A controlled link failure before upgrading to Proxmox VE 9.2 gives you a concrete baseline for SDN VLAN recovery, Linux bridge behavior, firewall logging, and VM network reachability. If the same 30-second failure produces the same packet loss and recovery pattern after the upgrade, you have much stronger evidence that the upgrade did not break the running network path. Run the pre-upgrade test now, save the logs, and repeat it after VE 9.2 is installed.

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 →