Hetzner Robot

Proxmox VE on Hetzner: install, bridge and additional IPs

Proxmox VE 9 goes on a Hetzner server on top of Debian 13: first installimage with RAID 1 and LVM, then the packages from the official Proxmox repository. The network for virtual machines is routed on the server itself, because Hetzner treats traffic from MAC addresses other than the server’s as a violation.

  • Time: 40 minutes
  • Level: Advanced

In short

  • The order: Rescue → installimage with Debian 13, RAID 1 and LVM → the Proxmox repository → the Proxmox kernel → the proxmox-ve package. The web interface opens at https://server-IP:8006.
  • Do not bridge VMs to the server’s network card: each VM has its own MAC, and Hetzner treats traffic from other MACs as a violation and may block the server.
  • The safe setup is routed: a vmbr0 bridge with no physical port, a route to each additional IP, and the server’s main IP as the gateway inside the VM.
  • The installation wipes the disks. Additional IPs and subnets are ordered through us.
Contents
  1. How the installation works
  2. Install Debian 13 with RAID 1 and LVM
  3. Prepare Debian: password, RAID and server name
  4. Install Proxmox VE 9 from the official repository
  5. Why you must not bridge VMs to the network card
  6. Set up a routed network for additional IPs
  7. Create a VM and set the IP inside it
  8. If you have a subnet instead of single IPs
  9. More VMs than IPs: internet access through NAT
  10. If something does not work
  11. FAQ

You will need

  • The Hetzner Robot login from ABCD and an SSH client: how to connect via SSH.
  • Additional IPs or a subnet for the VMs: we order them for you when you ask in a ticket (how to order additional IPs).
  • 30–40 minutes and a copy of your data: the installation wipes the disks.
  • No server yet? See Hetzner servers: you can choose Proxmox VE in the order window, and we install it before delivering the server.

How the installation works

Hetzner has no official Proxmox image. The installimage menu has Proxmox VE 7, 8 and 9 images under Other, but the whole item is marked !!NO SUPPORT!!: Hetzner does not support these images. So in this article installimage first installs a clean Debian 13: it sets up the disks, the boot loader and the server network itself. Then Proxmox VE 9 is installed on top of Debian from the official repository. This method is described in the Proxmox documentation and in a Hetzner Community tutorial, and it does not need a KVM console.

PuTTY window: the Other images submenu of installimage with Proxmox VE 7 (Debian bullseye), Proxmox VE 8 (Debian bookworm) and Proxmox VE 9 (Debian trixie); the three Proxmox lines are highlighted.
Proxmox images under Other: Hetzner does not support them.

The example below uses a server with two NVMe disks, the main IP 203.0.113.10 in the 203.0.113.0/26 network and three additional IPs: 203.0.113.20, 203.0.113.21 and 203.0.113.22. Use your own addresses. Commands run over SSH as root; copy the blocks whole.

Install Debian 13 with RAID 1 and LVM

installimage completely wipes both disks. If the server has data you need, copy it before you start.

  1. Boot the server into Rescue and connect via SSH

    In Robot, open Serveryour serverRescue, choose Linux, activate Rescue and save the password. Then reboot the server in the Reset tab with a hardware reset. After 2–3 minutes, connect:

    ssh root@203.0.113.10

    These steps are covered in more detail in the article on reinstalling Linux on a Hetzner server.

  2. Run installimage and choose Debian 13

    installimage

    In the menu, choose Debian, then the image with trixie in its name: that is Debian 13. The editor opens with the installation settings.

  3. Check the disks, RAID and server name

    These lines should look like this; change the name in HOSTNAME to your own:

    DRIVE1 /dev/nvme0n1
    DRIVE2 /dev/nvme1n1
    SWRAID 1
    SWRAIDLEVEL 1
    HOSTNAME pve1.example.com

    The server name must be in the name.domain format: Proxmox requires exactly this kind of name.

  4. Replace the partitions with LVM

    Find the lines that start with PART (without # in front) and replace them with these four. The one exception is the PART /boot/efi esp 256M line: if it is there, keep it first and put the four lines after it.

    PART /boot ext3 1024M
    PART swap swap 8G
    PART lvm vg0 all
    LV vg0 root / ext4 100G

    The system takes 100 GB in the vg0 volume group, and later we give the free space in it to virtual machine disks. Leave the other lines of the file unchanged. What each line means is explained in the article installimage: RAID and partitioning.

  5. Save the settings and wait for the installation

    F2 saves the file, F10 closes the editor (on a Mac, Fn+F2 and Fn+F10, or Esc, then 2, and Esc, then 0). Confirm the data deletion and wait for the INSTALLATION COMPLETE message. Then reboot the server:

    reboot

After 2–3 minutes the server boots into Debian 13. Next, prepare it for the Proxmox installation.

Prepare Debian: password, RAID and server name

  1. Log in to the new system

    ssh root@203.0.113.10

    The password is the same as the Rescue one. If SSH says REMOTE HOST IDENTIFICATION HAS CHANGED, run ssh-keygen -R 203.0.113.10 on your computer and connect again: the new system has its own key.

  2. Set the root password

    passwd

    You will log in to the Proxmox web interface with this password. The interface is reachable from the internet, so the password must be long and not used anywhere else.

  3. Check the RAID and the volume group

    cat /proc/mdstat
    vgs

    mdstat should show raid1 arrays made of nvme0n1 and nvme1n1. A resync line means background syncing; you do not need to wait for it to finish. The vgs output should list the vg0 group, with almost the whole disk capacity minus the 100 GB for the system in the VFree column.

  4. Check that the server name points to its IP

    hostname --ip-address

    The output should be 203.0.113.10, possibly with an IPv6 address next to it. If you see 127.0.1.1 or an error instead, the Proxmox pve-cluster service will not start. Fix the /etc/hosts file (with your own name) and check again:

    sed -i '/^127\.0\.1\.1/d' /etc/hosts
    echo "203.0.113.10 pve1.example.com pve1" >> /etc/hosts

    The first command deletes the line with 127.0.1.1, the second adds a line where the server name points to its main IP.

Once hostname --ip-address shows the main IP, Debian is ready for Proxmox.

Install Proxmox VE 9 from the official repository

  1. Add the Proxmox repository

    apt update && apt install -y wget
    wget https://enterprise.proxmox.com/debian/proxmox-archive-keyring-trixie.gpg -O /usr/share/keyrings/proxmox-archive-keyring.gpg
    cat > /etc/apt/sources.list.d/pve-install-repo.sources <<'EOF'
    Types: deb
    URIs: http://download.proxmox.com/debian/pve
    Suites: trixie
    Components: pve-no-subscription
    Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg
    EOF

    The first line installs wget, the second downloads the key that signs the Proxmox packages. The cat … EOF block creates the file for the free pve-no-subscription repository for Debian 13. Copy it whole, from cat through EOF inclusive.

  2. Update the system and install the Proxmox kernel

    apt update && apt full-upgrade -y
    apt install -y proxmox-default-kernel
    reboot

    If a window asking which disks to install GRUB on appears during the installation, select both disks /dev/nvme0n1 and /dev/nvme1n1 with the space bar (the whole disks, not partitions) and press Enter.

  3. Install the Proxmox VE packages

    After 2–3 minutes, connect again and check the kernel:

    uname -r

    The kernel name must end in -pve. Now install Proxmox itself:

    apt install -y proxmox-ve postfix open-iscsi chrony

    In the Postfix configuration window, choose Local only and keep the default name.

  4. Remove the Debian kernel

    apt remove -y linux-image-amd64 'linux-image-6.12*' os-prober
    update-grub
    reboot

    The standard Debian kernel is no longer needed. os-prober looks for systems on all partitions and may add virtual machine disks to the server’s boot menu, so we remove it too. update-grub rebuilds the boot menu.

  5. Open the web interface

    Open https://203.0.113.10:8006 in your browser: https and port 8006 exactly. The browser warns about the certificate, and that is expected: Proxmox created the certificate itself. Open the page through the advanced options of the warning.

    The user is root, in the Realm field choose Linux PAM standard authentication, and the password is the one you set with passwd. Close the No valid subscription window with the OK button: it does not affect how Proxmox works.

    The Proxmox VE login window: empty User name and Password fields, the Realm list set to Linux PAM standard authentication (highlighted), Language English and the Login button.
    Proxmox login: user root, Realm Linux PAM standard authentication, the password from the passwd command.
  6. Create storage for VM disks

    lvcreate -l 95%FREE --thinpool data vg0
    pvesm add lvmthin local-lvm --vgname vg0 --thinpool data --content images,rootdir
    pvesm status

    The first command creates a thin pool data in the free space of vg0, the second adds it to Proxmox as the local-lvm storage. The pvesm status output should list local (for ISO images and backups) and local-lvm (for VM disks), both active.

Proxmox is running. All that is left is to set up the network the virtual machines use to reach the internet.

Why you must not bridge VMs to the network card

Hetzner binds IP addresses to the server’s MAC address. In the usual Proxmox setup (bridged), VMs are bridged directly to the network card, and each one has its own MAC. Hetzner treats traffic from such MACs as a violation: it sends an abuse complaint and may block the server. The complaint comes to us, and we forward it to you (what to do with a complaint).

That is why we build the network as a routed setup. The VMs connect to an internal bridge vmbr0 that is not tied to the network card, and the server itself passes their traffic out from its own MAC. You do not need to order virtual MACs in Robot for this.

When a bridged setup with virtual MACs fits

A single additional IP can get a virtual MAC in Robot: you set that MAC in the network card settings of the VM and bridge the VM to the server’s network card. This works for no more than 6 single IPs per server. Subnets and failover IPs do not work in a bridged setup, only in a routed one. If your Robot login has no button for virtual MACs, message us.

Set up a routed network for additional IPs

Do not touch the main network card in the /etc/network/interfaces file: the Hetzner installer has already configured it. We only append a new bridge to the end of the file.

A mistake in the network settings will cut off SSH access to the server. In the first step you save a copy of the file, and if access is lost, you restore it from Rescue as shown at the end of the article.

  1. Save a copy of the settings and add the vmbr0 bridge

    Copy and paste the whole block, from the cp line to the EOF line. There must be as many post-up ip route add lines as you have additional IPs:

    cp /etc/network/interfaces /root/interfaces.backup
    cat >> /etc/network/interfaces <<'EOF'
    
    auto vmbr0
    iface vmbr0 inet static
        address 203.0.113.10/32
        bridge-ports none
        bridge-stp off
        bridge-fd 0
        post-up echo 1 > /proc/sys/net/ipv4/ip_forward
        post-up echo 1 > /proc/sys/net/ipv4/conf/vmbr0/proxy_arp
        post-up ip route add 203.0.113.20/32 dev vmbr0
        post-up ip route add 203.0.113.21/32 dev vmbr0
        post-up ip route add 203.0.113.22/32 dev vmbr0
    EOF
    LineWhat it does
    auto vmbr0Brings the bridge up when the server boots.
    iface vmbr0 inet staticA bridge with a static IPv4 address.
    address 203.0.113.10/32The server’s main IP on the bridge: it will be the gateway for the VMs. The /32 mask keeps a second route to the server’s network from appearing.
    bridge-ports noneThe bridge is not connected to the network card, so the VMs’ MAC addresses never reach the Hetzner network.
    bridge-stp off, bridge-fd 0Turn off the STP protocol and the delay when the bridge starts: a bridge without ports does not need them.
    post-up echo 1 > …/ip_forwardTurns on packet forwarding between the VMs and the outside network. On a Hetzner installation it is off by default.
    post-up echo 1 > …/proxy_arpThe server answers the VMs’ ARP requests for addresses in its network, so the VMs reach neighboring addresses through it.
    post-up ip route add 203.0.113.20/32 dev vmbr0A route to an additional IP through the bridge. One line per address.
  2. Apply the settings

    ifreload -a

    If the command prints a warning about route add or File exists for the main network card, there is nothing to worry about: this happens when the route already exists. If the reply is command not found, apply the settings with a reboot: reboot.

  3. Check the routes and packet forwarding

    ip route | grep vmbr0
    cat /proc/sys/net/ipv4/ip_forward

    The first command should show three routes: to 203.0.113.20, 203.0.113.21 and 203.0.113.22. While no VM is running, the word linkdown may appear next to them; that is normal. The second command should print 1. If something is missing, reboot the server and check again.

The bridge is ready, and you can now connect virtual machines to it.

Create a VM and set the IP inside it

  1. Upload an ISO image of the system

    In the web interface, select the local storage on the left, then ISO Images. The Upload button uploads an image from your computer, and Download from URL fetches it from a direct link to the ISO. The second way is faster: the server downloads the file itself.

  2. Create a VM with the Create VM button

    The Create VM button is at the top right. In the wizard, set:

    • the Disks tab: Storage local-lvm;
    • the CPU tab: Type host;
    • the Network tab: Bridge vmbr0, Model VirtIO.

    For Windows, choose the Intel E1000 model on the Network tab and the SATA bus on the Disks tab. Then the Windows installer sees the disk and the network without extra drivers.

    The Create VM wizard in the Proxmox VE 9 web interface, Network tab. The Bridge field is empty and outlined in red: this server has no vmbr0 bridge yet. VirtIO (paravirtualized) is selected in the Model field. The Bridge and Model fields are highlighted.
    Choose vmbr0 in Bridge. If the field is empty and outlined in red, as here, the bridge does not exist yet: add it with the steps above first. For Windows, choose Intel E1000 instead of VirtIO.
  3. Set the network manually when installing the system in the VM

    FieldValue in the example
    IP address203.0.113.20; for the second VM 203.0.113.21, for the third 203.0.113.22
    Netmask255.255.255.192; in the Ubuntu installer, the Subnet field: 203.0.113.0/26
    Gateway203.0.113.10, the server’s main IP
    DNSthe addresses from the nameserver lines in the /etc/resolv.conf file on the server: grep nameserver /etc/resolv.conf

    Take the netmask from the IP address details we send you. The gateway in the VM is the server’s main IP, not the gateway from those details (203.0.113.1 in the example): that one works only for the server itself. Use each IP in one VM only.

    If the additional IP is outside the main IP’s network

    In the example, the main IP’s network is 203.0.113.0/26, that is, the addresses from 203.0.113.1 to 203.0.113.62. If an additional IP is outside it, the gateway 203.0.113.10 ends up outside the VM’s network, and the installer may reject it. In that case, give the Linux VM an address with a /32 mask and specify that the gateway is directly reachable. The ip link command shows the name of the network card in the VM; in Proxmox it is usually ens18.

    Debian, the /etc/network/interfaces file inside the VM:

    auto ens18
    iface ens18 inet static
        address 192.0.2.25
        netmask 255.255.255.255
        pointopoint 203.0.113.10
        gateway 203.0.113.10

    Ubuntu, a file in /etc/netplan/ inside the VM:

    network:
      version: 2
      ethernets:
        ens18:
          addresses: [192.0.2.25/32]
          routes:
            - to: default
              via: 203.0.113.10
              on-link: true

    On the server, such an address needs the same kind of route line in vmbr0: post-up ip route add 192.0.2.25/32 dev vmbr0.

  4. Check the internet connection inside the VM

    ping -c 3 one.one.one.one

    If replies come back, the network works. If it says the name cannot be resolved, check the DNS. If the VM has no internet after you change its network settings, shut it down completely with the Stop button and start it again with the Start button.

The other VMs are set up the same way: each gets its own additional IP and the same gateway.

If you have a subnet instead of single IPs

Hetzner routes a subnet to the server’s main IP, so it needs no separate gateway. A separate bridge vmbr1 is more convenient for it: the server takes one address of the subnet and becomes the gateway for the VMs. An example for the 198.51.100.0/29 subnet:

cat >> /etc/network/interfaces <<'EOF'

auto vmbr1
iface vmbr1 inet static
    address 198.51.100.1/29
    bridge-ports none
    bridge-stp off
    bridge-fd 0
    post-up echo 1 > /proc/sys/net/ipv4/ip_forward
EOF
ifreload -a

Connect the VMs to the vmbr1 bridge and give them IPs from 198.51.100.2 to 198.51.100.6, the netmask 255.255.255.248 and the gateway 198.51.100.1. Of the eight addresses in a /29, five are left for VMs: the first is the network address, the last is the broadcast address, and one more is taken by the server.

More VMs than IPs: internet access through NAT

VMs without their own IP can reach the internet through the server’s main IP. They need one more bridge with an internal network. First, find the name of the server’s network card: it comes after the word dev in the output of this command:

ip route show default

In the example it is enp0s31f6; replace it with yours in the block below:

cat >> /etc/network/interfaces <<'EOF'

auto vmbr2
iface vmbr2 inet static
    address 10.10.10.1/24
    bridge-ports none
    bridge-stp off
    bridge-fd 0
    post-up echo 1 > /proc/sys/net/ipv4/ip_forward
    post-up iptables -t nat -A POSTROUTING -s 10.10.10.0/24 -o enp0s31f6 -j MASQUERADE
    post-down iptables -t nat -D POSTROUTING -s 10.10.10.0/24 -o enp0s31f6 -j MASQUERADE
EOF
ifreload -a

The iptables lines replace the VM’s internal address with the server’s main IP when traffic leaves the server. In the VMs on the vmbr2 bridge, set IPs from 10.10.10.2 to 10.10.10.254, the netmask 255.255.255.0, the gateway 10.10.10.1 and DNS from the server’s /etc/resolv.conf. These VMs cannot be reached from outside until you forward ports to them.

If something does not work

What you seeWhat to do
The server does not respond over SSH after the installation or a rebootWait 5 minutes. If there is no access for more than 10 minutes, activate Rescue and reboot the server; after that you can repeat the installation. If that does not help, message us: we will request a KVM console to see the server’s screen (about the KVM console).
installimage shows an errorCheck that the PART and LV lines are written exactly as in the example, with no extra characters, and that SWRAID 1 is set. If the error repeats, save debug.txt and send it to us together with the error text.
The web interface does not openThe address must be https://203.0.113.10:8006. Check the services with systemctl status pveproxy pve-cluster: both should be active (running). If pve-cluster does not start, the server name is almost always the cause: see the name check. If the Robot firewall is turned on for the server, allow incoming TCP port 8006 in it or turn the firewall off for a while.
The web interface rejects the passwordThe user is root, the Realm is Linux PAM standard authentication, not Proxmox VE authentication server. You can set the password again with passwd.
apt update returns a 401 error for enterprise.proxmox.comThis is the paid repository, and you do not need it. Turn it off: mv /etc/apt/sources.list.d/pve-enterprise.sources /etc/apt/sources.list.d/ceph.sources /root/ 2>/dev/null; apt update
When you start a VM: KVM virtualisation configured, but not availableRun grep -cE 'vmx|svm' /proc/cpuinfo. If the output is 0, hardware virtualization is turned off in the BIOS. Message us: we will ask Hetzner to turn it on.
In the Create VM wizard, the Bridge field is empty and outlined in redThe server has no vmbr0 bridge yet. Add it with the routed network steps, apply the settings with ifreload -a and open the wizard again.
The VM has no internetCheck in this order: on the server, ip route | grep vmbr0 shows routes to all your IPs, and cat /proc/sys/net/ipv4/ip_forward prints 1; in the VM settings (Hardware, Network Device), bridge=vmbr0 is set; inside the VM, the gateway is the server’s main IP, and the netmask and IP are correct; after the network change, the VM was shut down completely with Stop and started again.
A Hetzner complaint about foreign MAC addresses arrivedSome VM is bridged directly to the server’s network card. Check that the bridges in /etc/network/interfaces have bridge-ports none and that all VMs are connected only to them.

If you lose access to the server after setting up the network, activate Rescue and reboot the server, as at the start of the article. In Rescue, restore the saved copy of the settings and reboot:

mdadm --assemble --scan
vgchange -ay vg0
mount /dev/vg0/root /mnt
cp /mnt/root/interfaces.backup /mnt/etc/network/interfaces
umount /mnt
reboot

The commands assemble the RAID arrays, activate the volume group, mount the system root and restore the network file from the copy. If you get stuck at a step, send us the step number and the output of the last command as text.

FAQ

Can I install Proxmox from an ISO through the KVM console?
You can, but Hetzner does not guarantee that an ISO installation will work, and the console is free for only 3 hours. Installing on top of Debian needs no console. ESXi and other systems that are not among Hetzner’s images are also installed from an ISO through the KVM console: about the KVM console.
Do I need to order virtual MAC addresses?
Not for the routed setup in this article: all VM traffic leaves from the server’s MAC. A virtual MAC is needed only in a bridged setup, and only for single IPs.
Proxmox shows a No valid subscription window. Do I need a subscription?
No. Proxmox works fully without a subscription, and updates come from the free pve-no-subscription repository added in this guide. A subscription gives access to the paid enterprise repository and to Proxmox support.
How do I give my virtual machines IPv6?
A Hetzner server has its own /64 IPv6 subnet, but the setup in this article configures IPv4 only. If your VMs need IPv6, write to us in a ticket: we will send you a configuration for your server.

Did this guide help?

Questions?