basforge

Search titles, text and the “If it doesn’t work” cases.

    RU

    Pick the Proxmox node name and address during install: they can be changed later, but in several places

    A Proxmox node’s name is the name of the directory holding the configs of all its guests in /etc/pve/nodes/, and its address must match the entry in /etc/hosts, or pve-cluster won’t start and the UI is empty. So set the name and a permanent address in the installer. The address can later be changed in five minutes; the name only by moving configs, and in a cluster only by removing the node from the cluster.

    Proxmox ties more to the node name than plain Debian does:

    /etc/hostname                       pve1
    /etc/hosts                          10.0.0.2 pve1.home.arpa pve1
    /etc/pve/nodes/pve1/qemu-server/    configs of this node's VMs
    /etc/pve/nodes/pve1/lxc/            container configs
    /etc/pve/nodes/pve1/pve-ssl.pem     web UI certificate — for the node's name and address
    /etc/pve/storage.cfg                storages with a node set: nodes pve1

    Changing the address

    On a single node — three places:

    1. /etc/network/interfaces — address and gateway on vmbr0.
    2. /etc/hosts — the address on the line with the node name.
    3. Apply: ifreload -a, then systemctl restart pve-cluster pveproxy.

    The order matters: both files first, then apply. A new address on the network without the new address in /etc/hosts means pve-cluster won’t start (see “If it doesn’t work”). Make the edit from a monitor and keyboard or from the IPMI console: over the network the connection to the host drops halfway. The web UI certificate was also issued for the old address — the browser warns again; pvecm updatecerts --force issues a new certificate.

    Changing the name

    On a single node with no guests — same as the address, plus /etc/hostname, /etc/mailname and a reboot. With guests, moving the configs is added:

    mv /etc/pve/nodes/pve1/qemu-server/*.conf /etc/pve/nodes/pve2/qemu-server/
    mv /etc/pve/nodes/pve1/lxc/*.conf /etc/pve/nodes/pve2/lxc/

    The new node’s directory appears when pve-cluster starts with the new name. Guest disks don’t need moving: they’re on storages, and the configs reference them by storage name, not node name. Storages bound to the node (nodes pve1 in /etc/pve/storage.cfg) get the new name. The node’s graph history in /var/lib/rrdcached/db stays under the old name.

    In a cluster every node and corosync knows the node name, so it isn’t changed this way: the node is removed from the cluster, renamed as a single node and added back. Hence the rule from the install: pick the name and address once (Installing Proxmox).

    If it doesn’t work

    After an address change the UI doesn’t open, and systemctl status pve-cluster says failed

    Cause
    /etc/hosts still has the node name with the old address: pmxcfs looks up its address by name and doesn’t find it on the interfaces.
    Fix
    Change the node’s line in /etc/hosts to the new address and run systemctl restart pve-cluster pveproxy.

    After a name change the UI shows no VMs, and a node with the old name hangs around greyed out

    Cause
    The guest configs stayed in /etc/pve/nodes/<old name>/, and the node now looks for them in the directory with the new name.
    Fix
    Move qemu-server/*.conf and lxc/*.conf from the old directory to the new one, then delete the old directory.

    How to check

    hostname -f                     # pve1.home.arpa
    getent hosts "$(hostname)"      # 10.0.0.2 — the same address as on vmbr0
    ip -br addr show vmbr0          # vmbr0 UP 10.0.0.2/24
    systemctl is-active pve-cluster pveproxy    # active active
    qm list && pct list             # all guests are in place