Cloud-init on Centos8 ( set-hostname not working )

[him]

I have a problem with my Centos8 template
I created a template based on the cloud image provided by centos (official)
as a provider we give as hostname the same name as the vm but in this case (centos8 temp) the hostname of the vms deployed via centos 8 template keeps the previous hostname Localhost.
on other templates like debian or ubuntu the hostname is defined normally
I have tried several cloud image centos 8 & 9 and I have the same problem

pve version 6.4-14
cloud-init version 23

 

[bbgeek17]

a bit more information would be helpful:

I created a template based on the cloud image provided by centos (official)

does this mean its a non-modified cloud image or have you customized it?

as a provider we give as hostname the same name as the vm

Do you use built-in PVE cloud-init functionality or do you utilize custom cloud-init?
If former provide a "qm config [vmid]" if latter, add your custom cloud-init file.

There is also this thread https://forum.proxmox.com/threads/setting-host-name-via-cloud-init.45525 although you appear to have opposite problem OS wise.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[him]

does this mean its a non-modified cloud image or have you customized it?

I have tested both solutions:
-deploy a ready-made cloud image
-creat my own vm template on clean install

Utilisez-vous la fonctionnalité cloud-init PVE intégrée ou utilisez-vous cloud-init personnalisé ?
Si le premier fournit un "qm config [vmid]", si ce dernier, ajoutez votre fichier cloud-init personnalisé.

I don't understand the meaning of your question but I know that cloud-init was integrated in the pve since the beginning we use whmcs modul to fill the cloud init directory

 

[him]

update

when I specify the DNS domain by putting ex: hostname
the hostname is defined and the etc/hostname is overwritten and contains the string we defined ex : (hostname.hostname)

 

[bbgeek17]

I don't understand the meaning of your question but I know that cloud-init was integrated in the pve since the beginning we use whmcs modul to fill the cloud init directory

Proxmox implements a reduced set of Cloud-Init internally: https://pve.proxmox.com/pve-docs/qm.1.html#:~:text=Cloud-Init specific Options

You also have an option to use "cicustom" where you can supply a full blown cloud-init file of any supported format.
I dont know what whmcs does, it could be a conditional mixture of both.

I ran a quick test for you using "CentOS-8-GenericCloud-8.4.2105-20210603.0.x86_64.qcow2".

Using integrated cloudinit:

Name set to "vm3000" results in hostname not set properly. You can examine /etc/hosts and notice that vm3000 is there, but not in /etc/hostname

qm config 3000
cipassword: **********
ciuser: centos
ipconfig0: ip=dhcp
name: vm3000
scsi0: bb-nvme:vm-3000-disk-0,size=10G
scsi1: bb-nvme:vm-3000-cloudinit,media=cdrom

more /etc/hostname
127.0.0.1 localhost vm3000

Setting the name to FQDN results in a more expected config:

qm config 3000
cipassword: **********
ciuser: centos
ipconfig0: ip=dhcp
name: vm3000.blockbridge.com
scsi0: bb-nvme:vm-3000-disk-0,size=10G
scsi1: bb-nvme:vm-3000-cloudinit,media=cdrom

more /etc/hostname
127.0.0.1 vm3000.blockbridge.com vm3000

However, when using cicustom:

cicustom: meta=nfs:snippets/cirros.meta.3000,user=nfs:snippets/centos.user

There is no need for FQDN in hostname.

root@vm3000 centos]# mkdir /mnt/1
[root@vm3000 centos]# mount /dev/sr0 /mnt/1
mount: /mnt/1: WARNING: device write-protected, mounted read-only.
[root@vm3000 centos]# cd /mnt/1
[root@vm3000 1]# ls -al
total 5
drwxr-xr-x. 2 root root 2048 May 15 09:49 .
drwxr-xr-x. 3 root root   15 May 15 09:49 ..
-rw-r--r--. 1 root root   54 May 15 09:49 meta-data
-rw-r--r--. 1 root root  111 May 15 09:49 network-config
-rw-r--r--. 1 root root 1728 May 15 09:49 user-data
-rw-r--r--. 1 root root    0 May 15 09:49 vendor-data
[root@vm3000 1]# more meta-data
{
"instance-id": "3000",
"local-hostname": "vm3000"
}

root@pve7demo1:~# qm config 3000
cicustom: meta=nfs:snippets/cirros.meta.3000,user=nfs:snippets/centos.user
name: vm3000
scsi0: bb-nvme:vm-3000-disk-0,size=10G
scsi1: bb-nvme:vm-3000-cloudinit,media=cdrom

Notice that my custom userdata contained wrong hostname:

more user-data
#cloud-config

chpasswd:
    expire: false
hostname: vm2000
manage_etc_hosts: true

Looking at cloudinit log you can see where hostname is being changed from vm2000, to 3000 (similar messages are present for localhost in non-custom case):

grep 2000 /var/log/cloud-init*
/var/log/cloud-init.log:2023-05-15 13:49:21,939 - cc_set_hostname.py[DEBUG]: Setting the hostname to vm3000 (vm2000)
/var/log/cloud-init.log:2023-05-15 13:49:21,951 - cc_update_hostname.py[DEBUG]: Updating hostname to vm3000 (vm2000)

At high level the conclusion I make - modern cloud images prefer hostname set in metadata, which is consistent with most major cloud providers using metadata servers, rather than ISO based cloud-init.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox