Clout-init to create centos 7 guests on KVM

I wanted a quick way to provision a Centos VM in my Lab at home (Fedora host). Until now I was using virt-manager (GUI) to create a VM, attach the Centos DVD, boot it and go through the installer. It is ok but it takes too long. Luckily there is a faster way: download a cloud image, boot a VM based on it and very quickly we have a new guest ready. Furthermore this can be automated because everything is done at the command line.

I will document step by step how to create a Centos guest from a cloud image, all at the command line so that the guest creation is very fast and easy to automate. My guest will be called pg01 (I will use it for postgres), of course change occurences of pg01 in this document to what suits you.

Some additional documentation:

Install KVM and libvirt, config user

libvirt is normally installed by default on Fedora and on Centos 7, if not check for example

You don’t need to use root to create guest VM’s or to use virsh, you can keep your own user but make sure to add yourself in the libvirt group and check the env variable LIBVIRT_DEFAULT_URI

  • add user to group libvirt

In my case I am using user pierre, so:

sudo usermod -G libvirt -a pierre

nb: this requires login on again

  • add environment variable LIBVIRT_DEFAULT_URI
export LIBVIRT_DEFAULT_URI=qemu:///system

and add it to your profile also

echo export LIBVIRT_DEFAULT_URI=qemu:///system >> ~/.bash_profile

Download the cloud image.

This is fast because the image is small (895M).

# create a directory to store the downloaded image, mine is /data1/downloads
cd /data1/downloads 

We can get info about the image

qemu-img info CentOS-7-x86_64-GenericCloud.qcow2

As you can see the image virtual size is 8.0G but the disk size is only 895M

image: CentOS-7-x86_64-GenericCloud.qcow2
file format: qcow2
virtual size: 8.0G (8589934592 bytes)
disk size: 895M
cluster_size: 65536
Format specific information:
    compat: 0.10

I prefer to resize the image (the space will not be allocated until it is used) so that I get a bigger root partition in my new guest.

qemu-img resize CentOS-7-x86_64-GenericCloud.qcow2 20G

If we look again the info, the virtual size is 20G but the file size did not change (895M)

Next I create a storage pool for the new guest (just a directory on the host). I use an env variable to point to the directory in the remainder of this text. Change pg01 by whatever you decided for your new VM name.

export VMPOOLDIR=/data2/virtpool/pg01
sudo mkdir $VMPOOLDIR
sudo chown pierre:pierre $VMPOOLDIR
virsh pool-create-as --name pg01 --type dir --target $VMPOOLDIR

Prepare the iso for cloud-init

we need to create two files called user-data and meta-data, those two files will be put in the iso. Again change pg01 as suits you.

Since I like to have static IP for my guest, I included the section network-interfaces in the meta-data file. If you don’t need a fixed IP (or if you prefer to configure it after), remove this section and a few line from the file user-data as well (see below) so that an IP will be assigned via dhcp by libvirt.

cat > $VMPOOLDIR/meta-data <<EOF
instance-id: pg01
local-hostname: pg01
network-interfaces: |
  iface eth0 inet static

The second file is user-data. In this file I will reference my public key, so make sure to have a ssh keys pair in the directory HOME/.ssh. You can generate a key pair with ssh-keygen.

ssh-keygen -t rsa

the public key in $HOME/.ssh/ will be injected in the guest, so that you will be able to log on from the host via ssh.

To have a static IP configured, I added some hacking in the runcmd section in the user-data file. Maybe there is a way to express that I want NM_CONTROLLED to no and ONBOOT to yes via the network section in the meta-data file above ? But it did not work. The commands ifdown/ifup in the runcmd section come from red-hat documentation (workaround for a bug ?). If you don’t need a static IP, remove the network-interfaces section from meta-data and remove lines from section runcmd below (all lines except the first, i.e. keep the remove of cloud-init), remove also the DNS and resolv_conf settings.

Of course adapt the hostname to your need, in this example it is pg01

About the chpasswd section, it will set a password for root. This is super handy in case of problems because you can log through the console. But take care that the keyboard for the console might be qwerty and not azerty on first boot…

⚠️ On my Fedora host the section cloud_config_modules in which I enabled resolv_conf resulted in my guest not booting up. So I had to remove it but then /etc/resolv.conf contains wrong information. As a work-around I added a sed in the runcmd section, like this

  - [ sed, -i, -e, "s/", /etc/resolv.conf ]
cat > $VMPOOLDIR/user-data <<EOF
# Hostname management
preserve_hostname: False
hostname: pg01
fqdn: pg01.localnet
# Setup Users with ssh keys so that I can log in into new machine
  - default
  - name: ansible
    groups: ['wheel']
    shell: /bin/bash
      - $(cat $HOME/.ssh/
  - resolv_conf
# set timezone for VM
timezone: Europe/Brussels
# Remove cloud-init when finished with it
# some patches of network config that I could not do via network section in meta-data
  - [ yum, -y, remove, cloud-init ]
  - [ ifdown, eth0 ]
  - [ ifup, eth0 ]
  - [ sed, -i, -e, "s/ONBOOT=no/ONBOOT=yes/", /etc/sysconfig/network-scripts/ifcfg-eth0 ]
  - [ sed, -i, -e, "\$aNM_CONTROLLED=no", /etc/sysconfig/network-scripts/ifcfg-eth0 ]
# Set DNS
manage_resolv_conf: true
# we'll use dnsmask on the host
  nameservers: ['']

# Configure where output will go
  all: ">> /var/log/cloud-init.log"
# configure interaction with ssh server
ssh_deletekeys: True
ssh_genkeytypes: ['rsa', 'ecdsa']
# Install my public ssh key to the first user-defined user configured 
# in cloud.cfg in the template (which is centos for CentOS cloud images)
  - ssh-rsa $(cat $HOME/.ssh/
# So that we can logon from the console should ssh not be available
  list: |
  expire: False  

Now with cloud-localds we can create a bootable iso. The help of this command says “Create a disk for cloud-init to utilize nocloud”. Not sure what this means, but basically it uses user-data and meta-data to produce an iso with which we can boot our cloud image so that cloud-init will do its job at first boot.

sudo yum install cloud-utils mkisofs genisoimage
cloud-localds pg01.iso user-data meta-data

On fedora, I have a strange error image saying “missing ‘genisoimage’. Required for –filesystem=iso9660.”. However the command “dns install genisoimage” does nothing else than saying “mkisofs is already installed and is last version”. My assumption is that mkisofs is a new name for genisoimage but cloud-localds is looking for the old name (I should file a bug report to Fedora). To make it work I created a symlink from mkisofs to genisoimage

ln -s /usr/bin/mkisofs /usr/bin/genisoimage

With this iso and the cloud image we downloaded, we will be able to provision our new guest.

create a new guest

First copy the downloaded cloud image in the pool. I do that using qemu-img convert but a simple copy is OK

qemu-img convert -O qcow2 /data1/downloads/CentOS-7-x86_64-GenericCloud.qcow2 $VMPOOLDIR/pg01.qcow2

then run virt-install to create the VM

virt-install --name pg01 --memory 1024 --vcpus 2 --disk $VMPOOLDIR/pg01.qcow2,device=disk,bus=virtio --os-type generic --os-variant centos7.0 --virt-type kvm --network network=default,model=virtio --cdrom $VMPOOLDIR/pg01.iso 

If you do not specify –noautoconsole in the virt-install command, the program tries to start virt-viewer so that one can see the progress of the installation. When at the end, it prompts for a login, reboot the VM. If you have –noautoconsole then just wait long enough (a few minutes, it is very fast)

If you specified a static IP, then you know the IP otherwise you can get the IP of the new guest:

virsh domifaddr pg01

then I can ssh with user ansible (that I specified in the user-data file) or with user centos (defined in the cloud image). Since I copied my public key (via user-data file) I don’t need a password. I strongly recommand to change the root password so that you can connect via the console later on if you are stuck with ssh (wrong network config for example)

We can get rid of the iso and user-data and meta-data

rm $VMPOOLDIR/pg01.iso $VMPOOLDIR/user-data $VMPOOLDIR/meta-data

What’s left is:

  • configure the VM to have a static IP (if not done via cloud-config)
  • Add a disk and create a volume group

Static IP

What I would do

  1. change the dhcp range of the default network (NB: the default network uses NAT, if you need to access the VM’s from outside the host then you must create a bridge network)
virsh net-edit default

and change the dhcp range

      <range start='' end=''/>

so that IP’s between and are reserved for static allocation.

We can see that the change is not active

virsh net-dumpxml default 
virsh net-dumpxml default --inactive

We must stop and restart the default network

virsh net-destroy default
virsh net-start default
virsh net-dumpxml default 

then we must restart libvirtd

systemctl restart libvirtd

and we must reboot our VM then wait that the new IP is allocated

virsh reboot pg01
  1. Change the network configuration of the VM.

If doing it via cloud-init did not work, I would just do it manually…

get into the VM as root, and edit the file /etc/sysconfig/network-scripts/ifcfg-eth0

HWADDR=< keep existing one >
  1. add vm in /etc/hosts on the host

With libvird on the host, there is a dnsmask server automatically started so that the content of the /etc/hosts on the host will be made available to the guests via DNS. This is great because VM’s can connect to each other via DNS.

In /etc/hosts on the host pg01.localnet


first create a disk on the host

virsh vol-create-as --pool pg01 --name pg01-disk1.qcow2 --capacity 40G --allocation 10G --format qcow2

and allocate it to the VM

virsh attach-disk --domain pg01 --source $VMPOOLDIR/pg01-disk1.qcow2 --target vdb --persistent --driver qemu --subdriver qcow2

Then get into the VM and set-up the volume group, the logical volume and the file system

# as user root
yum install lvm2
pvcreate /dev/vdb
vgcreate vg01 /dev/vdb
lvcreate -l100%FREE -n lv_01 vg01
mkfs -t xfs /dev/vg01/lv_01
mkdir /u01

edit /etc/fstab and add

/dev/vg01/lv_01 /u01 xfs defaults 0 0

then mount the filesystem

mount /u01

The VM can be cloned, you can take snapshot, … you can do pretty everything via the command line. I’ll document that in a next post.

Written on March 5, 2019