Day 4: virt-install and cloud-init
Goal: Create a VM from a base image and cloud-init files with a single
virt-installcommand, fully set up after boot. · Duration: approx. 60 min · Prerequisite: Day 3
Briefing
virt-installwrites the domain XML for you, creates the disk, builds the seed for cloud-init and starts the VM. What you built by hand on Days 1 to 3 becomes one command.- cloud-init reads two files: meta-data says
who the VM is (
instance-id, hostname), user-data says what should happen (users, packages, commands). - The
instance-iddecides whether cloud-init runs “for the first time”. Same ID means: already done.
user-data + meta-data ──► virt-install ──► web1.qcow2 (overlay on basis/noble.qcow2)
├─► cloudinit.iso (first boot only)
└─► domain web1 + qemu-guest-agent channel
Drill 1: What does libvirt know about operating systems?
sudo apt-get install -y libosinfo-bin
osinfo-query os | grep ubuntu24Expected:
ubuntu24.04 | Ubuntu 24.04 LTS | 24.04 | …
With --osinfo ubuntu24.04, virt-install
picks suitable devices (virtio, machine type) and knows the system’s
minimum requirements.
Save your progress? Sign up for free
Drill 2: One base image for all
sudo mkdir -p /var/lib/libvirt/images/basis
sudo cp ~/kvm-lab/images/noble-server-cloudimg-amd64.img \
/var/lib/libvirt/images/basis/noble.qcow2
sudo chmod 444 /var/lib/libvirt/images/basis/noble.qcow2As on Day 2, the base is write-protected. From now on every VM gets a thin overlay on top of it.
Save your progress? Sign up for free
Drill 3: The cloud-init files
mkdir -p ~/kvm-lab/tag04 && cd ~/kvm-lab/tag04
cat > user-data <<EOF
#cloud-config
users:
- name: ubuntu
groups: [sudo]
sudo: "ALL=(ALL) NOPASSWD:ALL"
shell: /bin/bash
lock_passwd: false
plain_text_passwd: lab-console-pw-123
ssh_authorized_keys:
- $(cat ~/.ssh/id_ed25519.pub)
package_update: true
packages: [nginx, qemu-guest-agent]
runcmd:
- echo "web1 aus cloud-init" > /var/www/html/index.html
- systemctl enable --now qemu-guest-agent
EOF
cat > meta-data <<'EOF'
instance-id: web1-001
local-hostname: web1
EOFuserscreates the user completely: sudo without a password, console password, SSH key.packagesinstalls nginx and the QEMU guest agent, a small service through which libvirt talks to the guest (Drill 7).runcmdruns once, at the very end of the first boot.
Save your progress? Sign up for free
Drill 4: Look first, then build
virt-install --name web1 --memory 1024 --vcpus 2 --osinfo ubuntu24.04 --import \
--disk size=10,backing_store=/var/lib/libvirt/images/basis/noble.qcow2 \
--network network=default \
--cloud-init user-data=user-data,meta-data=meta-data \
--graphics none --noautoconsole \
--print-xml --dry-run | head -20Expected (shortened):
<domain type="kvm">
<name>web1</name>
<uuid>…</uuid>
<metadata>
<libosinfo:libosinfo xmlns:libosinfo="http://libosinfo.org/xmlns/libvirt/domain/1.0">
<libosinfo:os id="http://ubuntu.com/ubuntu/24.04"/>
</libosinfo:libosinfo>
</metadata>
<memory>1048576</memory>
<currentMemory>1048576</currentMemory>
<vcpu>2</vcpu>
<os>
<type arch="x86_64" machine="q35">hvm</type>
<smbios mode="sysinfo"/>
</os>
<features>
<acpi/>
<apic/>
</features>
<cpu mode="host-passthrough"/>
| Option | Meaning |
|---|---|
--import |
install nothing, the disk is already bootable |
--disk size=10,backing_store=… |
new 10 GB overlay web1.qcow2 on the base |
--cloud-init user-data=…,meta-data=… |
build the seed and insert it on first boot |
--graphics none --noautoconsole |
serial console, do not connect automatically |
Tried it: Without
--dry-run,--print-xmlstill creates the disk. The real invocation afterwards findsweb1.qcow2already there and silently names the new diskweb1-1.qcow2. What remains is an orphaned file that nobody uses anymore.
Save your progress? Sign up for free
Drill 5: Build
The same command without the last line:
virt-install --name web1 --memory 1024 --vcpus 2 --osinfo ubuntu24.04 --import \
--disk size=10,backing_store=/var/lib/libvirt/images/basis/noble.qcow2 \
--network network=default \
--cloud-init user-data=user-data,meta-data=meta-data \
--graphics none --noautoconsoleExpected:
WARNING Requested memory 1024 MiB is less than the recommended 3072 MiB for OS ubuntu24.04
Starting install...
Allocating 'virtinst-0ngrwmyt-cloudinit.iso' | 0 B 00:00 ...
Transferring 'virtinst-0ngrwmyt-cloudinit.iso' | 0 B 00:00 ...
Allocating 'web1.qcow2' | 0 B 00:00 ...
Creating domain... | 0 B 00:00
Domain creation completed.
The warning comes from the osinfo database. For an nginx, 1 GB is enough.
virsh domblklist web1
sudo qemu-img info -U /var/lib/libvirt/images/web1.qcow2 | grep -E 'virtual|^disk|backing'Expected:
Target Source
-----------------------------------------------------------------
vda /var/lib/libvirt/images/web1.qcow2
sda /var/lib/libvirt/boot/virtinst-0ngrwmyt-cloudinit.iso
virtual size: 10 GiB (10737418240 bytes)
disk size: 328 KiB
backing file: /var/lib/libvirt/images/basis/noble.qcow2
-U (force share) allows qemu-img info on a
disk that is currently in use. Without it you get the lock error from
Day 2.
Save your progress? Sign up for free
Drill 6: Fully set up?
After about a minute the guest has an IP:
virsh domifaddr web1Expected:
Name MAC address Protocol Address
-------------------------------------------------------------------------------
vnet4 52:54:00:5e:a8:8a ipv4 192.168.122.110/24
The packages take longer. cloud-init status --wait waits
until everything is done:
ssh -o StrictHostKeyChecking=no [email protected] cloud-init status --wait
ssh [email protected] cloud-init status --longExpected (shortened, after two to three minutes):
……………………………………………
status: done
status: done
extended_status: done
boot_status_code: enabled-by-generator
…
detail: DataSourceNoCloud [seed=dmi,/dev/sr0][dsmode=local]
errors: []
errors: [] is the most important line. A typo in
user-data ends up here, and the VM still starts.
curl http://192.168.122.110/Expected:
web1 aus cloud-init
Save your progress? Sign up for free
Drill 7: The QEMU guest agent
virt-install has built a channel to the QEMU guest agent
into the domain:
virsh dumpxml web1 | grep -A2 '<channel'Expected:
<channel type='unix'>
<source mode='bind' path='/run/libvirt/qemu/channel/7-web1/org.qemu.guest_agent.0'/>
<target type='virtio' name='org.qemu.guest_agent.0' state='connected'/>
state='connected': the QEMU guest agent is running. Now
libvirt can ask the guest itself instead of looking it up in the DHCP
server:
virsh domifaddr web1 --source agentExpected:
Name MAC address Protocol Address
-------------------------------------------------------------------------------
lo 00:00:00:00:00:00 ipv4 127.0.0.1/8
- - ipv6 ::1/128
enp1s0 52:54:00:5e:a8:8a ipv4 192.168.122.110/24
- - ipv6 fe80::5054:ff:fe5e:a88a/64
This also works with static IPs without DHCP. You will see what else the agent can do on Day 9.
Save your progress? Sign up for free
Drill 8: The first reboot
virsh reboot web1
virsh list --allExpected (after a few seconds):
Domain 'web1' is being rebooted
Id Name State
------------------------
- erste shut off
- web1 shut off
Off instead of rebooted. Why?
virsh dumpxml --inactive web1 | grep on_rebootExpected:
<on_reboot>restart</on_reboot>
Tried it: During the “installation”,
virt-installsetson_reboottodestroyin the running definition. Normally it waits for this first reboot, removes the seed and restarts the VM. With--noautoconsole, however,virt-installhas long since exited, and nobody starts the VM again. The stored definition is fine; from now onrestartapplies again.
virsh start web1
virsh domblklist web1Expected:
Domain 'web1' started
Target Source
----------------------------------------------
vda /var/lib/libvirt/images/web1.qcow2
sda -
The drive with the seed is empty. When creating the VM,
virt-install only attached the ISO file to the running
definition and deleted it afterwards. Good, because it contained the
console password.
Save your progress? Sign up for free
Checkpoint
virsh list --allExpected: web1 with running.
curl http://192.168.122.110/Expected: web1 aus cloud-init. Your IP is shown by
virsh domifaddr web1.
ssh [email protected] cloud-init statusExpected: status: done.
Challenge
Build a second VM web2 that serves
web2 aus cloud-init. Check that it shares the same base
with web1.
Hint: You need your own files with their own
instance-id and their own hostname.
Solution
cd ~/kvm-lab/tag04
cat > meta-data-web2 <<'EOF'
instance-id: web2-001
local-hostname: web2
EOF
cat > user-data-web2 <<EOF
#cloud-config
users:
- name: ubuntu
groups: [sudo]
sudo: "ALL=(ALL) NOPASSWD:ALL"
shell: /bin/bash
lock_passwd: false
plain_text_passwd: lab-console-pw-123
ssh_authorized_keys:
- $(cat ~/.ssh/id_ed25519.pub)
package_update: true
packages: [nginx, qemu-guest-agent]
runcmd:
- echo "web2 aus cloud-init" > /var/www/html/index.html
- systemctl enable --now qemu-guest-agent
EOF
virt-install --name web2 --memory 1024 --vcpus 2 --osinfo ubuntu24.04 --import \
--disk size=10,backing_store=/var/lib/libvirt/images/basis/noble.qcow2 \
--network network=default \
--cloud-init user-data=user-data-web2,meta-data=meta-data-web2 \
--graphics none --noautoconsoleAfter two to three minutes:
virsh domifaddr web2 --source agent | grep enp
curl http://192.168.122.51/Expected:
enp1s0 52:54:00:51:cc:46 ipv4 192.168.122.51/24
web2 aus cloud-init
sudo qemu-img info -U /var/lib/libvirt/images/web2.qcow2 | grep backingExpected:
backing file: /var/lib/libvirt/images/basis/noble.qcow2
backing file format: qcow2
Save your progress? Sign up for free
Enterprise Note
- cloud-init is the standard: Proxmox, OpenStack, OpenNebula and the big clouds understand cloud-config. Your user-data carries over largely unchanged; only the datasource differs.
- No secrets in user-data: In the guest, the file
lives at
/var/lib/cloud/instance/user-data.txtand is readable by root. Passwords belong in a secret store, not in the VM description. - Golden images: What every VM needs (agent, monitoring, hardening) goes into the base image, built for example with Packer. cloud-init then only handles what is individual.
- Always the QEMU guest agent: Backups with a consistent file system, IP display and clean shutdown depend on it.