Skip to content

All-in-one lab on a single VM, nested inside an existing OpenStack cloud

A complete OpenStack 2026.1 (Keystone, Glance, Nova, Neutron with Open vSwitch, Placement, Heat, Horizon) on one VM of an existing cloud, deployed with Kolla-Ansible's all-in-one inventory. It is the smallest lab that still boots real instances under KVM and gives them floating IPs. It was built in about 70 minutes, including about 17 minutes of image pull, and two errors had to be fixed on the way; both are described below with their fix.

Context

  • Purpose. A first lab: learn how Kolla-Ansible lays out a deployment, and try runbooks on something that can be broken and rebuilt. It does not need HA; it needs to be cheap, quick to rebuild, and close to Kolla-Ansible's defaults.
  • Constraints of the hosting cloud (the same as for the multinode pattern): project networks with an MTU of 1442; nested virtualisation available inside instances; no Internet egress for ports without a floating IP; outbound TCP 80 blocked. An existing deploy host on the same project network provides an HTTPS proxy, an NTP server and a local registry holding the Kolla images under a tag that never moves.
  • Budget. One VM of 4 vCPU, 16 GB of RAM and a 50 GB root disk.
  • Skills. Anyone who can follow Kolla-Ansible's quickstart; nothing here goes beyond it except what the hosting cloud forces.

Topology

flowchart LR
  deploy["deploy host<br>HTTPS proxy, NTP,<br>image registry"]
  subgraph vm["lab VM (all-in-one)"]
    svc["API services, MariaDB, RabbitMQ,<br>Neutron agents, nova-compute + libvirt (kvm)"]
    br["br-ex"] --- veth1 --- veth0["veth0<br>external gateway"]
  end
  deploy -- "management network (ens3)" --> vm
Role Count Size Runs
All-in-one node 1 4 vCPU, 16 GB RAM, 50 GB disk, one NIC every service of Kolla-Ansible's all-in-one inventory: control, network, compute, monitoring and deployment groups; kolla-ansible itself runs on the node, from a virtual environment
Deploy host 1 existing machine only serves the lab: HTTPS proxy, NTP, image registry

Network design

Interface Kolla-Ansible role Notes
ens3 (the VM's only port, on the project network) network_interface: internal API, service traffic, and the "VIP" port security enabled; the lab VM's security group allows SSH and ICMP from the deploy host only
veth1 neutron_external_interface, plugged into br-ex one end of a local veth pair, created persistently with systemd-networkd
veth0 none the other end; carries the gateway address of the lab's external network, so that the node reaches floating IPs
  • No second port for the external network. A veth pair gives Kolla-Ansible the address-less interface it wants for br-ex, without a second port and without disabling port security anywhere. The price: floating IPs are only reachable from the node itself.
  • No HAProxy, keepalived or ProxySQL. With one node there is nothing to fail over to; enable_haproxy: "no", enable_proxysql: "no", and kolla_internal_vip_address set to the node's own address. ProxySQL must be disabled explicitly in 2026.1 (see what went wrong).
  • MTU. The hosting cloud's 1442 bytes, given to Neutron with an override, /etc/kolla/config/neutron.conf:

    [DEFAULT]
    global_physnet_mtu = 1442
    
  • Egress. apt, pip, kolla-ansible install-deps, Docker and the containers all go through the deploy host's HTTPS proxy; the images come from its registry, without the proxy.

Storage design

No Cinder, no Ceph. Glance stores images in a Docker volume on the node, and Nova keeps instance disks on the node's root file system, so a lab that fills the disk fills it for every service. Enough for CirrOS instances; give the VM a larger disk, or a separate volume for the instance store, before trying larger images.

Inventory and configuration layout

~/kolla-venv/            kolla-ansible 22.2.0 and ansible-core 2.20.10, plus docker and dbus-python (see below)
~/all-in-one             Kolla-Ansible's all-in-one inventory, unchanged (localhost, ansible_connection=local)
/etc/kolla/
  globals.yml            the values below, and nothing else
  passwords.yml          generated by kolla-genpwd, never copied anywhere
  config/
    neutron.conf         global_physnet_mtu

Values that differ from Kolla-Ansible's defaults:

Setting Default Here Why
enable_haproxy yes no one node, no VIP to fail over
enable_proxysql yes no otherwise it takes port 3306 on the node's address and MariaDB cannot start
kolla_internal_vip_address none the node's address no VIP without HAProxy and keepalived
network_interface, neutron_external_interface eth0, eth1 ens3, veth1 the VM's interface names; the veth pair above
global_physnet_mtu (Neutron, override) 1500 1442 the hosting cloud's MTU
docker_registry, docker_registry_insecure, openstack_tag quay.io, moving tag the deploy host's registry, plain HTTP, a pinned tag pinned images
docker_*_proxy, container_*_proxy, docker_apt_key_env unset the deploy host's proxy the only egress
docker_client_timeout 120 600 pulls through one shared proxy

Everything else keeps Kolla-Ansible's defaults (neutron_plugin_agent: openvswitch, nova_compute_virt_type: kvm, kolla_container_engine: docker).

Build it step by step

Placeholders: <deploy-host> (192.0.2.10) is the deploy host; <lab-vm> (192.0.2.25) is the all-in-one VM, on the project network <mgmt-network>; <cloud> is the hosting cloud's clouds.yaml entry; <flavor-4c16g> a flavor with 4 vCPU, 16 GB of RAM and a 50 GB disk.

1. Create the VM (on the deploy host, against the hosting cloud)

⚠️ State-changing. Creates a key pair, a security group, a port and a server in your project of the hosting cloud, and uses quota there. Check the quota first. Undo with the commands of delete the lab.

export OS_CLOUD=<cloud>
S=~/.lab/<lab-vm>; mkdir -p $S && chmod 700 $S
ssh-keygen -q -t ed25519 -N '' -C <lab-vm> -f $S/id_ed25519
openstack keypair create --public-key $S/id_ed25519.pub <lab-vm>
openstack security group create <lab-vm>-sg
openstack security group rule create <lab-vm>-sg --protocol tcp --dst-port 22 --remote-ip 192.0.2.10/32
openstack security group rule create <lab-vm>-sg --protocol icmp --remote-ip 192.0.2.10/32
openstack port create --network <mgmt-network> --security-group <lab-vm>-sg <lab-vm>-mgmt
openstack image list --status active -f value -c Name | grep -i '24\.04'   # the exact image name on your cloud
openstack server create --flavor <flavor-4c16g> --image '<ubuntu-24.04-image>' \
  --port <lab-vm>-mgmt --key-name <lab-vm> --config-drive true --wait <lab-vm>

--config-drive true because, on this cloud, the SSH key did not always reach instances through the metadata service. The VM was active in about 2 minutes. Before going further, check nested virtualisation: if /dev/kvm is missing in the VM, set nova_compute_virt_type: qemu later.

ssh -i $S/id_ed25519 ubuntu@192.0.2.25 'ls -l /dev/kvm; grep -cE "vmx|svm" /proc/cpuinfo; cloud-init status --wait'

Every following command runs on the lab VM.

2. Prepare the host

⚠️ State-changing. Changes the lab VM only: apt proxy and sources, automatic upgrades turned off, NTP source, a new veth pair.

printf 'Acquire::http::Proxy "http://192.0.2.10:3128";\nAcquire::https::Proxy "http://192.0.2.10:3128";\n' \
  | sudo tee /etc/apt/apt.conf.d/90proxy
sudo sed -i 's#http://[^ ]*ubuntu.com/ubuntu/\?#https://archive.ubuntu.com/ubuntu/#' /etc/apt/sources.list.d/ubuntu.sources

sudo systemctl disable --now unattended-upgrades apt-daily.timer apt-daily-upgrade.timer
printf 'APT::Periodic::Update-Package-Lists "0";\nAPT::Periodic::Unattended-Upgrade "0";\n' \
  | sudo tee /etc/apt/apt.conf.d/20auto-upgrades

sudo mkdir -p /etc/systemd/timesyncd.conf.d
printf '[Time]\nNTP=192.0.2.10\n' | sudo tee /etc/systemd/timesyncd.conf.d/90-lab.conf
sudo systemctl restart systemd-timesyncd && timedatectl show -p NTPSynchronized   # NTPSynchronized=yes
sudo apt-get update

printf '[NetDev]\nName=veth0\nKind=veth\n[Peer]\nName=veth1\n' | sudo tee /etc/systemd/network/25-veth-ext.netdev
printf '[Match]\nName=veth0\n[Network]\nAddress=198.51.100.1/24\nLinkLocalAddressing=no\n' | sudo tee /etc/systemd/network/25-veth0.network
printf '[Match]\nName=veth1\n[Network]\nLinkLocalAddressing=no\n' | sudo tee /etc/systemd/network/25-veth1.network
sudo networkctl reload && ip -br a show veth0 && ip -br a show veth1

Right after the reload both ends may still show DOWN without an address; a few seconds later veth0 is UP with 198.51.100.1/24 and veth1 is UP without an address.

Automatic upgrades are turned off because they once upgraded libc6 and systemd in the middle of a deploy (deployment error page).

3. Install Kolla-Ansible

The Kolla-Ansible series must match the images: 22.x for 2026.1.

sudo apt-get install -y git python3-dev python3-venv libffi-dev gcc libssl-dev libdbus-glib-1-dev
export https_proxy=http://192.0.2.10:3128 http_proxy=http://192.0.2.10:3128
python3 -m venv ~/kolla-venv && source ~/kolla-venv/bin/activate
pip install -U pip
pip install 'ansible-core==2.20.10' 'kolla-ansible==22.2.0'
pip install docker dbus-python          # needed on an all-in-one, see "What went wrong"
kolla-ansible install-deps

sudo mkdir -p /etc/kolla/config && sudo chown -R $USER: /etc/kolla
cp ~/kolla-venv/share/kolla-ansible/etc_examples/kolla/* /etc/kolla/
cp ~/kolla-venv/share/kolla-ansible/ansible/inventory/all-in-one ~/all-in-one
kolla-genpwd

4. Configure

Replace /etc/kolla/globals.yml with:

kolla_base_distro: "ubuntu"
kolla_container_engine: "docker"
enable_haproxy: "no"
enable_proxysql: "no"
kolla_internal_vip_address: "192.0.2.25"
network_interface: "ens3"
neutron_external_interface: "veth1"
neutron_plugin_agent: "openvswitch"
nova_compute_virt_type: "kvm"
docker_registry: "192.0.2.10:4000"
docker_registry_insecure: true
openstack_tag: "2026.1-ubuntu-noble-<date>"
docker_client_timeout: 600
docker_http_proxy: "http://192.0.2.10:3128"
docker_https_proxy: "http://192.0.2.10:3128"
docker_no_proxy: "localhost,127.0.0.1,192.0.2.25,192.0.2.10,<lab-vm>"
docker_apt_key_env:
  http_proxy: "http://192.0.2.10:3128"
  https_proxy: "http://192.0.2.10:3128"
container_http_proxy: "http://192.0.2.10:3128"
container_https_proxy: "http://192.0.2.10:3128"
container_no_proxy: "localhost,127.0.0.1,192.0.2.25,192.0.2.10,<lab-vm>"

and add the MTU override:

printf '[DEFAULT]\nglobal_physnet_mtu = 1442\n' > /etc/kolla/config/neutron.conf

5. Deploy

⚠️ State-changing. Installs Docker and deploys every OpenStack service on the lab VM.

source ~/kolla-venv/bin/activate
ansible -i ~/all-in-one all -m ping
kolla-ansible bootstrap-servers -i ~/all-in-one   # about 2.5 min
kolla-ansible prechecks -i ~/all-in-one           # about 1 min
kolla-ansible pull -i ~/all-in-one                # about 17 min
kolla-ansible deploy -i ~/all-in-one              # about 20 min
kolla-ansible post-deploy -i ~/all-in-one         # writes /etc/kolla/admin-openrc.sh and clouds.yaml

During deploy, fatal: ... Timeout when waiting for search string MariaDB followed by ...ignoring is expected: Kolla-Ansible is checking whether a MariaDB cluster already exists. Real failures show as failed= greater than 0 in the PLAY RECAP.

6. Verify

The client is installed through the proxy, like everything else. In a new shell, export it again first:

source ~/kolla-venv/bin/activate
export https_proxy=http://192.0.2.10:3128 no_proxy=192.0.2.25,localhost,127.0.0.1
pip install python-openstackclient -c https://releases.openstack.org/constraints/upper/2026.1
. /etc/kolla/admin-openrc.sh
openstack compute service list      # nova-scheduler, nova-conductor, nova-compute: up
openstack network agent list        # DHCP, Open vSwitch, Metadata, L3 agents: alive

⚠️ State-changing. init-runonce creates a CirrOS image, flavors, a demo network, an external network and a router in the lab's own OpenStack; the commands after it create an instance and a floating IP.

export DEMO_NET_CIDR=203.0.113.0/24 DEMO_NET_GATEWAY=203.0.113.1
export EXT_NET_CIDR=198.51.100.0/24 EXT_NET_RANGE=start=198.51.100.100,end=198.51.100.199 EXT_NET_GATEWAY=198.51.100.1
~/kolla-venv/share/kolla-ansible/init-runonce

openstack server create --image cirros --flavor m1.tiny --key-name mykey --network demo-net --wait demo1
FIP=$(openstack floating ip create public1 -f value -c floating_ip_address)
openstack server add floating ip demo1 $FIP
sleep 30                     # the instance is still booting: a ping at once got no answer
ping -c3 $FIP
ssh -i ~/.ssh/id_ecdsa cirros@$FIP uname -a
sudo docker exec nova_libvirt sh -c 'virsh dumpxml $(virsh list --name | head -1)' | grep "domain type"   # kvm
kolla-ansible check -i ~/all-in-one

On the reference build: 31 containers running (35 on the second build), the instance under KVM, ping and SSH through the floating IP working, Horizon answering on http://192.0.2.25/, and kolla-ansible check with failed=0.

Delete the lab

🛑 Destructive. Deletes the lab VM and everything deployed on it, then its port, security group and key pair in the hosting cloud. Check that the names are those of the lab before running it. Cannot be undone: rebuild from step 1 (about 70 minutes).

export OS_CLOUD=<cloud>
openstack server delete --wait <lab-vm>
openstack port delete <lab-vm>-mgmt
openstack security group delete <lab-vm>-sg
openstack keypair delete <lab-vm>
rm -rf ~/.lab/<lab-vm>

Key choices and their consequences

Choice Alternative(s) Why this one here Consequence in operation
One node, all-in-one inventory The multinode pattern Cheapest lab that runs real instances; rebuilt in about an hour No HA: failover, cluster and multi-compute runbooks cannot be tried here
No HAProxy, keepalived or ProxySQL Keep HAProxy with a VIP reserved by allowed address pairs Nothing to fail over to; one port is enough API endpoints point at the node's address; ProxySQL has to be disabled by hand (below)
External network on a veth pair A second port with port security disabled No second port, port security stays on everywhere Floating IPs are reachable only from the node itself
kolla-ansible runs on the node Run it from the deploy host Exactly Kolla-Ansible's quickstart; the deploy host stays unchanged The Docker SDK must also be in the virtual environment (below)
Pinned images from a local registry Pull from quay.io Two builds run the same images Needs the registry to hold every image the default services use
kvm qemu Nested virtualisation available If the hosting cloud stops offering it, switch to qemu and expect slow instances

What went wrong during deployment

  • prechecks failed with No module named 'docker'.

    fatal: [localhost]: FAILED! => {"cmd": [".../kolla-venv/bin/python3.12", "-c", "import docker; ..."],
     "stderr": "ModuleNotFoundError: No module named 'docker'"}
    

    On an all-in-one, Ansible manages localhost with the virtual environment's Python, while bootstrap-servers installs the Docker SDK into the system Python (package python3-docker). Fixed with pip install docker dbus-python in the virtual environment, then prechecks again.

  • MariaDB exited and deploy hung. deploy stayed on RUNNING HANDLER [mariadb : Wait for first MariaDB service port liveness]; docker ps -a showed mariadb Exited (1), and /var/lib/docker/volumes/kolla_logs/_data/mariadb/mariadb.log said Can't start server: Bind on TCP/IP port. Got error: 98: Address already in use. ss -ltnp | grep 3306 showed proxysql on the node's address. In 2026.1 ProxySQL is enabled by default and listens on kolla_internal_vip_address:3306; without HAProxy that address is the node's own, where MariaDB listens too. ProxySQL also kept the load balancer role enabled, so a haproxy container ran despite enable_haproxy: "no". Fixed by stopping deploy, adding enable_proxysql: "no", removing the three containers and the empty MariaDB volume, then prechecks and deploy again:

    🛑 Destructive. Removes the MariaDB container and its data volume. Only on a fresh deployment that never completed, where the volume holds no data yet; on any other deployment this deletes every OpenStack database.

    sudo docker rm -f proxysql haproxy mariadb
    sudo docker volume rm mariadb
    
  • A remote pkill -f kolla-ansible killed its own SSH session. The shell running the command had kolla-ansible in its command line too, so the command returned 255. pkill -f ansible-playbook from another session, or Ctrl-C, stops the deploy cleanly.

Version notes

  • ProxySQL is on by default in 2026.1 (Kolla-Ansible 22.x, enable_proxysql: "{{ enable_mariadb }}"), as it already was in 2025.1 (enable_proxysql: "yes"): an all-in-one without HAProxy needs enable_proxysql: "no" as well as enable_haproxy: "no".
  • On Ubuntu 24.04, bootstrap-servers installs the Docker SDK from the python3-docker package, in the system Python only.

Lessons learned

  • Install docker and dbus-python in the virtual environment from the start on an all-in-one.
  • Turning off HAProxy is not enough on 2026.1; turn off ProxySQL too, or MariaDB never starts.
  • Read the PLAY RECAP, not the fatal: lines: one of them is expected during a first deploy.
  • A veth pair is an easy external network for a one-NIC lab, as long as you only need to reach floating IPs from the node.
  • Keep the build in one log, command by command: both errors were quick to fix because the exact output was there.

Validity and limits

Built twice: on 2026-10-21, and on 2026-10-24 by following this page from step 1 on a new VM, with the corrections now in it (the image name, the proxy before pip in step 6, the wait before the ping). The second build met neither of the errors above, since the page already avoids them: bootstrap-servers 2.5 min, prechecks 36 s, pull 17.7 min, deploy 22 min (failed=0, the expected MariaDB ...ignoring), about 52 minutes of machine time from step 1 to the end of step 6; kolla-ansible check failed=0, the instance under KVM, ping and SSH through its floating IP, Horizon answering. Both builds used OpenStack 2026.1, Kolla-Ansible 22.2.0, ansible-core 2.20.10, images 2026.1-ubuntu-noble, Docker and Open vSwitch, on Ubuntu 24.04 inside one particular hosting cloud with nested KVM. Not tried: OVN, Podman, Cinder, other distributions, qemu, a deploy straight from quay.io, or Kolla-Ansible run from the deploy host instead of the node. This is a learning lab, not a pattern for anything that must stay up.


Provenance: written by agent:lab-reviewer, from a lab build on 2026.1. Reviewed in its pull request (see the page history). Checked by a human operator: not yet.

Worked for you, or not? Tell us in one line: your release, your topology, and what happened.