Once a private cloud with OpenStack is up and running, virtual machines can be created quickly, but you have to get there first. This process involves several steps, including installing hardware in the data center, wiring, and installing an operating system. The operating system installation is where the OpenStack component Ironic [1] enters the field. In this article, we introduce the Ironic module, which specializes in bare metal installations.

Ironic first appeared as an Incubator Project in the OpenStack Icehouse release of April 2017 [2] as an offshoot of the Nova bare metal driver. It was further integrated in the course of the Juno release and finally found its place as an integrated project in the Kilo release.

Bifrost

Installing an operating system raises the chicken and egg problem: The operating system is installed on disk by an installer, which in turn needs something like an operating system under which it can run. The Bifrost subproject aims to solve this conundrum. Underneath, you have to imagine a standalone, running Ironic that OpenStack provides with all the components it requires.

This article assumes the following situation: A Bifrost instance and OSISM (Open Source Infrastructure and Service Manager) are used to deploy an operating system on a node. OSISM [3], the deployment and lifecycle management framework, installs and configures all the required OpenStack components on that node so that another node can subsequently be installed by Ironic (Figure 1).

Figure 1: The sample setup for this article includes three servers with two networks each.

Bifrost bridges the gap between "the hardware is ready and wired in the data center" and "now you can log on to the operating system." The name comes from Norse mythology and refers to the rainbow bridge between Midgard and Asgard (i.e., the connection between middle earth, the home of humans, and the kingdom of the gods and goddesses).

Bifrost requires two configuration files in the environments/kolla/files/overlays/ directory: one for itself and one on which the server is to be installed (node1 in this case). The most important parameters in the Bifrost configuration file (bifrost.yml, Listing 1) are enabled_hardware_types, use_tinyipa, cirros_deploy_image_upstream_url, and dhcp_pool*.

enabled_hardware_types: ipmi
download_ipa: true
use_tinyipa: false
ipa_upstream_release: stable-{{   openstack_release }}
cirros_deploy_image_upstream_url: https://share/ironic-ubuntu-osism-20.04.qcow2
dhcp_pool_start: 192.168.21.200
dhcp_pool_end: 192.168.21.250
dnsmasq_router: 192.168.21.254
domain: osism.test

The enabled_hardware_types configuration option describes the interface for controlling the hardware [4], and the use_tinyipa variable gives you a choice between CoreOS (true) and CentOS (false); CoreOS is intended for continuous integration and test environments. The cirros_deploy_image_upstream_url parameter contains the link to the image to be installed, and dhcp_pool* contains the DHCP pool for the PXE environment.

Listing 2 shows the servers.yml configuration file for server(s) to be installed. It is important to use the MAC address of the deployment network interface controller (NIC) here and not, say, that of the remote NIC interface. In the case of Supermicro, for example, you determine this with the command:

---
node1:
  uuid: "UUID"
  driver_info:
    power:
      ipmi_username: "<LOGIN>"
      ipmi_address:  "<IP/Hostname>"
      ipmi_password: "<lesssecret>"
  nics:
    -
    mac: "<MAC of NIC>"
  driver: "ipmi"
  ipv4_interface_mac: "<MAC of NIC>"
  ipv4_address: "192.168.21.21"
  ipv4_subnet_mask: "255.255.255.0"
  ipv4_gateway: "192.168.21.254"
  ipv4_nameserver:
    - "8.8.8.8"
    - "8.8.4.4"
  properties:
    cpu_arch: "x86_64"
    ram: "65536"
    disk_size: "1024"
    cpus: "16"
  name: "node1"

# ipmitool -H <IP address> -U <user> -P <password> raw 0x30 0x21 | tail -c 18 | sed 's/ /:/g'

To avoid the password specified here appearing in the process list and history, use the -f /<path>/<to>/<passwordfile> parameter instead of -P <password>, if necessary, and enter the path to a file containing the password.

Once you have committed both files to a Git repository, you can now turn to the OSISM Manager inventory. To begin, assign roles by modifying the inventory/20-roles file as shown in Listing 3; you need to define the host variables for node1 in inventory/host_vars/node1.osism.test.yml (Listing 4).

[generic]
manager.osism.test
node1.osism.test
[manager]
manager.osism.test
[monitoring]
manager.osism.test
[control]
node1.osism.test
[compute]
node1.osism.test
[network]
node1.osism.test

--
ansible_host: 192.168.21.21
console_interface: eno1
management_interface: eno1
internal_address: 192.168.21.21
fluentd_host: 192.168.21.21
network_interfaces:
  - device: eno1
    auto: true
    family: inet
    method: static
    address: 192.168.21.21
    netmask: 255.255.255.0
    gateway: 192.168.21.254
    mtu: 1500
  - device: eno2
    auto: true
    family: inet
    method: manual
    mtu: 1500
network_interface: eno1

These files also should be maintained in the Git repository and retrieved with the osism-generic configuration command in the OSISM Manager, which updates the repository contents in /opt/configuration/. This workflow always serves as the basis for an OSISM environment: maintaining the Git repository and retrieving with osism-generic configuration:

local: git add <path>/<to>/<file>
local: git commit -m "<Commit Message>"
manager: osism-generic configuration

Now Bifrost can be installed with the:

osism-kolla deploy bifrost

command to create the bifrost_deploy container and four volumes (Listing 5, lines 1-5). There are two operating system images: One is the Ironic Python Agent (IPA), and the other is the deployment image (lines 7-14). The IPA image is CentOS (lines 16 and 17).

01 # docker volume ls
02 local     bifrost_httpboot
03 local     bifrost_ironic
04 local     bifrost_mariadb
05 local     bifrost_tftpboot
06
07 # ll -h /var/lib/docker/volumes/bifrost_httpboot/_data/
08 [...]
09 -rw-r--r-- 1 42422 42422 6.1G Mar  2 10:33 deployment_image.qcow2
10 -rw-r--r-- 1 root  root    57 Mar  2 10:21 deployment_image.qcow2.CHECKSUMS
11 -rw-r--r-- 1 42422 42422 318M Mar  2 10:13 ipa.initramfs
12 -rw-r--r-- 1 42422 42422  104 Mar  2 10:13 ipa.initramfs.sha256
13 -rw-r--r-- 1 42422 42422 9.1M Mar  2 10:13 ipa.kernel
14 -rw-r--r-- 1 42422 42422  101 Mar  2 10:12 ipa.kernel.sha256
15
16 # file /var/lib/docker/volumes/bifrost_httpboot/_data/ipa.kernel
17 /var/lib/docker/volumes/bifrost_httpboot/_data/ipa.kernel: Linux kernel x86 boot executable bzImage, version 4.18.0-240.10.1.el8_3.x86_64 (mockbuild@kbuilder.bsys.centos.org) #1 SMP Mon Jan 18 17:05:51 UT, RO-rootFS, swap_dev 0x9, Normal VGA

The deployment image here is ironic-ubuntu-osism-21.04.qcow2, which was configured in Listing 1. The osism-kolla deploy-servers bifrost command triggers the installation of node1. Bifrost creates a PXE environment for the MAC address configured for node1 – started by Intelligent Platform Management Interface (IPMI) boot into the PXE environment – and the IPA image is loaded and started. The Ironic Python Agent connects to Bifrost and writes the deployment image to disk. If this is successful, Bifrost dismantles the PXE environment, restarts node1, and boots from disk.

Up to Neutron with OSISM

Installing and setting up the OpenStack components requires an SSH connection between the Manager and node1. The private SSH key is available in the OSISM Manager in /opt/ansible/secrets/id_rsa.deploy and also in the Ansible Vault in environments/kolla/secrets.yml; look for the bifrost_ssh_key.private_key variable. You can then connect:

$ ssh -i /opt/ansible/secrets/id_rsa.deploy ubuntu@192.168.21.21

After successfully opening a connection, the OpenStack components can be installed from the OSISM framework. The statement:

# osism-generic operator --limit node1.osism.test --user ubuntu --ask-become-pass --private-key /ansible/secrets/id_rsa.deploy

sets up the operator user that you need for all further osism-* calls.

The osism-generic facts command collects the facts, and the network configuration stored in host_vars is written with the osism-generic network command. The call to osism-generic reboot verifies the network configuration. After the configuration of the remaining OpenStack components in the Git repository, the commands in Listing 6 continue the installation up to the Nova phase.

# osism-generic hosts
# osism-generic bootstrap
# osism-kolla deploy common
# osism-kolla deploy haproxy
# osism-kolla deploy elasticsearch
# osism-kolla deploy kibana
# osism-kolla deploy memcached
# osism-kolla deploy mariadb
# osism-kolla deploy rabbitmq
# osism-kolla deploy openvswitch
# osism-kolla deploy keystone
# osism-kolla deploy horizon
# osism-kolla deploy glance
# osism-kolla deploy placement
# osism-kolla deploy nov

OSISM relies on local name resolution and maintains the /etc/hosts file on manager and node1 for this purpose. As part of the bootstrap, it sets kernel parameters, (de)installs packages, and hardens the SSH server. In this setup, the manager also takes on the controller role, which you would distribute across at least three nodes in a production setup (Figure 2).

Figure 2: A rough overview of how the OpenStack components interact with Ironic.

HAProxy handles the distribution of queries to the controller nodes. The common, elasticsearch, and kibana components provide centralized logging that can be accessed from the Kibana web interface. Memcached accelerates the Apache server behind OpenStack's Horizon web interface. OpenStack requires a database (MariaDB in this case), a message queue (RabbitMQ), and a virtual switch (Open vSwitch).

Now the Keystone authentication/authorization, Horizon, Glance image store, Placement resource management, Nova, and Hypervisor components are installed and set up. The following customizations for the OpenStack Neutron module allow Ironic to talk to the hardware:

neutron_type_drivers: "flat,vxlan"
neutron_tenant_network_types: "flat"
enable_ironic_neutron_agent: "yes"

The mandatory network type flat provides a way to communicate with the hardware at the network level. The ironic_neutron_agent component is responsible for communication between the Neutron network component and Ironic. According to the git commands entered earlier, osism-kolla deploy neutron installs and configures the Neutron component.

En Route to Ironic

Ironic deployment requires a provision network and a cleaning network. The remote board network serves as the cleaning network, and the provision network handles the deployment (Listing 7). The UUIDs of the two networks can be used to populate Ironic's configuration (Listings 8 and 9 in the environments/kolla path).

# openstack network create --provider-network-type flat --provider-physical-network physnet1 --share provisionnet
# openstack subnet create  --network pxenet --subnet-range 192.168.21.0/24 --ip-version 4 --gateway 192.168.21.254 --allocation-pool start=192.168.21.22,end=192.168.21.28 --dhcp provisionsubnet
# openstack network create --provider-network-type flat --provider-physical-network physnet1 --share cleannet
# openstack subnet create  --network pxenet --subnet-range 10.21.21.0/24 --ip-version 4 --gateway 10.21.21.254 --allocation-pool start=10.21.21.22,end=10.21.21.28 --dhcp cleansubnet

:enable_ironic: "yes"
ironic_dnsmasq_dhcp_range: "192.168.21.150,192.168.21.199"
ironic_cleaning_network: "<cleannet-UUID>"
ironic_dnsmasq_default_gateway: "192.168.21.254"
ironic_dnsmasq_boot_file: pxelinux.0

[DEFAULT]
enabled_network_interfaces = noop,flat,neutron
default_network_interface = neutron
[neutron]
provisioning_network = <provisionnet-UUID>

The command:

osism-kolla deploy ironic

installs Ironic, with three other steps in addition to adjusting the quota: Send the deployment kernel, including the ramdisk and the image to be written to Glance (Listing 10); create a flavor with appropriate parameters [5];

# openstack flavor create --ram 65536 --disk 500 --vcpus 16 --property resources:CUSTOM_BAREMETAL_RESOURCE_CLASS=1 baremetal-flavor

# openstack image create --disk-format aki   --container-format aki  --file /data/ironic-agent.kernel    --public deploy-kernel
# openstack image create --disk-format ari   --container-format ari  --file /data/ironic-agent.initramfs --public deploy-ramdisk
# openstack image create --disk-format qcow2 --container-format bare --file /data/osism-image.qcow2      --public deployment-image

and create the node and port in Ironic (Listing 11).

# openstack baremetal node create --driver ipmi --name node2 --driver-info ipmi_username=ADMIN --driver-info ipmi_password=<lesssecret> --driver-info ipmi_address=192.168.33.22 --resource-class baremetal-resource-class --driver-info deploy_kernel=<Deploy-Kernel-UUID> --driver-info deploy_ramdisk=<Deploy-Ramdisk-UUID> --driver-info cleaning_network=<cleannet-UUID> --driver-info provisioning_network=<provisionnet-UUID> --property cpu_arch=x86_64 --property capabilities='boot_mode=uefi' --inspect-interface inspector --deploy-interface direct --boot-interface ipxe

Make sure you have the right IPMI data and that the resource class matches that stored in the flavor. The flavor uses CUSTOM_BAREMETAL_RESOURCE_CLASS, which gives you a baremetal-resource-class. Other examples include CUSTOM_BAREMETAL_LARGE or CUSTOM_BAREMETAL_HANA, which resolve to baremetal-large and baremetal-hana.

As the driver-info, you need to specify the cleaning and provision networks in addition to the deploy kernel and the associated ramdisk. To enable deployment by iPXE, you need to specify direct as the deploy interface and UEFI as the boot_mode, and then create the Ironic port for node2 manually:

# openstack baremetal port create <MAC-Address> --node <Baremetal-Node-UUID> --physical-network physnet1

or load it by in-band inspection.