Example LXC based Ironic deployment

This section describes a specific deployment of Ironic using Openstack-Ansible. A number of design choices are made which illustrate how to configure the Ironic service for a specific set of requirements.

Deployment design decisions:

  • LXC containers are used in the openstack control plane

  • A single bmaas network is used for Ironic provisioning, cleaning and inspection

  • The bmaas network is not routable to any other networks nor to the internal VIP

  • Multitenancy is used with Ironic servers attached to project networks

  • networking-generic-switch will be used to control network devices to achieve multitenancy

  • Cisco NXOS switches

  • The deployment uses VXLAN project networks

A number of these design decisions are opinionated and could be changed, e.g. allowing the bmaas network to be routed to other networks including the internal VIP would result in some simplification.

This example is illustrative of a specific set of deployment requirements but is not intended to be followed rigidly. It demonstrates some of the capabilities of Ironic and Openstack-Ansible and how to approach most parts of a practical deployment of Ironic at small to modest scales.


Consideration should be given to the security of IPMI or other out-of-band interfaces which are notoriously buggy and often have vendor specific in-band tools which allow the BMC and its firmware to be manipulated from userspace. Understand the risks of allowing IPMI/BMC interfaces to share a physical interface with onboard ethernet ports, this feature will allow full access to the management interface of a deployed Ironic node. For high security environments or where the users are untrusted consult your hardware vendor for appropriate hardening steps.


BMAAS network address plan

In this example the subnet for the bmaas network is The size of the subnet determines the maximum number of nodes in the Ironic deployment.


Purpose to .10

Reserve for potential physical routers / SVI to .29

Control plane Ironic container interfaces to .95

Neutron DHCP allocation pool in Ironic_Network to .127

Ironic Inspector DHCP pool range to .254

Static IPs manually assigned to IPMI/iDRAC

In this deployment the bmaas network combines several functions for simplicity. It would be possible to use separate networks for inspection, cleaning and provisioning but that is beyond the scope of this example.

The subnet is divided into several address ranges, a reservation for future interfaces on physical routers, or a gateway address if the subnet is to be made routable in the future. A small number of addresses are then reserved for the bmaas network to connect to the Ironic containers in the control plane, and the remainder of the addresses are shared between a neutron network for provisioning, a range for Ironic Inspector to allocate with DHCP and finally a block of addresses for the IPMI (or other protocol) management ports of the Ironic nodes.


This example will use VLAN ID 3003 for the bmaas network but any available VLAN ID could be used.


This example collapses the Ironic IPMI, provisioning, cleaning and inspection networks into the same subnet. It is possible to make these be separate from each other by creating a network for each different function and updating the container networks and Ironic service configuration suitably. In particular it should be understood that the nodes are “untrusted” during the cleaning phase and will be in an arbitrary state as left by the previous user.

IPMI Interfaces

When configuring IPMI interfaces for Ironic nodes, the following steps are recommended:

  • Use static IP allocations for the IPMI interfaces, unless there is already a very reliable means of allocating addresses with DHCP. The Ironic team do not consider the Neutron DHCP agent to be suitable for assigning addresses to the IPMI interfaces.

  • Use dedicated IPMI ports on Ironic nodes especially if multitenancy is required. A node with a shared onboard ethernet/IPMI port will have that port moved into the tenant network when deployment is complete and the Ironic control plane will no longer be able to communicate with the management interface of the node.

Maximum size of the deployment

The maximum size of this Ironic deployment is limited by the address alloction in the bmaas network. In this example there can be a maximum of 127 server BMC interfaces in the range

The maxiumum number of servers that can be simultanously provisioned is determined by the address allocation to the Neutron DHCP pool.

The maximum number of servers that can be simultanously inspected is determined by the address allocation to the Ironic Inspector DHCP pool.

To increase the size of the deployment, the size of the bmaas subnet should be increased and the addresses allocated to meet the number of required nodes and maximum number of simultaneous nodes being provisioned and inspected.

Openstack-Ansible configuration

Once the address plan has been decided, the Openstack-Ansible configuration can be updated to match.

The existing cidr_networks and used_ips sections in /etc/openstack_deploy/openstack_user_config.yml must have extra entries to describe the network range available for the Ansible inventory to assign to Ironic control plane containers, in this example all addresses in the bmaas network are marked as used except the range to

An additional network is defined in the provider_networks list which represents the connection between the bridge br-bmaas on the controller and eth15 inside the ironic service containers.

The bmaas network must be extended from the control plane hosts to the switch ports connected to the Ironic node IPMI interfaces, and also to switch ports connected to the interfaces on the Ironic nodes that will be used for PXEboot. This will typically be a VLAN allocated specifically for Ironic provisioning.

The hosts for the Ironic control plane containers are assigned.


It is the responsibility of the deployer to create br-bmaas on the controller nodes and ensure that it is connected to the correct VLAN ID for the bmaas network. Configuration of host networking is outside the scope of Openstack-Ansible.


The range key in the provider network definition is not used but its useful as an reminder in the config file of the VLAN ID.

  <existing entries>
  bmaas:        # for containers on the bmaas network

  <existing entries>
  # bmaas ips
  - ","     # reserve for routers or other infrastructure
  - ","   # reserve for ironic IPMI and provisioning

  <existing entries>
  # Network definition to connect Ironic LXC containers to the bmaas network
  # on the infra hosts
  - network:
      net_name: physnet_neutron
      container_type: "veth"
      container_bridge: "br-bmaas"
      container_interface: "eth15"
      ip_from_q: bmaas
      type: "vlan"
      range: "3003:3003"
        - ironic_api_container
        - ironic_compute_container
        - ironic_inspector_container

# ironic API and conductor
  infra1: *_infra1_
  infra2: *_infra2_
  infra3: *_infra3_

# nova used by ironic for machine state management
  infra1: *_infra1_
  infra2: *_infra2_
  infra3: *_infra3_

# Ironic-inspector can only support a single instance at the moment
# High availability for ironic-inspector is not yet implemented
  infra1: *_infra1_


This example uses YAML Anchors to simplify openstack_user_config.yml allowing the IP addresses of the infra nodes to be defined only once. See https://yaml.org/spec/1.2.2/#alias-nodes.

Create the Neutron configuration

Enable the Neutron baremetal and genericswitch mechanism drivers by updating /etc/openstack_deploy/group_vars/neutron_server.yml

  - ml2.genericswitch
  - ml2.baremetal

# keep the ml2 drivers in this order
# see https://storyboard.openstack.org/#!/story/2008686
neutron_ml2_mechanism_drivers: "genericswitch,baremetal"

Configure neutron networking-generic-switch to know about the switches that the Ironic nodes are connected to in /etc/openstack_deploy/user_variables.yml. These switches are programmed by neutron to switch the Ironic nodes between the provisioning and project networks once deployment is complete. This is enabling multitenancy for Ironic.

This example is for a Cisco NXOS based switch, which uses the same command set as a Cisco IOS based switch for the functions needed by networking-generic-switch. There is no specific device_type for NXOS.


A MAC address for the switch must be specified in the neutron config, but Cisco and some other switch vendors present a unique MAC address per port so the MAC address as seen from the client cannot be used to identify the switch. For IOS/NXOS networking-generic-switch uses the field switch_info from the Ironic node local_link_connection information rather than match a MAC address when choosing which switch to configure for a particular node.

  genericswitch:my-switch-name:            # This should match the hostname configured on the switch
    device_type: netmiko_cisco_ios         # It is really NXOS but the commands are the same
    ngs_mac_address: "cc:46:d6:64:4b:41"   # Doesn't seem to matter but is required - this is taken from an SVI on the mgmt network
    ip: ""                      # An IP on the switch which has ssh access from the br-mgmt network, loopback, SVI or mgmt0 as needed
    username: "neutron"                    # The user that Neutron will SSH to the switch as
    password: "supersecret"                # The password that Neutron will use to SSH to the switch
    ##  key_file: <ssh key file>           # An SSH key may be used instead of a password
    ngs_manage_vlans: "False"              # VLANs are already provisioned on the switch so tell neutron not to create/delete VLANs


The configuration for networking-generic-switch is added to neturon.conf rather than ml2_conf_genericswitch.ini as the config needs to be read by both neutron-rpc-server and neutron-server. neutron-server is a uwsgi service in openstack-ansible so is only passed one config file, see https://bugs.launchpad.net/openstack-ansible/+bug/1987405


If there is already an override in place for this variable then extend the existing override rather than making a second one.

Configure switch to allow ssh from Neutron

To achieve multitenancy, Neutron will connect to the specified switch and configure the port for the Ironic node being provisioned to be in the correct project VLAN once the deployment is complete. During deployment Neutron will ensure that the node is in the bmaas provisioning network as specified in the Ironic config.

A suitable user and credential must exist on the switch. The SSH connection will originate from the Neutron processes running on the OpenStack control plane, on the mgmt network. There must be an IP route from the mgmt network to an interface on the switch which permits SSH login. That interface could be a physical management port (mgmt0 on NXOS), a loopback interface, an SVI or another interface with an IP address. SSH communication with the switch can happen either in-band or out-of-band depending on the requirements of the deployment.

This example config is for a neutron user using password authentication on an NXOS switch as seen by show run. The config applied on the switch gives the neutron user access to a minimal set of commands for configuring VLAN membership on specific ports.

To control the commands that the neutron user is allowed to issue on the Cisco Nexus switch create a role:

role name neutron-role
  rule 3 permit command configure t
  rule 2 permit read-write feature interface
  rule 1 permit read
  vlan policy deny
    permit vlan 3003-3003
    permit vlan 3100-3200
  interface policy deny
    permit interface Ethernet1/1
    permit interface Ethernet1/2
    permit interface Ethernet1/3
    permit interface Ethernet1/4
    permit interface Ethernet1/5
    permit interface Ethernet1/6
    permit interface Ethernet1/7
    permit interface Ethernet1/8

This role allows the neutron user assign a port to VLAN 3003 which is the bmaas network and is used during node provisioning. Any project VLANS that nodes should be able to be moved into after deployment should also be permitted, range 3100-3200 here.

The interfaces which the neutron user is permitted to modify are listed, in this case individually but consult the switch documentation for other options such as a regular expression.

A similar config can be made on an Arista switch, where a much more explicit list of allowed CLI commands must be defined using regular expressions.

role neutron-role
   10 permit mode exec command configure
   20 permit mode exec command terminal width 511
   30 permit mode exec command terminal length 0
   40 permit mode exec command enable
   50 permit mode exec command copy running-config startup-config
   60 permit mode config command interface
   70 permit mode if-Et([1-9]|27|29)\/1 command switchport mode access
   80 permit mode if-Et([1-9]|27|29)\/1 command (no )*switchport access vlan (3003|3966)
   90 permit mode if-Et([1-9]|27|29)\/1 command no switchport mode trunk
   100 permit mode if-Et([1-9]|27|29)\/1 command switchport trunk allowed vlan none
   110 permit mode config command copy running-config startup-config

Create the user and password, which must match those in the neutron.conf / genericswitch config file options:

username neutron password 5 <ENCRYPTED-PASSWORD-HERE> role neutron-role

Allow SSH to the switch from the expected IP addresses, for example a pair out of band management hosts and the OpenStack mgmt network

ip access-list ACL_ALLOW_SSH_VTY
  10 permit tcp any eq 22
  20 permit tcp any eq 22

line vty
  session-limit 5
  exec-timeout 10
  access-class ACL_ALLOW_SSH_VTY in

Create the Neutron network for Ironic provisioning, cleaning and inspection

openstack network create \
  --internal \
  --provider-network-type vlan \
  --provider-physical-network physnet_neutron \
  --provider-segment 3003 \

openstack subnet create \
  --allocation-pool \
  --dhcp \
  --gateway none

Create the Ironic configuration

In /etc/openstack_deploy/user_variables_ironic.yml

## IRONIC ##

ironic_ipxe_enabled: yes                  # use HTTP image download from the ironic conductor container
ironic_enable_web_server_for_images: yes  # use same web server to cache user images

# Ensure values used during PXEboot refer directly to the correct interface on Ironic API container
# instead of the internal VIP
ironic_http_url: "{{ ironic_ipxe_proto }}://{{ container_networks['bmaas_address']['address'] }}:{{ ironic_ipxe_port }}"
ironic_tftp_server_address: "{{ container_networks['bmaas_address']['address'] }}"

# Enable ironic drivers
ironic_drivers_enabled:     # Use PXE boot and IPMItool
  - agent_ipmitool
  - pxe_ipmitool
  - pxe_drac                # enables drivers for Dell iDrac interface

# Configure Ironic to use Neutron networking
ironic_enabled_network_interfaces_list: "noop,neutron"
ironic_default_network_interface: neutron

# Enable the default set of cleaning steps
ironic_automated_clean: yes

# Configure the neutron networks that Ironc should use
ironic_neutron_provisioning_network_name: "Ironic_Network"
ironic_neutron_cleaning_network_name:     "Ironic_Network"
ironic_neutron_inspection_network_name:   "Ironic_Network"

# Ensure ironic API (using uwsgi) listens on br-bmaas for agent callbacks
# as well as the mgmt interface for the loadbalancer

# Add ipa-insecure=1 to kernel parameters
# Needed when ironic endpoint is available over https with self-signed cert.
ironic_kernel_append_params: "ipa-debug=1 systemd.journald.forward_to_console=yes ipa-insecure=1"

# INI file overrides
  # Disable full device erasure (slow) and just metadata erasure, and replace with "Express erasure"
  # which tries to use firmware secure-erase command, but if that fails, reverts to metadata erasure.
  # See: https://docs.openstack.org/ironic/yoga/admin/cleaning.html#storage-cleaning-options
    erase_devices_priority: 0
    erase_devices_metadata_priority: 0
    clean_step_priority_override: "deploy.erase_devices_express:5"

  # Direct IPA to callback directly to deploying ironic container (via BMAAS network)
  # instead of going via HAProxy on mgmt network. Only applies when bmaas network is isolated.
    endpoint_override: "http://{{ container_networks['bmaas_address']['address'] }}:6385"

  # Enable ipmitool's Serial-over-LAN terminal console for baremetal nodes
    enabled_console_interfaces: "ipmitool-socat,no-console"


# Direct Inspector to callback directly to deploying ironic container (via BMAAS network)
# instead of going via HAProxy on mgmt network. Only applies when bmaas network is isolated.
ironic_inspector_callback_url: "{{ ironic_inspector_service_internaluri_proto }}://{{ container_networks['bmaas_address']['address'] }}:{{ ironic_inspector_service_port }}/v1/continue"

# Add ipa-insecure=1 to kernel parameters
# Needed when inspector is available over https with self-signed cert.
ironic_inspector_extra_callback_parameters: "ipa-collect-lldp=1 ipa-insecure=1"

# Ensure inspector API (using uwsgi) listens on br-bmaas for agent callbacks
# as well as the mgmt interface for the loadbalancer
ironic_inspector_service_address: ""

# dnsmasq/dhcp information for inspector
ironic_inspector_dhcp_enable_gateway: False
ironic_inspector_dhcp_enable_nameservers: False

ironic_inspector_dhcp_interface: eth15   # connected to br-bmaas on the host

Deploy Neutron changes

openstack-ansible playbooks/os-neutron-install.yml

Deploy the ironic-specific nova services

This deploys nova compute and nova console services to the ironic compute containers.

playbooks/os-nova-install.yml --limit ironic_all

Deploy changes to HAProxy

This will bring up the required Ironic, Inspector, and console endpoints.

openstack-ansible playbooks/haproxy-install.yml --tags haproxy_server-config

Deploy the Ironic and Inspector services

openstack-ansible playbooks/os-ironic-install.yml

Deploy the Horizon dashbaords for Ironic

openstack-ansible playbooks/os-horizon-install.yml

Using Ironic

Please refer to the general instructions in the Configuring Ironic section of this documentation.

VXLAN project networks

In this example Ironic multitenancy is implemented using VLANs. In an OpenStack deployment where project networks are implemented using an overlay such as VXLAN, it will not be possible to attach Ironic nodes directly to these networks. In addition, it is not possible for an end user to request that the underlying implementation is VLAN when creating a project network.

In a cloud using overlay project networks it will be necessary for the cloud administrator to create VLAN provider networks for users to attach Ironic nodes to and to share these into individual projects using Neutron RBAC.