How-To

Use the OpenStack CLI

If you wish to utilize the OpenStack CLI in no-auth mode, there are two options for configuring the authentication parameters.

clouds.yaml

During installation, Bifrost creates a clouds.yaml file with credentials necessary to access Ironic. A cloud called bifrost is always available. For example:

export OS_CLOUD=bifrost
openstack baremetal node list

In noauth mode, a cloud called bifrost-inspector is also included that allows access to the Ironic Inspector API. For example:

export OS_CLOUD=bifrost-inspector
openstack baremetal introspection list

Environment variables

Note

Previous versions of Bifrost recommended to use the ironic CLI rather than the openstack CLI. Doing this requires setting the OS_AUTH_TOKEN environment variable, however this causes Ansible enroll-dynamic.yaml and deploy-dynamic.yaml playbooks to fail, so OS_AUTH_TOKEN should be unset before running either of these.

The following two environment variables can be set:

  • OS_AUTH_TYPE - set to none to bypass authentication.

  • OS_ENDPOINT - A URL to the ironic API, such as http://localhost:6385/

For convenience, an environment file called env-vars is provided that contains default values for these variables and can be sourced to allow the CLI to connect to a local Ironic installation operating in noauth mode. For example:

. env-vars
openstack baremetal node list

This should display a table of nodes, or nothing if there are no nodes registered in Ironic.

Enroll Hardware

The following requirements are installed during the install process as documented in the install documentation.

  • openstack/shade library

  • openstack/os-client-config

In order to enroll hardware, you will naturally need an inventory of your hardware. When utilizing the dynamic inventory module and accompanying roles the inventory can be supplied in one of three ways, all of which ultimately translate to JSON data that Ansible parses.

The original method is to utilize a CSV file. This format is covered below in the Legacy CSV File Format section. This has a number of limitations, but does allow a user to bulk load hardware from an inventory list with minimal data transformations.

The newer method is to utilize a JSON or YAML document which the inventory parser will convert and provide to Ansible.

In order to use, you will need to define the environment variable BIFROST_INVENTORY_SOURCE to equal a file, which then allows you to execute Ansible utilizing the bifrost_inventory.py file as the data source.

Conversion from CSV to JSON formats

The inventory/bifrost_inventory.py program additionally features a mode that allows a user to convert a CSV file to the JSON data format utilizing a --convertcsv command line setting when directly invoked.

Example:

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.csv
inventory/bifrost_inventory.py --convertcsv >/tmp/baremetal.json

JSON file format

The JSON format closely resembles the data structure that ironic utilizes internally. The name, driver_info, nics, driver, and properties fields are directly mapped through to ironic. This means that the data contained within can vary from host to host, such as drivers and their parameters thus allowing a mixed hardware environment to be defined in a single file.

Example:

{
    "testvm1": {
      "uuid": "00000000-0000-0000-0000-000000000001",
      "driver_info": {
        "power": {
          "ipmi_address": "192.168.122.1",
          "ipmi_username": "admin",
          "ipmi_password": "pa$$w0rd"
        }
      },
      "nics": [
        {
          "mac": "52:54:00:f9:32:f6"
        }
      ],
      "driver": "ipmi",
      "ansible_ssh_host": "192.168.122.2",
      "ipv4_address": "192.168.122.2",
      "provisioning_ipv4_address": "10.0.0.9",
      "properties": {
        "cpu_arch": "x86_64",
        "ram": "3072",
        "disk_size": "10",
        "cpus": "1"
      },
      "name": "testvm1"
    }
}

The additional power of this format is easy configuration parameter injection, which could potentially allow a user to provision different operating system images onto different hardware chassis by defining the appropriate settings in an instance_info variable.

Examples utilizing JSON and YAML formatting, along host specific variable injection can be found in the playbooks/inventory/ folder.

Legacy CSV file format

The CSV file has the following columns:

  1. MAC Address

  2. Management username

  3. Management password

  4. Management Address

  5. CPU Count

  6. Memory size in MB

  7. Disk Storage in GB

  8. Flavor (Not Used)

  9. Type (Not Used)

  10. Host UUID

  11. Host or Node name

  12. Host IP Address to be set

  13. ipmi_target_channel - Requires: ipmi_bridging set to single

  14. ipmi_target_address - Requires: ipmi_bridging set to single

  15. ipmi_transit_channel - Requires: ipmi_bridging set to dual

  16. ipmi_transit_address - Requires: ipmi_bridging set to dual

  17. ironic driver

  18. Host provisioning IP Address

Example definition:

00:11:22:33:44:55,root,undefined,192.168.122.1,1,8192,512,NA,NA,aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee,hostname_100,192.168.2.100,,,,ipmi,10.0.0.9

This file format is fairly flexible and can be easily modified although the enrollment and deployment playbooks utilize the model of a host per line model in order to process through the entire list, as well as reference the specific field items.

An example file can be found at: playbooks/inventory/baremetal.csv.example

How this works?

Utilizing the dynamic inventory module, enrollment is as simple as setting the BIFROST_INVENTORY_SOURCE environment variable to your inventory data source, and then executing the enrollment playbook.:

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.json
ansible-playbook -vvvv -i inventory/bifrost_inventory.py enroll-dynamic.yaml

When ironic is installed on remote server, a regular ansible inventory with a target server should be added to ansible. This can be achieved by specifying a directory with files, each file in that directory will be part of the ansible inventory. Refer to ansible documentation http://docs.ansible.com/ansible/intro_dynamic_inventory.html#using-inventory-directories-and-multiple-inventory-sources

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.json
rm inventory/*.example
ansible-playbook -vvvv -i inventory/ enroll-dynamic.yaml

Note that enrollment is a one-time operation. The Ansible module does not synchronize data for existing nodes. You should use the ironic CLI to do this manually at the moment.

Additionally, it is important to note that the playbooks for enrollment are split into three separate playbooks based on the ipmi_bridging setting.

Deploy Hardware

How this works?

After the nodes are enrolled, they can be deployed upon. Bifrost is geared to utilize configuration drives to convey basic configuration information to the each host. This configuration information includes an SSH key to allow a user to login to the system.

To utilize the newer dynamic inventory based deployment:

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.json
ansible-playbook -vvvv -i inventory/bifrost_inventory.py deploy-dynamic.yaml

When ironic is installed on remote server, a regular ansible inventory with a target server should be added to ansible. This can be achieved by specifying a directory with files, each file in that directory will be part of the ansible inventory. Refer to ansible documentation http://docs.ansible.com/ansible/intro_dynamic_inventory.html#using-inventory-directories-and-multiple-inventory-sources

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.json
rm inventory/*.example
ansible-playbook -vvvv -i inventory/ deploy-dynamic.yaml

Note:

Before running the above command, ensure that the value for
`ssh_public_key_path` in ``./playbooks/inventory/group_vars/baremetal``
refers to a valid public key file, or set the ssh_public_key_path option
on the ansible-playbook command line by setting the variable.
Example: "-e ssh_public_key_path=~/.ssh/id_rsa.pub"

If the hosts need to be re-deployed, the dynamic redeploy playbook may be used:

export BIFROST_INVENTORY_SOURCE=/tmp/baremetal.json
ansible-playbook -vvvv -i inventory/bifrost_inventory.py redeploy-dynamic.yaml

This playbook will undeploy the hosts, followed by a deployment, allowing a configurable timeout for the hosts to transition in each step.

Execute local testing

A simple scripts/test-bifrost.sh script can be utilized to install pre-requisite software packages, Ansible, and then execute the test-bifrost-create-vm.yaml and test-bifrost.yaml playbooks in order to provide a single step testing mechanism.

playbooks/test-bifrost-create-vm.yaml creates one or more VMs for testing and saves out a baremetal.json file which is used by playbooks/test-bifrost.yaml to execute the remaining roles. Two additional roles are invoked by this playbook which enables Ansible to connect to the new nodes by adding them to the inventory, and then logging into the remote machine via the user’s ssh host key. Once that has successfully occurred, additional roles will unprovision the host(s) and delete them from ironic.

Command:

scripts/test-bifrost.sh

Note:

  • Cleaning mode is explicitly disabled in the test-bifrost.yaml playbook due to the fact that is an IO-intensive operation that can take a great deal of time.

  • In order to cap requirements for installation, an upper_constraints_file setting is defined. This is consuming the UPPER_CONSTRAINTS_FILE env var by default, to properly integrate with CI systems, and will default to /opt/stack/requirements/upper-constraints.txt file if not present.

Manually test with Virtual Machines

Bifrost supports using virtual machines to emulate the hardware.

The VirtualBMC project is used as an IPMI proxy, so that the same ipmi hardware type can be used as for real hardware.

  1. Set testing to true in the playbooks/inventory/group_vars/target file.

  2. You may need to adjust the value for ssh_public_key_path.

  3. Run the install step, as documented above, however adding -e testing=true to the Ansible command line.

  4. Execute the ansible-playbook -vvvv -i inventory/target test-bifrost-create-vm.yaml command to create a test virtual machine.

  5. Set the environment variable of BIFROST_INVENTORY_SOURCE to the path to the JSON file, which by default has been written to /tmp/baremetal.json.

  6. Run the enrollment step, using the JSON file you created in the previous step.

  7. Run the deployment step, as documented in Deploy Hardware.

Deployment and configuration of operating systems

By default, Bifrost deploys a configuration drive which includes the user SSH public key, hostname, and the network configuration in the form of network_data.json that can be read/parsed by the glean utility. This allows for the deployment of Ubuntu, CentOS, or Fedora “tenants” on baremetal. This file format is not yet supported by Cloud-Init, however it is on track for inclusion in cloud-init 2.0.

By default, Bifrost utilizes a utility called simple-init which leverages the previously noted glean utility to apply network configuration. This means that by default, root file systems may not be automatically expanded to consume the entire disk, which may, or may not be desirable depending upon operational needs. This is dependent upon what base OS image you utilize, and if the support is included in that image or not. At present, the standard Ubuntu cloud image includes cloud-init which will grow the root partition, however the ubuntu-minimal image does not include cloud-init and thus will not automatically grow the root partition.

Due to the nature of the design, it would be relatively easy for a user to import automatic growth or reconfiguration steps either in the image to be deployed, or in post-deployment steps via custom Ansible playbooks.

Build Custom Ironic Python Agent (IPA) images

Bifrost supports the ability for a user to build a custom IPA ramdisk utilizing diskimage-builder and ironic-python-agent-builder. In order to utilize this feature, the download_ipa setting must be set to false and the create_ipa_image must be set to “true”. By default, the install playbook will build a Debian stretch based IPA image, if a pre-existing IPA image is not present on disk. If you wish to explicitly set a specific release to be passed to diskimage-create, then the setting dib_os_release can be set in addition to dib_os_element.

If you wish to include an extra element into the IPA disk image, such as a custom hardware manager, you can pass the variable ipa_extra_dib_elements as a space-separated list of elements. This defaults to an empty string.

Configuring the integrated DHCP server

Setting static DHCP assignments with the integrated DHCP server

You can set up a static DHCP reservation using the ipv4_address parameter and setting the inventory_dhcp setting to a value of true. This will result in the first MAC address defined in the list of hardware MAC addresses to receive a static address assignment in dnsmasq.

Forcing DNS to resolve to ipv4_address

dnsmasq will resolve all entries to the IP assigned to each server in the leases file. However, this IP will not always be the desired one, if you are working with multiple networks. To force DNS to always resolve to ipv4_address please set the inventory_dns setting to a value of true. This will result in each server to resolve to ipv4_address by explicitly using address capabilities of dnsmasq.

Extending dnsmasq configuration

Bifrost manages the dnsmasq configuration file in /etc/dnsmasq.conf. It is not recommended to make manual modifications to this file after it has been written. dnsmasq supports the use of additional configuration files in /etc/dnsmasq.d, allowing extension of the dnsmasq configuration provided by bifrost. It is possible to use this mechanism provide additional DHCP options to systems managed by ironic, or even to create a DHCP boot environment for systems not managed by ironic. For example, create a file /etc/dnsmasq.d/example.conf with the following contents:

dhcp-match=set:<tag>,<match criteria>
dhcp-boot=tag:<tag>,<boot options>

The tag, match critera and boot options should be modified for your environment. Here we use dnsmasq tags to match against hosts that we want to manage. dnsmasq will use the last matching tagged dhcp-boot option for a host or an untagged default dhcp-boot option if there were no matches. These options will be inserted at the conf-dir=/etc/dnsmasq.d line of the dnsmasq configuration file. Once configured, send the HUP signal to dnsmasq, which will cause it to reread its configuration:

killall -HUP dnsmasq

Using Bifrost with your own DHCP server

The possibility exists that a user may already have a Dynamic Host Configuration Protocol (DHCP) server on their network.

Currently Ironic, when configured with Bifrost in standalone mode, does not utilize a DHCP provider. This would require a manual configuration of the DHCP server to deploy an image. Bifrost utilizes dnsmasq for this functionality; however, any DHCP server can be utilized. This is largely intended to function in the context of a single flat network although conceivably the nodes can be segregated.

What is required:

  • DHCP server on the network segment

  • Appropriate permissions to change DHCP settings

  • Network access to the API and conductor. Keep in mind the iPXE image does not support ICMP redirects.

Example DHCP server configurations

In the examples below port 8080 is used. However, the port number may vary depending on the environment configuration.

dnsmasq:

dhcp-match=set:ipxe,175 # iPXE sends a 175 option.
dhcp-boot=tag:ipxe,http://<Bifrost Host IP Address>:8080/boot.ipxe
dhcp-boot=/undionly.kpxe,<TFTP Server Hostname>,<TFTP Server IP Address>

Internet Systems Consortium DHCPd:

if exists user-class and option user-class = "iPXE" {
      filename "http://<Bifrost Host IP Address>:8080/boot.ipxe";
} else {
      filename "/undionly.kpxe";
      next-server <TFTP Server IP Address>;
}

Architecture

It should be emphasized that Ironic in standalone mode is intended to be used only in a trusted environment.

         +-------------+
         | DHCP Server |
         +-------------+
                |
+--------Trusted-Network----------+
       |                    |
+-------------+       +-----------+
|Ironic Server|       |   Server  |
+-------------+       +-----------+

Use Bifrost with Keystone

Note

Use of keystone with bifrost is a very new feature and should be considered an advanced topic. Please feel free to reach out to the bifrost contributors and the ironic community as a whole in the project’s IRC channel.

Bifrost execution with Keystone

Ultimately, as bifrost was designed for relatively short-lived installations to facilitate rapid hardware deployment, the default operating mode is referred to as noauth mode. With that, in order to leverage keystone authentication for the roles, one of the following steps need to take place.

  1. Update the role defaults for each role you plan to make use. This may not make much sense for most users, unless they are carrying such changes as downstream debt.

  2. Invoke ansible-playbook with variables being set to override the default behavior. Example:

    -e noauth_mode=false -e cloud_name=bifrost
    
  3. Set the global defaults for tagret (master/playbooks/inventory/group_vars/target).

OpenStack Client use with bifrost installed Keystone

A user wishing to invoke OSC commands against the bifrost installation, should set the OS_CLOUD environment variable. An example of setting the environment variable and then executing the OSC command to list all baremetal nodes:

export OS_CLOUD=bifrost
openstack baremetal node list

Keystone roles

Ironic, which is the underlying OpenStack component bifrost helps a user leverage, supports two different roles in keystone that helps govern the rights a user has in keystone.

These roles are baremetal_admin and baremetal_observer and a user can learn more about the roles from the ironic install guide.

Individual playbook use with os-client-config

The OpenStack Ansible modules utilize os-client-config to obtain authentication details to connect to determine details.

If noauth_mode is explicitly disabled, the bifrost roles that speak with Ironic for actions such as enrollment of nodes and deployment, automatically attempt to collect authentication data from os-client-config. Largely these details are governed as environment variables.

That being said, os-client-config supports the concept of clouds and an a user can explicitly select the cloud they wish to deploy to via the cloud_name parameter.