Provisioning Baremetal Before Overcloud Deploy

Baremetal provisioning is a feature which interacts directly with the Bare Metal service to provision baremetal before the overcloud is deployed. This adds a new provision step before the overcloud deploy, and the output of the provision is a valid Using Already Deployed Servers configuration.

In the Wallaby release the baremetal provisining was extended to also manage the neutron API resources for Configuring Network Isolation and Deploying with Custom Networks, and apply network configuration on the provisioned nodes using os-net-config.

Undercloud Components For Baremetal Provisioning

A new YAML file format is introduced to describe the baremetal required for the deployment, and the new command openstack overcloud node provision will consume this YAML and make the specified changes. The provision command interacts with the following undercloud components:

  • A baremetal provisioning workflow which consumes the YAML and runs to completion

  • The metalsmith tool which deploys nodes and associates ports. This tool is responsible for presenting a unified view of provisioned baremetal while interacting with:

    • The Ironic baremetal node API for deploying nodes

    • The Ironic baremetal allocation API which allocates nodes based on the YAML provisioning criteria

    • The Neutron API for managing ports associated with the node’s NICs

In a future release this will become the default way to deploy baremetal, as the Nova compute service and the Glance image service will be switched off on the undercloud.

Baremetal Provision Configuration

A declarative YAML format specifies what roles will be deployed and the desired baremetal nodes to assign to those roles. Defaults can be relied on so that the simplest configuration is to specify the roles, and the count of baremetal nodes to provision for each role

- name: Controller
  count: 3
- name: Compute
  count: 100

Often it is desirable to assign specific nodes to specific roles, and this is done with the instances property

- name: Controller
  count: 3
  instances:
  - hostname: overcloud-controller-0
    name: node00
  - hostname: overcloud-controller-1
    name: node01
  - hostname: overcloud-controller-2
    name: node02
- name: Compute
  count: 100
  instances:
  - hostname: overcloud-novacompute-0
    name: node04

Here the instance name refers to the logical name of the node, and the hostname refers to the generated hostname which is derived from the overcloud stack name, the role, and an incrementing index. In the above example, all of the Controller servers are on predictable nodes, as well as one of the Compute servers. The other 99 Compute servers are on nodes allocated from the pool of available nodes.

The properties in the instances entries can also be set in the defaults section so that they do not need to be repeated in every entry. For example, the following are equivalent

- name: Controller
  count: 3
  instances:
  - hostname: overcloud-controller-0
    name: node00
    image:
      href: overcloud-full-custom
  - hostname: overcloud-controller-1
    name: node01
    image:
      href: overcloud-full-custom
  - hostname: overcloud-controller-2
    name: node02
    image:
      href: overcloud-full-custom

- name: Controller
  count: 3
  defaults:
    image:
      href: overcloud-full-custom
  instances:
  - hostname: overcloud-controller-0
    name: node00
  - hostname: overcloud-controller-1
    name: node01
  - hostname: overcloud-controller-2
    name: node02

When using Configuring Network Isolation, Deploying with Custom Networks or a combination of the two the networks and network_configuration must either be set in the defaults for the role or for each specific node (instance). The following example extends the first simple configuration example adding typical TripleO network isolation by setting defaults for each role

- name: Controller
  count: 3
  defaults:
    networks:
    - network: ctlplane
      vif: true
    - network: external
      subnet: external_subnet
    - network: internalapi
      subnet: internal_api_subnet01
    - network: storage
      subnet: storage_subnet01
    - network: storagemgmt
      subnet: storage_mgmt_subnet01
    - network: tenant
      subnet: tenant_subnet01
    network_config:
      template: /home/stack/nic-config/controller.j2
      default_route_network:
      - external
- name: Compute
  count: 100
  defaults:
    networks:
    - network: ctlplane
      vif: true
    - network: internalapi
      subnet: internal_api_subnet02
    - network: tenant
      subnet: tenant_subnet02
    - network: storage
      subnet: storage_subnet02
    network_config:
      template: /home/stack/nic-config/compute.j2

Role Properties

Each role entry supports the following properties:

  • name: Mandatory role name

  • hostname_format: Override the default hostname format for this role. The default format uses the lower case role name, so for the Controller role the default format is %stackname%-controller-%index%. Only the Compute role doesn’t follow the role name rule, the Compute default format is %stackname%-novacompute-%index%

  • count: Number of nodes to provision for this role, defaults to 1

  • defaults: A dict of default values for instances entry properties. An instances entry property will override a default specified here See Instance and Defaults Properties for supported properties

  • instances: A list of dict for specifying attributes for specific nodes. See Instance and Defaults Properties for supported properties. The length of this list must not be greater than count

  • ansible_playbooks: A list of dict for Ansible playbooks and Ansible vars, the playbooks are run against the role instances after node provisioning, prior to the node network configuration. See Ansible Playbooks for more details and examples.

Instance and Defaults Properties

These properties serve three purposes:

  • Setting selection criteria when allocating nodes from the pool of available nodes

  • Setting attributes on the baremetal node being deployed

  • Setting network configuration properties for the deployed nodes

Each instances entry and the defaults dict support the following properties:

  • capabilities: Selection criteria to match the node’s capabilities

  • config_drive: Add data and first-boot commands to the config-drive passed to the node. See Config Drive

  • hostname: If this complies with the hostname_format pattern then other properties will apply to the node allocated to this hostname. Otherwise, this allows a custom hostname to be specified for this node. (Cannot be specified in defaults)

  • image: Image details to deploy with. See Image Properties

  • managed: Boolean to determine whether the instance is actually provisioned with metalsmith, or should be treated as preprovisioned.

  • name: The name of a node to deploy this instance on (Cannot be specified in defaults)

  • nics: (DEPRECATED: Replaced by networks in Wallaby) List of dicts representing requested NICs. See Nics Properties

  • networks: List of dicts representing instance networks. See Networks Properties

  • network_config: Network configuration details. See Network Config Properties

  • profile: Selection criteria to use Node matching with resource classes and profiles

  • provisioned: Boolean to determine whether this node is provisioned or unprovisioned. Defaults to true, false is used to unprovision a node. See Scaling Down

  • resource_class: Selection criteria to match the node’s resource class, defaults to baremetal. See Node matching with resource classes and profiles

  • root_size_gb: Size of the root partition in GiB, defaults to 49

  • swap_size_mb: Size of the swap partition in MiB, if needed

  • traits: A list of traits as selection criteria to match the node’s traits

Image Properties

  • href: Glance image reference or URL of the root partition or whole disk image. URL schemes supported are file://, http://, and https://. If the value is not a valid URL, it is assumed to be a Glance image reference

  • checksum: When the href is a URL, the MD5 checksum of the root partition or whole disk image

  • kernel: Glance image reference or URL of the kernel image (partition images only)

  • ramdisk: Glance image reference or URL of the ramdisk image (partition images only)

Networks Properties

The instances networks property supports a list of dicts, one dict per network.

  • network: Neutron network to create the port for this network:

  • fixed_ip: Specific IP address to use for this network

  • network: Neutron network to create the port for this network

  • subnet: Neutron subnet to create the port for this network

  • port: Existing Neutron port to use instead of creating one

  • vif: When true the network is attached as VIF (virtual-interface) by metalsmith/ironic. When false the baremetal provisioning workflow creates the Neutron API resource, but no VIF attachment happens in metalsmith/ironic. (Typically only the provisioning network (ctlplane) has this set to true.)

By default there is one network representing

- network: ctlplane
  vif: true

Other valid network entries would be

- network: ctlplane
  fixed_ip: 192.168.24.8
  vif: true
- port: overcloud-controller-0-ctlplane
- network: internal_api
  subnet: internal_api_subnet01

Network Config Properties

The network_config property contains os-net-config related properties.

  • template: The ansible j2 nic config template to use when applying node network configuration. (default: templates/net_config_bridge.j2)

  • physical_bridge_name: Name of the OVS bridge to create for accessing external networks. (default: br-ex)

  • public_interface_name: Which interface to add to the public bridge (default: nic1)

  • network_config_update: Whether to apply network configuration changes, on update or not. Boolean value. (default: false)

  • net_config_data_lookup: Per node and/or per node group os-net-config nic mapping config.

  • default_route_network: The network to use for the default route (default: ctlplane)

  • networks_skip_config: List of networks that should be skipped when configuring node networking

  • dns_search_domains: A list of DNS search domains to be added (in order) to resolv.conf.

  • bond_interface_ovs_options: The ovs_options or bonding_options string for the bond interface. Set things like lacp=active and/or bond_mode=balance-slb for OVS bonds or like mode=4 for Linux bonds using this option.

Nics Properties

The instances nics property supports a list of dicts, one dict per NIC.

  • fixed_ip: Specific IP address to use for this NIC

  • network: Neutron network to create the port for this NIC

  • subnet: Neutron subnet to create the port for this NIC

  • port: Existing Neutron port to use instead of creating one

By default there is one NIC representing

- network: ctlplane

Other valid NIC entries would be

- subnet: ctlplane-subnet
  fixed_ip: 192.168.24.8
- port: overcloud-controller-0-ctlplane

Config Drive

The instances config_drive property supports two sub-properties:

  • cloud_config: Dict of cloud-init cloud-config data for tasks to run on node boot. A task specified in an instances cloud_config will overwrite a task with the same name in in defaults cloud_config.

  • meta_data: Extra metadata to include with the config-drive cloud-init metadata. This will be added to the generated metadata public_keys, uuid, name, hostname, and instance-type which is set to the role name. Cloud-init makes this metadata available as instance-data. A key specified in an instances meta_data entry will overwrite the same key in defaults meta_data.

Below are some examples of what can be done with config_drive.

Run arbitrary scripts on first boot:

config_drive:
  cloud_config:
    bootcmd:
      # temporary workaround to allow steering in ConnectX-3 devices
      - echo "options mlx4_core log_num_mgm_entry_size=-1" >> /etc/modprobe.d/mlx4.conf
      - /sbin/dracut --force

Enable and configure ntp:

config_drive:
  cloud_config:
    enabled: true
    ntp_client: chrony  # Uses cloud-init default chrony configuration

Allow root ssh login (for development environments only):

config_drive:
  cloud_config:
    ssh_pwauth: true
    disable_root: false
    chpasswd:
      list: |-
        root:sekrit password
      expire: False

Use values from custom metadata:

config_drive:
  meta_data:
    foo: bar
  cloud_config:
    runcmd:
      - echo The value of foo is `jq .foo < /run/cloud-init/instance-data.json`

Ansible Playbooks

The role ansible_playbooks takes a list of playbook definitions, supporting the playbook and extra_vars sub-properties.

  • playbook: The path (relative to the roles definition YAML file) to the ansible playbook.

  • extra_vars: Extra Ansible variables to set when running the playbook.

Note

Playbooks only run if ‘–network-config’ is enabled.

Run arbitrary playbooks:

ansible_playbooks:
  - playbook: a_playbook.yaml
  - playbook: b_playbook.yaml

Run arbitrary playbooks with extra variables defined for one of the playbooks:

ansible_playbooks:
  - playbook: a_playbook.yaml
    extra_vars:
      param1: value1
      param2: value2
  - playbook: b_playbook.yaml

Grow volumes playbook

After custom playbooks are run, an in-built playbook is run to grow the LVM volumes of any node deployed with the whole-disk overcloud image overcloud-hardened-uefi-full.qcow2. The implicit ansible_playbooks would be:

ansible_playbooks:
  - playbook: /usr/share/ansible/tripleo-playbooks/cli-overcloud-node-growvols.yaml
    extra_vars:
      growvols_args: >
        /=8GB
        /tmp=1GB
        /var/log=10GB
        /var/log/audit=2GB
        /home=1GB
        /var=100%

Each LVM volume is grown by the amount specified until the disk is 100% allocated, and any remaining space is given to the / volume. In some cases it may be necessary to specify different growvols_args. For example the ObjectStorage role deploys swift storage which stores state in /srv, so this volume needs the remaining space instead of /var. The playbook can be explicitly written to override the default growvols_args value, for example:

ansible_playbooks:
  - playbook: /usr/share/ansible/tripleo-playbooks/cli-overcloud-node-growvols.yaml
    extra_vars:
      growvols_args: >
        /=8GB
        /tmp=1GB
        /var/log=10GB
        /var/log/audit=2GB
        /home=1GB
        /var=1GB
        /srv=100%

Set kernel arguments playbook

Features such as DPDK require that kernel arguments are set and the node is rebooted before the network configuration is run. A playbook is provided to allow this. Here it is run with the default variables set:

ansible_playbooks:
  - playbook: /usr/share/ansible/tripleo-playbooks/cli-overcloud-node-kernelargs.yaml
    extra_vars:
      kernel_args: ''
      reboot_wait_timeout: 900
      defer_reboot: false
      tuned_profile: 'throughput-performance'
      tuned_isolated_cores: ''

Here is an example for a specific DPDK deployment:

ansible_playbooks:
  - playbook: /usr/share/ansible/tripleo-playbooks/cli-overcloud-node-kernelargs.yaml
    extra_vars:
      kernel_args: 'default_hugepagesz=1GB hugepagesz=1G hugepages=64 intel_iommu=on iommu=pt'
      tuned_isolated_cores: '1-11,13-23'
      tuned_profile: 'cpu-partitioning'

Deploying the Overcloud

This example assumes that the baremetal provision configuration file has the filename ~/overcloud_baremetal_deploy.yaml and the resulting deployed server environment file is ~/overcloud-baremetal-deployed.yaml. It also assumes overcloud networks are pre-deployed using the openstack overcloud network provision command and the deployed networks environment file is ~/overcloud-networks-deployed.yaml.

The baremetal nodes are provisioned with the following command:

openstack overcloud node provision \
  --stack overcloud \
  --network-config \
  --output ~/overcloud-baremetal-deployed.yaml \
  ~/overcloud_baremetal_deploy.yaml

Note

Removing the --network-config argument will disable the management of non-VIF networks and post node provisioning network configuration with os-net-config via ansible.

The overcloud can then be deployed using the output from the provision command:

openstack overcloud deploy \
  -e /usr/share/openstack-tripleo-heat-templates/environments/deployed-server-environment.yaml \
  -e ~/overcloud-networks-deployed.yaml \
  -e ~/templates/vips-deployed-environment.yaml \
  -e ~/overcloud-baremetal-deployed.yaml \
  --deployed-server \
  --disable-validations \ # optional, see note below
  # other CLI arguments

Note

The validation which is part of openstack overcloud node provision may fail with the default overcloud image unless the Ironic node has more than 4 GB of RAM. For example, a VBMC node provisioned with 4096 MB of memory failed because the image size plus the reserved RAM size were not large enough (Image size: 4340 MiB, Memory size: 3907 MiB).

Viewing Provisioned Node Details

The commands baremetal node list and baremetal node show continue to show the details of all nodes, however there are some new commands which show a further view of the provisioned nodes.

The metalsmith tool provides a unified view of provisioned nodes, along with allocations and neutron ports. This is similar to what Nova provides when it is managing baremetal nodes using the Ironic driver. To list all nodes managed by metalsmith, run:

metalsmith list

The baremetal allocation API keeps an association of nodes to hostnames, which can be seen by running:

baremetal allocation list

The allocation record UUID will be the same as the Instance UUID for the node which is allocated. The hostname can be seen in the allocation record, but it can also be seen in the baremetal node show property instance_info, display_name.

Scaling the Overcloud

Scaling Up

To scale up an existing overcloud, edit ~/overcloud_baremetal_deploy.yaml to increment the count in the roles to be scaled up (and add any desired instances entries) then repeat the Deploying the Overcloud steps.

Scaling Down

Scaling down is done with the openstack overcloud node delete command but the nodes to delete are not passed as command arguments.

To scale down an existing overcloud edit ~/overcloud_baremetal_deploy.yaml to decrement the count in the roles to be scaled down, and also ensure there is an instances entry for each node being unprovisioned which contains the following:

  • The name of the baremetal node to remove from the overcloud

  • The hostname which is assigned to that node

  • A provisioned: false property

  • A YAML comment explaining the reason for making the node unprovisioned (optional)

For example the following would remove overcloud-compute-1

- name: Compute
  count: 1
  instances:
  - hostname: overcloud-compute-0
    name: node10
    # Removed from deployment due to disk failure
    provisioned: false
  - hostname: overcloud-compute-1
    name: node11

Then the delete command will be called with --baremetal-deployment instead of passing node arguments:

openstack overcloud node delete \
--stack overcloud \
--baremetal-deployment ~/overcloud_baremetal_deploy.yaml

Before any node is deleted, a list of nodes to delete is displayed with a confirmation prompt.

What to do when scaling back up depends on the situation. If the scale-down was to temporarily remove baremetal which is later restored, then the scale-up can increment the count and set provisioned: true on nodes which were previously provisioned: false. If that baremetal node is not going to be re-used in that role then the provisioned: false can remain indefinitely and the scale-up can specify a new instances entry, for example

- name: Compute
  count: 2
  instances:
  - hostname: overcloud-compute-0
    name: node10
    # Removed from deployment due to disk failure
    provisioned: false
  - hostname: overcloud-compute-1
    name: node11
  - hostname: overcloud-compute-2
    name: node12

Unprovisioning All Nodes

After openstack overcloud delete is called, all of the baremetal nodes can be unprovisioned without needing to edit ~/overcloud_baremetal_deploy.yaml by running the unprovision command with the --all argument:

openstack overcloud node unprovision --all \
  --stack overcloud \
  --network-ports \
  ~/overcloud_baremetal_deploy.yaml

Note

Removing the --network-ports argument will disable the management of non-VIF networks, non-VIF ports will _not_ be deleted in that case.