Authoring Site Topology

Drydock uses a YAML-formatted site topology definition to configure downstream drivers to provision baremetal nodes. This topology describes the networking configuration of a site as well as the set of node configurations that will be deployed. A node configuration consists of network attachment, network addressing, local storage, kernel selection and configuration and metadata.

The best source for a sample of the YAML schema for a topology is the unit test input source in ./tests/yaml_samples/fullsite.yaml.

Defining Networking

Network definitions in the topology are described by two document types: NetworkLink and Network. NetworkLink describes a physical or logical link between a node and switch. It is concerned with attributes that must be agreed upon by both endpoints: bonding, media speed, trunking, etc. A Network describes the layer 2 and layer 3 networks accessible over a link.


The Network document defines the layer 2 and layer 3 networks nodes will access. Each Network is accessible over exactly one NetworkLink. However that NetworkLink can be attached to different interfaces on different nodes to support changing hardware configurations.

Example YAML schema of the Network spec:

  vlan: '102'
  mtu: 1500
  routedomain: storage
    - type: static
    - type: dhcp
    - subnet:
      metric: 10
    - gateawy:
      metric: 10
      routedomain: storage

If a Network is accessible over a NetworkLink using 802.1q VLAN tagging, the vlan attribute specified the VLAN tag for this Network. It should be omitted for non-tagged Networks.

mtu is the maximum transmission unit for this Network. Must be equal or less than the mtu defined for the hosting NetworkLink. Can be omitted to default to the NetworkLink mtu.

cidr is the classless inter-domain routing address for the network.

routedomain is a logical grouping of L3 networks such that a network that describes a static route for accessing the route domain will yield a list of static routes for all the networks in the routedomain. See the description of routes below for more information.

ranges defines a sequence of IP addresses within the defined cidr. Ranges cannot overlap.

  • type: The type of address range.

    • static: A range used for static, explicit address assignments for nodes.
    • dhcp: A range used for assigning DHCP addresses. Note that a network being used for PXE booting must have a DHCP range defined.
    • reserved: A range of addresses that will not be used by MaaS.
  • start: The starting IP of the range, inclusive.

  • end: The last IP of the range, inclusive

routes defines a list of static routes to be configured on nodes attached to this network. The routes can defined in one of two ways: an explicit destination subnet where the route will be configured exactly as described or a destination routedomain where Drydock will calculate all the destination L3 subnets for the routedomain and add routes for each of them using the gateway and metric defined.

  • subnet: Destination CIDR for the route
  • gateway: The gateway IP on this Network to use for accessing the destination
  • metric: The metric or weight for this route
  • routedomain: Use this route’s gateway and metric for accessing networks in the
    defined routedomain.

dns is used for specifying the list of DNS servers to use if this network is the primary network for the node.

  • servers: A comma-separated list of IP addresses to use for DNS resolution
  • domain: A domain that can be used for automated registration of IP addresses assigned from this Network

DHCP Relay

DHCP relaying is used when a DHCP server is not attached to the same layer 2 broadcast domain as nodes that are being PXE booted. The DHCP requests from the node are consumed by the relay (generally configured on a top-of-rack switch) which then encapsulates the request in layer 3 routing and sends it to an upstream DHCP server. The Network spec supports a dhcp_relay key for Networks that should relay DHCP requests.

  • The Network must have a configured DHCP relay, this is not configured by Drydock or MaaS.
  • The upstream_target IP address must be a host IP address for a MaaS rack controller
  • The Network must have a defined DHCP address range.
  • The upstream target network must have a defined DHCP address range.

The dhcp_relay stanza:


Defining Node Configuration

Node configuration is defined in three documents: HostProfile, HardwareProfile and BaremetalNode. HardwareProfile defines attributes directly related to hardware configuration such as card-slot layout and firmware levels. HostProfile is a generic definition for how a node should be configured such that many nodes can reference a single HostProfile and each will be configured identically. A BaremetalNode is a concrete reference to the particular physical node. The BaremetalNode definition will reference a HostProfile and can then extend or override any of the configuration values.

NOTE: Drydock does not support hostnames containing ‘__’ (double underscore)

Hardware Profile

The hardware profile is used to convert some abstractions in the HostProfile documents into concrete configurations based a particular hardware build. A host profile will designate how the bootdisk should be configured, but the hardware profile will designate which exact device is used for the bootdisk. This allows a heterogeneous mix of hardware in a site without duplicating definitions of how that hardware should be configured.

An example HardwareProfile document:

schema: 'drydock/HardwareProfile/v1'
  schema: 'metadata/Document/v1'
  name: AcmeServer
  storagePolicy: 'cleartext'
    application: 'drydock'
  vendor: HP
  generation: '8'
  hw_version: '3'
  bios_version: '2.2.3'
  boot_mode: bios
  bootstrap_protocol: pxe
  pxe_interface: 0
      address: '0000:00:03.0'
      dev_type: '82540EM Gigabit Ethernet Controller'
      bus_type: 'pci'
      address: '0000:00:04.0'
      dev_type: '82540EM Gigabit Ethernet Controller'
      bus_type: 'pci'
      address: '2:0.0.0'
      dev_type: 'VBOX HARDDISK'
      bus_type: 'scsi'
    sriov: '2,4'
      size: '1G'
      count: 300
      size: '2M'
      count: 530000

Device Aliases

Device aliases are a way of mapping a particular device bus address to an alias. In the example above we map the PCI address 0000:00:03.0 to the alias prim_nic01. A host profile or baremetal node definition can then provide a configuration using prim_nic01 and Drydock will translate that to the correct operating system device name for the NIC device at PCI address 0000.00.03.0. Currently device aliases are supported for network interface slave devices and storage physical devices.

Kernel Parameter References

Some kernel parameters specified in a host profile rely on particular hardware builds, such as isolcpus. To support the greatest flexibility in building host profiles, you can specify a few values in a hardware profile that will then be sourced when needed by a host profile or baremetal node definition.

  • cpu_sets: Each key should have a value of a comma-separated list of CPUs/cores/hyperthreads that would be appropriate for the isolcpus kernel parameters. A host profile can then select any one of these CPU sets for a host.
  • hugepages: Each key should have a value of a mapping containing two keys: size and count. Again, a host profile can then select these values when defining kernel parameters for a host. Note the size field is a string and will be used as-is, so the format must be usable by the kernel.

Host Profiles and Baremetal Nodes

Example HostProfile and BaremetalNode configuration:

apiVersion: 'drydock/v1'
kind: HostProfile
  name: defaults
  region: sitename
  date: 17-FEB-2017
  # configuration values
apiVersion: 'drydock/v1'
kind: HostProfile
  name: compute_node
  region: sitename
  date: 17-FEB-2017
  host_profile: defaults
  # compute_node customizations to defaults
apiVersion: 'drydock/v1'
kind: BaremetalNode
  name: compute01
  region: sitename
  date: 17-FEB-2017
  host_profile: compute_node
  # configuration customization specific to single node compute01

In the above example, the compute_node HostProfile adopts all values from the defaults HostProfile and can then override defined values or append additional values. BaremetalNode compute01 then adopts all values from the compute_node HostProfile (which includes all the configuration items it adopted from defaults) and can then again override or append any configuration that is specific to that node.

Defining Node Out-Of-Band Management

Drydock supports plugin-based OOB management. At a minimum a OOB driver supports configuring a node to PXE boot during the next boot cycle and power cycling the node to initiate the provisioning process. Richer features might also be supported such as BIOS configuration or BMC log analysis. The value of oob.type in the host profile or baremetal node definition will define what additional parameters are required for that type and what capabilities are available via OOB driver tasks.


The ipmi OOB type requires additional configuration to allow OOB management:

  1. The oob parameters account and credential must be populated with a valid account and password that can access the BMC via IPMI over LAN.
  2. The oob parameter network must reference which node network is used for OOB access.
  3. The addressing section of the node definition must contain an IP address assignment for the network referenced in

Currently the IPMI driver supports only basic management by setting nodes to PXE boot and power-cycling the node.


The libvirt OOB type requires additional configuration within the site definition as well as particular configuration in the deployment of Drydock (and likely the node provisioning driver.):

  1. A SSH public/private key-pair should be generated with the public key being added to the authorized_keys file on all hypervisors hosting libvirt-based VMs being deployed. The account for this must be in the libvirt group.
  2. The private key should be provided in the Drydock and MAAS charts as an override to conf.ssh.private_key
  3. The Drydock and MAAS chart should override manifests.secret_ssh_key: true.
  4. In the site definition, each libvirt-based node must define oob parameter libvirt_uri of the form qemu+ssh://account@hostname/system where account is an account in the libvirt group on the hypervisor with an authorized_key and hostname is an IP address or FQDN for the hypervisor hosting the VM.

Currently the Libvirt driver supports only basic management by setting nodes to PXE boot and power-cycling the node.

Defining Node Interfaces and Network Addressing

Node network attachment can be described in a HostProfile or a BaremetalNode document. Node addressing is allowed only in a BaremetalNode document. If a HostProfile or BaremetalNode needs to remove a defined interface from an inherited configuration, it can set the mapping value for the interface name to null.

Once the interface attachments to networks is defined, HostProfile and BaremetalNode specs must define a primary_network attribute to denote which network the node should use as the primary route.


Interfaces for a node can be described in either a HostProfile or BaremetalNode definition. This will attach a defined NetworkLink to a host interface and define which Networks should be configured to use that interface.

Example interface definition YAML schema:

    device_link: pxe
      pxe: true
      - prim_nic01
      - pxe
    device_link: gp
      - prim_nic01
      - prim_nic02
      - mgmt
      - private

Each key in the interfaces mapping is a defined interface. The key is the name that will be used on the deployed node for the interface. The value must be a mapping defining the interface configuration or null to denote removal of that interface for an inherited configuration.

  • device_link: The name of the defined NetworkLink that will be attached to this interface. The NetworkLink definition includes part of the interface configuration such as bonding.
  • labels: Metadata for describing this interface.
  • slaves: The list of hardware interfaces used for creating this interface. This value can be a device alias defined in the HardwareProfile or the kernel name of the hardware interface. For bonded interfaces, this would list all the slaves. For non-bonded interfaces, this should list the single hardware interface used.
  • networks: This is the list of networks to enable on this interface. If multiple networks are listed, the NetworkLink attached to this interface must have trunking enabled or the design validation will fail.

Addressing for a node can only be defined in a BaremetalNode definition. The addressing stanza simply defines a static IP address or dhcp for each network a node should have a configured layer 3 interface on. It is a valid design to omit networks from the addressing stanza, in that case the interface attached to the omitted network will be configured as link up with no address.

Example addressing YAML schema:

  - network: pxe
    address: dhcp
  - network: mgmt
  - network: private
  - network: oob

Defining Node Storage

Storage can be defined in the storage stanza of either a HostProfile or BaremetalNode document. The storage configuration can describe the creation of partitions on physical disks, the assignment of physical disks and/or partitions to volume groups, and the creation of logical volumes. Drydock will make a best effort to parse out system-level storage such as the root filesystem or boot filesystem and take appropriate steps to configure them in the active node provisioning driver. At a minimum, the storage configuration must contain a root filesystem partition.

Example YAML schema of the storage stanza:

        bootdrive: true
        - name: 'root'
          size: '10g'
          bootable: true
            mountpoint: '/'
            fstype: 'ext4'
            mount_options: 'defaults'
        - name: 'boot'
          size: '1g'
            mountpoint: '/boot'
            fstype: 'ext4'
            mount_options: 'defaults'
      volume_group: 'log_vg'
        - name: 'log_lv'
          size: '500m'
            mountpoint: '/var/log'
            fstype: 'xfs'
            mount_options: 'defaults'

The storage stanza can contain two top-level keys: physical_devices and volume_groups. The latter is optional.

Physical Devices and Partitions

A physical device can either be carved up in partitions (including a single partition consuming the entire device) or added to a volume group as a physical volume. Each key in the physical_devices mapping represents a device on a node. The key should either be a device alias defined in the HardwareProfile or the name of the device published by the OS. The value of each key must be a mapping with the following keys

  • labels: A mapping of key/value strings providing generic labels for the device
  • partitions: A sequence of mappings listing the partitions to be created on the device. The mapping is described below. Incompatible with the volume_group specification.
  • volume_group: A volume group name to add the device to as a physical volume. Incompatible with the partitions specification.

A partition mapping describes a GPT partition on a physical disk. It can be left as a raw block device or formatted and mounted as a filesystem.

  • name: Metadata describing the partition in the topology

  • size: The size of the partition. See the Size Format section below

  • bootable: Boolean whether this partition should be the bootable device

  • part_uuid: A UUID4 formatted UUID to assign to the partition. If not specified one will be generated

  • filesystem: An optional mapping describing how the partition should be formatted and mounted

    • mountpoint: Where the filesystem should be mounted. If not specified the partition will be left as a raw device
    • fstype: The format of the filesystem. Defaults to ext4
    • mount_options: fstab style mount options. Default is ‘defaults’
    • fs_uuid: A UUID4 formatted UUID to assign to the filesystem. If not specified one will be generated
    • fs_label: A filesystem label to assign to the filesystem. Optional.
Size Format

The size specification for a partition or logical volume is formed from three parts:

  • The first character can optionally be > indicating that the size specified is a minimum and the calculated size should be at least the minimum and should take the rest of the available space on the physical device or volume group.

  • The second part is the numeric portion and must be an integer

  • The third part is a label

    • m|M|mb|MB: Megabytes or 10^6 * the numeric
    • g|G|gb|GB: Gigabytes or 10^9 * the numeric
    • t|T|tb|TB: Terabytes or 10^12 * the numeric
    • %: The percentage of total device or volume group space
Volume Groups and Logical Volumes

Logical volumes can be used to create RAID-0 volumes spanning multiple physical disks or partitions. Each key in the volume_groups mapping is a name assigned to a volume group. This name must be specified as the volume_group attribute on one or more physical devices or partitions or the configuration is invalid. Each mapping value is another mapping describing the volume group.

  • vg_uuid: A UUID4 format uuid applied to the volume group. If not specified, one is generated
  • logical_volumes: A sequence of mappings listing the logical volumes to be created in the volume group
Logical Volume

A logical volume is a RAID-0 volume. Using logical volumes for / and /boot is supported

  • name: Required field. Used as the logical volume name.
  • size: The logical volume size. See Size Format above for details.
  • lv_uuid: A UUID4 format uuid applied to the logical volume: If not specified, one is generated
  • filesystem: A mapping specifying how the logical volume should be formatted and mounted. See the Partition section above for filesystem details.

Platform Configuration

In the platform stanza you can define the operating system image and kernel to use as well as customize the kernel configuration with kernel_params.

The valid image and kernel values are dependent on what is supported by your node provisioner. In the example of Canonical MaaS using the 16.04 LTS image, the values would be image: 'xenial' and kernel: 'ga-16.04' for the LTS kernel or kernel: hwe-16.04 for the hardware-enablement kernel.

The kernel_params configuration is a mapping. Each key should either be a string or boolean value. For boolean true values, the key will be added to the kernel parameter list as a flag. For string values, the key:value pair will be added to the kernel parameter list as key=value.

One special case is supported for values that match a hardware profile reference. When the parameter is rendered for a particular node, the value included in the kernel parameter list will be sourced from the effective HardwareProfile assigned to the node.

  • hardwareprofile:cpuset.<name>: Sourced from the hardware profile cpu_sets.<name> value.
  • hardwareprofile.hugepages.<name>.size: Source from the hardware profile hugepages.<name>.size value.
  • hardwareprofile.hugepages.<name>.count: Source from the hardware profile hugepages.<name>.count value.