Chapter 6. Composable services and custom roles

The overcloud usually consists of nodes in predefined roles such as Controller nodes, Compute nodes, and different storage node types. Each of these default roles contains a set of services defined in the core heat template collection on the director node. However, you can also create custom roles that contain specific sets of services.

You can use this flexibility to create different combinations of services on different roles. This chapter explores the architecture of custom roles, composable services, and methods for using them.

6.1. Supported role architecture

The following architectures are available when you use custom roles and composable services:

Default architecture
Uses the default roles_data files. All controller services are contained within one Controller role.
Supported standalone roles
Use the predefined files in /usr/share/openstack-tripleo-heat-templates/roles to generate a custom roles_data file`. For more information, see Section 6.4, “Supported custom roles”.
Custom composable services
Create your own roles and use them to generate a custom roles_data file. Note that only a limited number of composable service combinations have been tested and verified and Red Hat cannot support all composable service combinations.

6.2. Examining the roles_data file

The roles_data file contains a YAML-formatted list of the roles that director deploys onto nodes. Each role contains definitions of all of the services that comprise the role. Use the following example snippet to understand the roles_data syntax:

- name: Controller
  description: |
    Controller role that has all the controller services loaded and handles
    Database, Messaging and Network functions.
    - OS::TripleO::Services::AuditD
    - OS::TripleO::Services::CACerts
    - OS::TripleO::Services::CephClient
- name: Compute
  description: |
    Basic Compute Node role
    - OS::TripleO::Services::AuditD
    - OS::TripleO::Services::CACerts
    - OS::TripleO::Services::CephClient

The core heat template collection contains a default roles_data file located at /usr/share/openstack-tripleo-heat-templates/roles_data.yaml. The default file contains definitions of the following role types:

  • Controller
  • Compute
  • BlockStorage
  • ObjectStorage
  • CephStorage.

The openstack overcloud deploy command includes the default roles_data.yaml file during deployment. However, you can use the -r argument to override this file with a custom roles_data file:

$ openstack overcloud deploy --templates -r ~/templates/roles_data-custom.yaml

6.3. Creating a roles_data file

Although you can create a custom roles_data file manually, you can also generate the file automatically using individual role templates. Director provides several commands to manage role templates and automatically generate a custom roles_data file.


  1. List the default role templates:

    $ openstack overcloud roles list
  2. View the role definition in YAML format with the openstack overcloud roles show command:

    $ openstack overcloud roles show Compute
  3. Generate a custom roles_data file. Use the openstack overcloud roles generate command to join multiple predefined roles into a single file. For example, run the following command to generate a roles_data.yaml file that contains the Controller, Compute, and Networker roles:

    $ openstack overcloud roles generate -o ~/roles_data.yaml Controller Compute Networker

    Use the -o option to define the name out of the output file.

    This command creates a custom roles_data file. However, the previous example uses the Controller and Networker roles, which both contain the same networking agents. This means that the networking services scale from the Controller role to the Networker role and the overcloud balances the load for networking services between the Controller and Networker nodes.

    To make this Networker role standalone, you can create your own custom Controller role, as well as any other role that you require. This allows you to generate a roles_data file from your own custom roles.

  4. Copy the directory from the core heat template collection to the home directory of the stack user:

    $ cp -r /usr/share/openstack-tripleo-heat-templates/roles ~/.
  5. Add or modify the custom role files in this directory. Use the --roles-path option with any of the role sub-commands to use this directory as the source for your custom roles:

    $ openstack overcloud roles generate -o my_roles_data.yaml \
      --roles-path ~/roles \
      Controller Compute Networker

    This command generates a single my_roles_data.yaml file from the individual roles in the ~/roles directory.


The default roles collection also contains the ControllerOpenStack role, which does not include services for Networker, Messaging, and Database roles. You can use the ControllerOpenStack in combination with the standalone Networker, Messaging, and Database roles.

6.4. Supported custom roles

The following table contains information about the available custom roles. You can find custom role templates in the /usr/share/openstack-tripleo-heat-templates/roles directory.



OpenStack Block Storage (cinder) node.



Full standalone Ceph Storage node. Includes OSD, MON, Object Gateway (RGW), Object Operations (MDS), Manager (MGR), and RBD Mirroring.



Standalone scale-out Ceph Storage file role. Includes OSD and Object Operations (MDS).



Standalone scale-out Ceph Storage object role. Includes OSD and Object Gateway (RGW).



Ceph Storage OSD node role.



Alternate Compute node role.



DVR enabled Compute node role.



Compute node with hyper-converged infrastructure. Includes Compute and Ceph OSD services.



Compute Instance HA node role. Use in conjunction with the environments/compute-instanceha.yaml` environment file.



Compute node with Cavium Liquidio Smart NIC.



Compute OVS DPDK RealTime role.



Compute OVS DPDK role.



Compute role for ppc64le servers.



Compute role optimized for real-time behaviour. When using this role, it is mandatory that an overcloud-realtime-compute image is available and the role specific parameters IsolCpusList, NovaComputeCpuDedicatedSet and NovaComputeCpuSharedSet are set according to the hardware of the real-time compute nodes.



Compute SR-IOV RealTime role.



Compute SR-IOV role.



Standard Compute node role.



Controller role that does not contain the database, messaging, networking, and OpenStack Compute (nova) control components. Use in combination with the Database, Messaging, Networker, and Novacontrol roles.



Controller role with core Controller services loaded but no Ceph Storage (MON) components. This role handles database, messaging, and network functions but not any Ceph Storage functions.



Controller role that does not contain the OpenStack Compute (nova) control component. Use in combination with the Novacontrol role.



Controller role that does not contain the database, messaging, and networking components. Use in combination with the Database, Messaging, and Networker roles.



Controller role with all core services loaded and uses Ceph NFS. This roles handles database, messaging, and network functions.



Controller role with all core services loaded. This roles handles database, messaging, and network functions.


ControllerSriov (ML2/OVN)

Same as the normal Controller role but with the OVN Metadata agent deployed.



Standalone database role. Database managed as a Galera cluster using Pacemaker.



Compute node with hyper-converged infrastructure and all Ceph Storage services. Includes OSD, MON, Object Gateway (RGW), Object Operations (MDS), Manager (MGR), and RBD Mirroring.



Compute node with hyper-converged infrastructure and Ceph Storage file services. Includes OSD and Object Operations (MDS).



Compute node with hyper-converged infrastructure and Ceph Storage block services. Includes OSD, MON, and Manager.



Compute node with hyper-converged infrastructure and Ceph Storage object services. Includes OSD and Object Gateway (RGW).



Ironic Conductor node role.



Standalone messaging role. RabbitMQ managed with Pacemaker.



Standalone networking role. Runs OpenStack networking (neutron) agents on their own. If your deployment uses the ML2/OVN mechanism driver, see additional steps in Deploying a Custom Role with ML2/OVN.



Same as the normal Networker role but with the OVN Metadata agent deployed. See additional steps in Deploying a Custom Role with ML2/OVN.



Standalone nova-control role to run OpenStack Compute (nova) control agents on their own.



Swift Object Storage node role.



Telemetry role with all the metrics and alarming services.


6.5. Examining role parameters

Each role contains the following parameters:

(Mandatory) The name of the role, which is a plain text name with no spaces or special characters. Check that the chosen name does not cause conflicts with other resources. For example, use Networker as a name instead of Network.
(Optional) A plain text description for the role.

(Optional) A YAML list of tags that define role properties. Use this parameter to define the primary role with both the controller and primary tags together:

- name: Controller
    - primary
    - controller

If you do not tag the primary role, the first role that you define becomes the primary role. Ensure that this role is the Controller role.


A YAML list or dictionary of networks that you want to configure on the role. If you use a YAML list, list each composable network:

    - External
    - InternalApi
    - Storage
    - StorageMgmt
    - Tenant

If you use a dictionary, map each network to a specific subnet in your composable networks.

      subnet: external_subnet
      subnet: internal_api_subnet
      subnet: storage_subnet
      subnet: storage_mgmt_subnet
      subnet: tenant_subnet

Default networks include External, InternalApi, Storage, StorageMgmt, Tenant, and Management.

(Optional) Defines the default number of nodes that you want to deploy for this role.

(Optional) Defines the default hostname format for the role. The default naming convention uses the following format:


For example, the default Controller nodes are named:

(Optional) Defines whether to disable OpenStack Compute (nova) and OpenStack Image Storage (glance) constraints when deploying with director. Use this parameter when you deploy an overcloud with pre-provisioned nodes. For more information, see Configuring a Basic Overcloud with Pre-Provisioned Nodes in the Director Installation and Usage guide.

(Optional) Defines how many nodes to update simultaneously during the OpenStack update options. In the default roles_data.yaml file:

  • The default is 1 for Controller, Object Storage, and Ceph Storage nodes.
  • The default is 25 for Compute and Block Storage nodes.

If you omit this parameter from a custom role, the default is 1.

(Optional) Defines the default list of services to include on the node. For more information, see Section 6.8, “Examining composable service architecture”.

You can use these parameters to create new roles and also define which services to include in your roles.

The openstack overcloud deploy command integrates the parameters from the roles_data file into some of the Jinja2-based templates. For example, at certain points, the overcloud.j2.yaml heat template iterates over the list of roles from roles_data.yaml and creates parameters and resources specific to each respective role.

For example, the following snippet contains the resource definition for each role in the overcloud.j2.yaml heat template:

    type: OS::Heat::ResourceGroup
    depends_on: Networks
      count: {get_param: {{}}Count}
      removal_policies: {get_param: {{}}RemovalPolicies}
        type: OS::TripleO::{{}}
          CloudDomain: {get_param: CloudDomain}
          ServiceNetMap: {get_attr: [ServiceNetMap, service_net_map]}
          EndpointMap: {get_attr: [EndpointMap, endpoint_map]}

This snippet shows how the Jinja2-based template incorporates the {{}} variable to define the name of each role as an OS::Heat::ResourceGroup resource. This in turn uses each name parameter from the roles_data file to name each respective OS::Heat::ResourceGroup resource.

6.6. Creating a new role

You can use the composable service architecture to create new roles according to the requirements of your deployment. For example, you might want to create a new Horizon role to host only the OpenStack Dashboard (horizon).


Role names must start with a letter, end with a letter or digit, and contain only letters, digits, and hyphens. Underscores must never be used in role names.


  1. Create a custom copy of the default roles directory:

    $ cp -r /usr/share/openstack-tripleo-heat-templates/roles ~/.
  2. Create a new file called ~/roles/Horizon.yaml and create a new Horizon role that contains base and core OpenStack Dashboard services:

    - name: Horizon
      CountDefault: 1
      HostnameFormatDefault: '%stackname%-horizon-%index%'
        - OS::TripleO::Services::CACerts
        - OS::TripleO::Services::Kernel
        - OS::TripleO::Services::Ntp
        - OS::TripleO::Services::Snmp
        - OS::TripleO::Services::Sshd
        - OS::TripleO::Services::Timezone
        - OS::TripleO::Services::TripleoPackages
        - OS::TripleO::Services::TripleoFirewall
        - OS::TripleO::Services::SensuClient
        - OS::TripleO::Services::FluentdClient
        - OS::TripleO::Services::AuditD
        - OS::TripleO::Services::Collectd
        - OS::TripleO::Services::MySQLClient
        - OS::TripleO::Services::Apache
        - OS::TripleO::Services::Horizon

    Set the CountDefault to 1 so that a default overcloud always includes the Horizon node.

  3. Optional: If you want to scale the services in an existing overcloud, retain the existing services on the Controller role. If you want to create a new overcloud and you want the OpenStack Dashboard to remain on the standalone role, remove the OpenStack Dashboard components from the Controller role definition:

    - name: Controller
      CountDefault: 1
        - OS::TripleO::Services::GnocchiMetricd
        - OS::TripleO::Services::GnocchiStatsd
        - OS::TripleO::Services::HAproxy
        - OS::TripleO::Services::HeatApi
        - OS::TripleO::Services::HeatApiCfn
        - OS::TripleO::Services::HeatApiCloudwatch
        - OS::TripleO::Services::HeatEngine
        # - OS::TripleO::Services::Horizon                # Remove this service
        - OS::TripleO::Services::IronicApi
        - OS::TripleO::Services::IronicConductor
        - OS::TripleO::Services::Iscsid
        - OS::TripleO::Services::Keepalived
  4. Generate the new roles_data-horizon.yaml file using the ~/roles directory as the source:

    $ openstack overcloud roles generate -o roles_data-horizon.yaml \
      --roles-path ~/roles \
      Controller Compute Horizon
  5. Define a new flavor for this role so that you can tag specific nodes. For this example, use the following commands to create a horizon flavor:

    1. Create a horizon flavor:

      (undercloud)$ openstack flavor create --id auto --ram 6144 --disk 40 --vcpus 4 horizon

      These properties are not used for scheduling instances, however, the Compute scheduler does use the disk size to determine the root partition size.

    2. Tag each bare metal node that you want to designate for the Dashboard service (horizon) with a custom resource class:

      (undercloud)$ openstack baremetal node set --resource-class baremetal.HORIZON <NODE>

      Replace <NODE> with the ID of the bare metal node.

    3. Associate the horizon flavor with the custom resource class:

      (undercloud)$ openstack flavor set --property resources:CUSTOM_BAREMETAL_HORIZON=1 horizon

      To determine the name of a custom resource class that corresponds to a resource class of a bare metal node, convert the resource class to uppercase, replace punctuation with an underscore, and prefix the value with CUSTOM_.


      A flavor can request only one instance of a bare metal resource class.

    4. Set the following flavor properties to prevent the Compute scheduler from using the bare metal flavor properties for scheduling instances:

      (undercloud)$ openstack flavor set --property resources:VCPU=0 --property resources:MEMORY_MB=0 --property resources:DISK_GB=0 horizon
  6. Define the Horizon node count and flavor using the following environment file snippet:

      OvercloudHorizonFlavor: horizon
      HorizonCount: 1
  7. Include the new roles_data-horizon.yaml file and environment file in the openstack overcloud deploy command, along with any other environment files relevant to your deployment:

    $ openstack overcloud deploy --templates -r ~/templates/roles_data-horizon.yaml -e ~/templates/node-count-flavor.yaml

    This configuration creates a three-node overcloud that consists of one Controller node, one Compute node, and one Networker node. To view the list of nodes in your overcloud, run the following command:

    $ openstack server list

6.7. Guidelines and limitations

Note the following guidelines and limitations for the composable role architecture.

For services not managed by Pacemaker:

  • You can assign services to standalone custom roles.
  • You can create additional custom roles after the initial deployment and deploy them to scale existing services.

For services managed by Pacemaker:

  • You can assign Pacemaker-managed services to standalone custom roles.
  • Pacemaker has a 16 node limit. If you assign the Pacemaker service (OS::TripleO::Services::Pacemaker) to 16 nodes, subsequent nodes must use the Pacemaker Remote service (OS::TripleO::Services::PacemakerRemote) instead. You cannot have the Pacemaker service and Pacemaker Remote service on the same role.
  • Do not include the Pacemaker service (OS::TripleO::Services::Pacemaker) on roles that do not contain Pacemaker-managed services.
  • You cannot scale up or scale down a custom role that contains OS::TripleO::Services::Pacemaker or OS::TripleO::Services::PacemakerRemote services.

General limitations:

  • You cannot change custom roles and composable services during a major version upgrade.
  • You cannot modify the list of services for any role after deploying an overcloud. Modifying the service lists after Overcloud deployment can cause deployment errors and leave orphaned services on nodes.

6.8. Examining composable service architecture

The core heat template collection contains two sets of composable service templates:

  • deployment contains the templates for key OpenStack services.
  • puppet/services contains legacy templates for configuring composable services. In some cases, the composable services use templates from this directory for compatibility. In most cases, the composable services use the templates in the deployment directory.

Each template contains a description that identifies its purpose. For example, the deployment/time/ntp-baremetal-puppet.yaml service template contains the following description:

description: >
  NTP service deployment using puppet, this YAML file
  creates the interface between the HOT template
  and the puppet manifest that actually installs
  and configure NTP.

These service templates are registered as resources specific to a Red Hat OpenStack Platform deployment. This means that you can call each resource using a unique heat resource namespace defined in the overcloud-resource-registry-puppet.j2.yaml file. All services use the OS::TripleO::Services namespace for their resource type.

Some resources use the base composable service templates directly:

  OS::TripleO::Services::Ntp: deployment/time/ntp-baremetal-puppet.yaml

However, core services require containers and use the containerized service templates. For example, the keystone containerized service uses the following resource:

  OS::TripleO::Services::Keystone: deployment/keystone/keystone-container-puppet.yaml

These containerized templates usually reference other templates to include dependencies. For example, the deployment/keystone/keystone-container-puppet.yaml template stores the output of the base template in the ContainersCommon resource:

    type: ../containers-common.yaml

The containerized template can then incorporate functions and data from the containers-common.yaml template.

The overcloud.j2.yaml heat template includes a section of Jinja2-based code to define a service list for each custom role in the roles_data.yaml file:

  description: A list of service resources (configured in the heat
               resource_registry) which represent nested stacks
               for each service that should get installed on the {{}} role.
  type: comma_delimited_list
  default: {{role.ServicesDefault|default([])}}

For the default roles, this creates the following service list parameters: ControllerServices, ComputeServices, BlockStorageServices, ObjectStorageServices, and CephStorageServices.

You define the default services for each custom role in the roles_data.yaml file. For example, the default Controller role contains the following content:

- name: Controller
  CountDefault: 1
    - OS::TripleO::Services::CACerts
    - OS::TripleO::Services::CephMon
    - OS::TripleO::Services::CephExternal
    - OS::TripleO::Services::CephRgw
    - OS::TripleO::Services::CinderApi
    - OS::TripleO::Services::CinderBackup
    - OS::TripleO::Services::CinderScheduler
    - OS::TripleO::Services::CinderVolume
    - OS::TripleO::Services::Core
    - OS::TripleO::Services::Kernel
    - OS::TripleO::Services::Keystone
    - OS::TripleO::Services::GlanceApi
    - OS::TripleO::Services::GlanceRegistry

These services are then defined as the default list for the ControllerServices parameter.


You can also use an environment file to override the default list for the service parameters. For example, you can define ControllerServices as a parameter_default in an environment file to override the services list from the roles_data.yaml file.

6.9. Adding and removing services from roles

The basic method of adding or removing services involves creating a copy of the default service list for a node role and then adding or removing services. For example, you might want to remove OpenStack Orchestration (heat) from the Controller nodes.


  1. Create a custom copy of the default roles directory:

    $ cp -r /usr/share/openstack-tripleo-heat-templates/roles ~/.
  2. Edit the ~/roles/Controller.yaml file and modify the service list for the ServicesDefault parameter. Scroll to the OpenStack Orchestration services and remove them:

        - OS::TripleO::Services::GlanceApi
        - OS::TripleO::Services::GlanceRegistry
        - OS::TripleO::Services::HeatApi            # Remove this service
        - OS::TripleO::Services::HeatApiCfn         # Remove this service
        - OS::TripleO::Services::HeatApiCloudwatch  # Remove this service
        - OS::TripleO::Services::HeatEngine         # Remove this service
        - OS::TripleO::Services::MySQL
        - OS::TripleO::Services::NeutronDhcpAgent
  3. Generate the new roles_data file:

    $ openstack overcloud roles generate -o roles_data-no_heat.yaml \
      --roles-path ~/roles \
      Controller Compute Networker
  4. Include this new roles_data file when you run the openstack overcloud deploy command:

    $ openstack overcloud deploy --templates -r ~/templates/roles_data-no_heat.yaml

    This command deploys an overcloud without OpenStack Orchestration services installed on the Controller nodes.


You can also disable services in the roles_data file using a custom environment file. Redirect the services to disable to the OS::Heat::None resource. For example:

  OS::TripleO::Services::HeatApi: OS::Heat::None
  OS::TripleO::Services::HeatApiCfn: OS::Heat::None
  OS::TripleO::Services::HeatApiCloudwatch: OS::Heat::None
  OS::TripleO::Services::HeatEngine: OS::Heat::None

6.10. Enabling disabled services

Some services are disabled by default. These services are registered as null operations (OS::Heat::None) in the overcloud-resource-registry-puppet.j2.yaml file. For example, the Block Storage backup service (cinder-backup) is disabled:

  OS::TripleO::Services::CinderBackup: OS::Heat::None

To enable this service, include an environment file that links the resource to its respective heat templates in the puppet/services directory. Some services have predefined environment files in the environments directory. For example, the Block Storage backup service uses the environments/cinder-backup.yaml file, which contains the following entry:


  1. Add an entry in an environment file that links the CinderBackup service to the heat template that contains the cinder-backup configuration:

      OS::TripleO::Services::CinderBackup: ../podman/services/pacemaker/cinder-backup.yaml

    This entry overrides the default null operation resource and enables the service.

  2. Include this environment file when you run the openstack overcloud deploy command:

    $ openstack overcloud deploy --templates -e /usr/share/openstack-tripleo-heat-templates/environments/cinder-backup.yaml

6.11. Creating a generic node with no services

You can create generic Red Hat Enterprise Linux 8.2 nodes without any OpenStack services configured. This is useful when you need to host software outside of the core Red Hat OpenStack Platform (RHOSP) environment. For example, RHOSP provides integration with monitoring tools such as Kibana and Sensu. For more information, see the Monitoring Tools Configuration Guide. While Red Hat does not provide support for the monitoring tools themselves, director can create a generic Red Hat Enterprise Linux 8.2 node to host these tools.


The generic node still uses the base overcloud-full image rather than a base Red Hat Enterprise Linux 8 image. This means the node has some Red Hat OpenStack Platform software installed but not enabled or configured.


  1. Create a generic role in your custom roles_data.yaml file that does not contain a ServicesDefault list:

    - name: Generic
    - name: Controller
      CountDefault: 1
        - OS::TripleO::Services::AuditD
        - OS::TripleO::Services::CACerts
        - OS::TripleO::Services::CephClient
    - name: Compute
      CountDefault: 1
        - OS::TripleO::Services::AuditD
        - OS::TripleO::Services::CACerts
        - OS::TripleO::Services::CephClient

    Ensure that you retain the existing Controller and Compute roles.

  2. Create an environment file generic-node-params.yaml to specify how many generic Red Hat Enterprise Linux 8 nodes you require and the flavor when selecting nodes to provision:

      OvercloudGenericFlavor: baremetal
      GenericCount: 1
  3. Include both the roles file and the environment file when you run the openstack overcloud deploy command:

    $ openstack overcloud deploy --templates \
    -r ~/templates/roles_data_with_generic.yaml \
    -e ~/templates/generic-node-params.yaml

    This configuration deploys a three-node environment with one Controller node, one Compute node, and one generic Red Hat Enterprise Linux 8 node.