Chapter 9. Configuring a basic overcloud with pre-provisioned nodes

This chapter contains basic configuration procedures that you can use to configure a Red Hat OpenStack Platform (RHOSP) environment with pre-provisioned nodes. This scenario differs from the standard overcloud creation scenarios in several ways:

  • You can provision nodes with an external tool and let the director control the overcloud configuration only.
  • You can use nodes without relying on the director provisioning methods. This is useful if you want to create an overcloud without power management control, or use networks with DHCP/PXE boot restrictions.
  • The director does not use OpenStack Compute (nova), OpenStack Bare Metal (ironic), or OpenStack Image (glance) to manage nodes.
  • Pre-provisioned nodes can use a custom partitioning layout that does not rely on the QCOW2 overcloud-full image.

This scenario includes only basic configuration with no custom features. However, you can add advanced configuration options to this basic overcloud and customize it to your specifications with the instructions in the Advanced Overcloud Customization guide.

Important

You cannot combine pre-provisioned nodes with director-provisioned nodes.

9.1. Pre-provisioned node requirements

Before you begin deploying an overcloud with pre-provisioned nodes, ensure that the following configuration is present in your environment:

  • The director node that you created in Chapter 4, Installing director.
  • A set of bare metal machines for your nodes. The number of nodes required depends on the type of overcloud you intend to create. These machines must comply with the requirements set for each node type. These nodes require Red Hat Enterprise Linux 8.1 or later installed as the host operating system. Red Hat recommends using the latest version available.
  • One network connection for managing the pre-provisioned nodes. This scenario requires uninterrupted SSH access to the nodes for orchestration agent configuration.
  • One network connection for the Control Plane network. There are two main scenarios for this network:

    • Using the Provisioning Network as the Control Plane, which is the default scenario. This network is usually a layer-3 (L3) routable network connection from the pre-provisioned nodes to director. The examples for this scenario use following IP address assignments:

      Table 9.1. Provisioning Network IP assignments

      Node nameIP address

      Director

      192.168.24.1

      Controller 0

      192.168.24.2

      Compute 0

      192.168.24.3

    • Using a separate network. In situations where the director’s Provisioning network is a private non-routable network, you can define IP addresses for nodes from any subnet and communicate with director over the Public API endpoint. For more information about the requirements for this scenario, see Section 9.6, “Using a separate network for pre-provisioned nodes”.
  • All other network types in this example also use the Control Plane network for OpenStack services. However, you can create additional networks for other network traffic types.
  • If any nodes use Pacemaker resources, the service user hacluster and the service group haclient must have a UID/GID of 189. This is due to CVE-2018-16877. If you installed Pacemaker together with the operating system, the installation creates these IDs automatically. If the ID values are set incorrectly, follow the steps in the article OpenStack minor update / fast-forward upgrade can fail on the controller nodes at pacemaker step with "Could not evaluate: backup_cib" to change the ID values.
  • To prevent some services from binding to an incorrect IP address and causing deployment failures, make sure that the /etc/hosts file does not include the node-name=127.0.0.1 mapping.

9.2. Creating a user on pre-provisioned nodes

When you configure an overcloud with pre-provisioned nodes, director requires SSH access to the overcloud nodes as the stack user. To create the stack user, complete the following steps.

Procedure

  1. On each overcloud node, create the stack user and set a password. For example, run the following commands on the Controller node:

    [root@controller-0 ~]# useradd stack
    [root@controller-0 ~]# passwd stack  # specify a password
  2. Disable password requirements for this user when using sudo:

    [root@controller-0 ~]# echo "stack ALL=(root) NOPASSWD:ALL" | tee -a /etc/sudoers.d/stack
    [root@controller-0 ~]# chmod 0440 /etc/sudoers.d/stack
  3. After you create and configure the stack user on all pre-provisioned nodes, copy the stack user’s public SSH key from the director node to each overcloud node. For example, to copy the director’s public SSH key to the Controller node, run the following command:

    [stack@director ~]$ ssh-copy-id stack@192.168.24.2

9.3. Registering the operating system for pre-provisioned nodes

Each node requires access to a Red Hat subscription. Complete the following steps on each node to register your nodes with the Red Hat Content Delivery Network.

Procedure

  1. Run the registration command and enter your Customer Portal user name and password when prompted:

    [root@controller-0 ~]# sudo subscription-manager register
  2. Find the entitlement pool for the Red Hat OpenStack Platform 16:

    [root@controller-0 ~]# sudo subscription-manager list --available --all --matches="Red Hat OpenStack"
  3. Use the pool ID located in the previous step to attach the Red Hat OpenStack Platform 16 entitlements:

    [root@controller-0 ~]# sudo subscription-manager attach --pool=pool_id
  4. Disable all default repositories:

    [root@controller-0 ~]# sudo subscription-manager repos --disable=*
  5. Enable the required Red Hat Enterprise Linux repositories.

    1. For x86_64 systems, run:

      [root@controller-0 ~]# sudo subscription-manager repos --enable=rhel-8-for-x86_64-baseos-eus-rpms --enable=rhel-8-for-x86_64-appstream-eus-rpms --enable=rhel-8-for-x86_64-highavailability-eus-rpms --enable=ansible-2.8-for-rhel-8-x86_64-rpms --enable=openstack-16-for-rhel-8-x86_64-rpms --enable=rhceph-4-osd-for-rhel-8-x86_64-rpms --enable=rhceph-4-mon-for-rhel-8-x86_64-rpms --enable=rhceph-4-tools-for-rhel-8-x86_64-rpms --enable=advanced-virt-for-rhel-8-x86_64-rpms --enable=fast-datapath-for-rhel-8-x86_64-rpms
    2. For POWER systems, run:

      [root@controller-0 ~]# sudo subscription-manager repos --enable=rhel-8-for-ppc64le-baseos-rpms --enable=rhel-8-for-ppc64le-appstream-rpms --enable=rhel-8-for-ppc64le-highavailability-rpms --enable=ansible-2.8-for-rhel-8-ppc64le-rpms --enable=openstack-16-for-rhel-8-ppc64le-rpms --enable=advanced-virt-for-rhel-8-ppc64le-rpms
    Important

    Enable only the repositories listed. Additional repositories can cause package and software conflicts. Do not enable any additional repositories.

  6. Update your system to ensure you have the latest base system packages:

    [root@controller-0 ~]# sudo dnf update -y
    [root@controller-0 ~]# sudo reboot

The node is now ready to use for your overcloud.

9.4. Configuring SSL/TLS access to director

If the director uses SSL/TLS, the pre-provisioned nodes require the certificate authority file used to sign the director’s SSL/TLS certificates. If you use your own certificate authority, perform the following actions on each overcloud node.

Procedure

  1. Copy the certificate authority file to the /etc/pki/ca-trust/source/anchors/ directory on each pre-provisioned node.
  2. Run the following command on each overcloud node:

    [root@controller-0 ~]#  sudo update-ca-trust extract

These steps ensure that the overcloud nodes can access the director’s Public API over SSL/TLS.

9.5. Configuring networking for the control plane

The pre-provisioned overcloud nodes obtain metadata from director using standard HTTP requests. This means all overcloud nodes require L3 access to either:

  • The director Control Plane network, which is the subnet that you define with the network_cidr parameter in your undercloud.conf file. The overcloud nodes require either direct access to this subnet or routable access to the subnet.
  • The director Public API endpoint, that you specify with the undercloud_public_host parameter in your undercloud.conf file. This option is available if you do not have an L3 route to the Control Plane or if you want to use SSL/TLS communication. For more information about configuring your overcloud nodes to use the Public API endpoint, see Section 9.6, “Using a separate network for pre-provisioned nodes”.

Director uses the Control Plane network to manage and configure a standard overcloud. For an overcloud with pre-provisioned nodes, your network configuration might require some modification to accommodate communication between the director and the pre-provisioned nodes.

Using network isolation

You can use network isolation to group services to use specific networks, including the Control Plane. There are multiple network isolation strategies in the the Advanced Overcloud Customization guide. You can also define specific IP addresses for nodes on the Control Plane. For more information about isolating networks and creating predictable node placement strategies, see the following sections in the Advanced Overcloud Customizations guide:

Note

If you use network isolation, ensure that your NIC templates do not include the NIC used for undercloud access. These templates can reconfigure the NIC, which introduces connectivity and configuration problems during deployment.

Assigning IP addresses

If you do not use network isolation, you can use a single Control Plane network to manage all services. This requires manual configuration of the Control Plane NIC on each node to use an IP address within the Control Plane network range. If you are using the director Provisioning network as the Control Plane, ensure that the overcloud IP addresses that you choose are outside of the DHCP ranges for both provisioning (dhcp_start and dhcp_end) and introspection (inspection_iprange).

During standard overcloud creation, director creates OpenStack Networking (neutron) ports and automatically assigns IP addresses to the overcloud nodes on the Provisioning / Control Plane network. However, this can cause director to assign different IP addresses to the ones that you configure manually for each node. In this situation, use a predictable IP address strategy to force director to use the pre-provisioned IP assignments on the Control Plane.

For example, you can use an environment file ctlplane-assignments.yaml with the following IP assignments to implement a predictable IP strategy:

resource_registry:
  OS::TripleO::DeployedServer::ControlPlanePort: /usr/share/openstack-tripleo-heat-templates/deployed-server/deployed-neutron-port.yaml

parameter_defaults:
  DeployedServerPortMap:
    controller-0-ctlplane:
      fixed_ips:
        - ip_address: 192.168.24.2
      subnets:
        - cidr: 192.168.24.0/24
      network:
        tags:
          192.168.24.0/24
    compute-0-ctlplane:
      fixed_ips:
        - ip_address: 192.168.24.3
      subnets:
        - cidr: 192.168.24.0/24
      network:
        tags:
          - 192.168.24.0/24

In this example, the OS::TripleO::DeployedServer::ControlPlanePort resource passes a set of parameters to director and defines the IP assignments of your pre-provisioned nodes. Use the DeployedServerPortMap parameter to define the IP addresses and subnet CIDRs that correspond to each overcloud node. The mapping defines the following attributes:

  1. The name of the assignment, which follows the format <node_hostname>-<network> where the <node_hostname> value matches the short host name for the node, and <network> matches the lowercase name of the network. For example: controller-0-ctlplane for controller-0.example.com and compute-0-ctlplane for compute-0.example.com.
  2. The IP assignments, which use the following parameter patterns:

    • fixed_ips/ip_address - Defines the fixed IP addresses for the control plane. Use multiple ip_address parameters in a list to define multiple IP addresses.
    • subnets/cidr - Defines the CIDR value for the subnet.

A later section in this chapter uses the resulting environment file (ctlplane-assignments.yaml) as part of the openstack overcloud deploy command.

9.6. Using a separate network for pre-provisioned nodes

By default, director uses the Provisioning network as the overcloud Control Plane. However, if this network is isolated and non-routable, nodes cannot communicate with the director Internal API during configuration. In this situation, you might need to define a separate network for the nodes and configure them to communicate with the director over the Public API.

There are several requirements for this scenario:

The examples in this section use IP address assignments that differ from the main scenario:

Table 9.2. Provisioning network IP assignments

Node NameIP address or FQDN

Director (Internal API)

192.168.24.1 (Provisioning Network and Control Plane)

Director (Public API)

10.1.1.1 / director.example.com

Overcloud Virtual IP

192.168.100.1

Controller 0

192.168.100.2

Compute 0

192.168.100.3

The following sections provide additional configuration for situations that require a separate network for overcloud nodes.

IP address assignments

The method for IP assignments is similar to Section 9.5, “Configuring networking for the control plane”. However, since the Control Plane is not routable from the deployed servers, you must use the DeployedServerPortMap parameter to assign IP addresses from your chosen overcloud node subnet, including the virtual IP address to access the Control Plane. The following example is a modified version of the ctlplane-assignments.yaml environment file from Section 9.5, “Configuring networking for the control plane” that accommodates this network architecture:

resource_registry:
  OS::TripleO::DeployedServer::ControlPlanePort: /usr/share/openstack-tripleo-heat-templates/deployed-server/deployed-neutron-port.yaml
  OS::TripleO::Network::Ports::ControlPlaneVipPort: /usr/share/openstack-tripleo-heat-templates/deployed-server/deployed-neutron-port.yaml
  OS::TripleO::Network::Ports::RedisVipPort: /usr/share/openstack-tripleo-heat-templates/network/ports/noop.yaml 1

parameter_defaults:
  NeutronPublicInterface: eth1
  EC2MetadataIp: 192.168.100.1 2
  ControlPlaneDefaultRoute: 192.168.100.1
  DeployedServerPortMap:
    control_virtual_ip:
      fixed_ips:
        - ip_address: 192.168.100.1
      subnets:
        - cidr: 24
    controller-0-ctlplane:
      fixed_ips:
        - ip_address: 192.168.100.2
      subnets:
        - cidr: 24
    compute-0-ctlplane:
      fixed_ips:
        - ip_address: 192.168.100.3
      subnets:
        - cidr: 24
1
The RedisVipPort resource is mapped to network/ports/noop.yaml. This mapping is necessary because the default Redis VIP address comes from the Control Plane. In this situation, use a noop to disable this Control Plane mapping.
2
The EC2MetadataIp and ControlPlaneDefaultRoute parameters are set to the value of the Control Plane virtual IP address. The default NIC configuration templates require these parameters and you must set them to use a pingable IP address to pass the validations performed during deployment. Alternatively, customize the NIC configuration so that they do not require these parameters.

9.7. Mapping pre-provisioned node hostnames

When you configure pre-provisioned nodes, you must map heat-based hostnames to their actual hostnames so that ansible-playbook can reach a resolvable host. Use the HostnameMap to map these values.

Procedure

  1. Create an environment file, for example hostname-map.yaml, and include the HostnameMap parameter and the hostname mappings. Use the following syntax:

    parameter_defaults:
      HostnameMap:
        [HEAT HOSTNAME]: [ACTUAL HOSTNAME]
        [HEAT HOSTNAME]: [ACTUAL HOSTNAME]

    The [HEAT HOSTNAME] usually conforms to the following convention: [STACK NAME]-[ROLE]-[INDEX]:

    parameter_defaults:
      HostnameMap:
        overcloud-controller-0: controller-00-rack01
        overcloud-controller-1: controller-01-rack02
        overcloud-controller-2: controller-02-rack03
        overcloud-novacompute-0: compute-00-rack01
        overcloud-novacompute-1: compute-01-rack01
        overcloud-novacompute-2: compute-02-rack01
  2. Save the hostname-map.yaml file.

9.8. Configuring Ceph Storage for pre-provisioned nodes

Complete the following steps on the undercloud host to configure ceph-ansible for nodes that are already deployed.

Procedure

  1. On the undercloud host, create an environment variable, OVERCLOUD_HOSTS, and set the variable to a space-separated list of IP addresses of the overcloud hosts that you want to use as Ceph clients:

    export OVERCLOUD_HOSTS="192.168.1.8 192.168.1.42"
  2. Run the enable-ssh-admin.sh script to configure a user on the overcloud nodes that Ansible can use to configure Ceph clients:

    bash /usr/share/openstack-tripleo-heat-templates/deployed-server/scripts/enable-ssh-admin.sh

When you run the openstack overcloud deploy command, Ansible configures the hosts that you define in the OVERCLOUD_HOSTS variable as Ceph clients.

9.9. Creating the overcloud with pre-provisioned nodes

The overcloud deployment uses the standard CLI methods from Section 7.13, “Deployment command”. For pre-provisioned nodes, the deployment command requires some additional options and environment files from the core heat template collection:

  • --disable-validations - Use this option to disable basic CLI validations for services not used with pre-provisioned infrastructure. If you do not disable these validations, the deployment fails.
  • environments/deployed-server-environment.yaml - Include this environment file to create and configure the pre-provisioned infrastructure. This environment file substitutes the OS::Nova::Server resources with OS::Heat::DeployedServer resources.

The following command is an example overcloud deployment command with the environment files specific to the pre-provisioned architecture:

$ source ~/stackrc
(undercloud) $ openstack overcloud deploy \
  [other arguments] \
  --disable-validations \
  -e /usr/share/openstack-tripleo-heat-templates/environments/deployed-server-environment.yaml \
  -e /home/stack/templates/hostname-map.yaml \
  --overcloud-ssh-user stack \
  --overcloud-ssh-key ~/.ssh/id_rsa \
  [OTHER OPTIONS]

The --overcloud-ssh-user and --overcloud-ssh-key options are used to SSH into each overcloud node during the configuration stage, create an initial tripleo-admin user, and inject an SSH key into /home/tripleo-admin/.ssh/authorized_keys. To inject the SSH key, specify the credentials for the initial SSH connection with --overcloud-ssh-user and --overcloud-ssh-key (defaults to ~/.ssh/id_rsa). To limit exposure to the private key that you specify with the --overcloud-ssh-key option, director never passes this key to any API service, such as heat or the Workflow service (mistral), and only the director openstack overcloud deploy command uses this key to enable access for the tripleo-admin user.

9.10. Overcloud deployment output

When the overcloud creation completes, director provides a recap of the Ansible plays that were executed to configure the overcloud:

PLAY RECAP *************************************************************
overcloud-compute-0     : ok=160  changed=67   unreachable=0    failed=0
overcloud-controller-0  : ok=210  changed=93   unreachable=0    failed=0
undercloud              : ok=10   changed=7    unreachable=0    failed=0

Tuesday 15 October 2018  18:30:57 +1000 (0:00:00.107) 1:06:37.514 ******
========================================================================

Director also provides details to access your overcloud.

Ansible passed.
Overcloud configuration completed.
Overcloud Endpoint: http://192.168.24.113:5000
Overcloud Horizon Dashboard URL: http://192.168.24.113:80/dashboard
Overcloud rc file: /home/stack/overcloudrc
Overcloud Deployed

9.11. Accessing the overcloud

The director generates a script to configure and help authenticate interactions with your overcloud from the undercloud. The director saves this file, overcloudrc, in the home directory of the stack user. Run the following command to use this file:

(undercloud) $ source ~/overcloudrc

This command loads the environment variables that are necessary to interact with your overcloud from the undercloud CLI. The command prompt changes to indicate this:

(overcloud) $

To return to interacting with the undercloud, run the following command:

(overcloud) $ source ~/stackrc
(undercloud) $

Each node in the overcloud also contains a heat-admin user. The stack user has SSH access to this user on each node. To access a node over SSH, find the IP address of the node that you want to access:

(undercloud) $ openstack server list

Then connect to the node using the heat-admin user and the IP address of the node:

(undercloud) $ ssh heat-admin@192.168.24.23

9.12. Scaling pre-provisioned nodes

The process for scaling pre-provisioned nodes is similar to the standard scaling procedures in Chapter 15, Scaling overcloud nodes. However, the process to add new pre-provisioned nodes differs because pre-provisioned nodes do not use the standard registration and management process from OpenStack Bare Metal (ironic) and OpenStack Compute (nova).

Scaling up pre-provisioned nodes

When scaling up the overcloud with pre-provisioned nodes, you must configure the orchestration agent on each node to correspond to the director node count.

Perform the following actions to scale up overcloud nodes:

  1. Prepare the new pre-provisioned nodes according to Section 9.1, “Pre-provisioned node requirements”.
  2. Scale up the nodes. For more information, see Chapter 15, Scaling overcloud nodes.
  3. After you execute the deployment command, wait until the director creates the new node resources and launches the configuration.

Scaling down pre-provisioned nodes

When scaling down the overcloud with pre-provisioned nodes, follow the scale down instructions in Chapter 15, Scaling overcloud nodes.

In most scaling operations, you must obtain the UUID value of the node that you want to remove and pass this value to the openstack overcloud node delete command. To obtain this UUID, list the resources for the specific role:

$ openstack stack resource list overcloud -c physical_resource_id -c stack_name -n5 --filter type=OS::TripleO::<RoleName>Server

Replace <RoleName> with the name of the role that you want to scale down. For example, for the ComputeDeployedServer role, run the following command:

$ openstack stack resource list overcloud -c physical_resource_id -c stack_name -n5 --filter type=OS::TripleO::ComputeDeployedServerServer

Use the stack_name column in the command output to identify the UUID associated with each node. The stack_name includes the integer value of the index of the node in the heat resource group:

+------------------------------------+----------------------------------+
| physical_resource_id               | stack_name                       |
+------------------------------------+----------------------------------+
| 294d4e4d-66a6-4e4e-9a8b-           | overcloud-ComputeDeployedServer- |
| 03ec80beda41                       | no7yfgnh3z7e-1-ytfqdeclwvcg      |
| d8de016d-                          | overcloud-ComputeDeployedServer- |
| 8ff9-4f29-bc63-21884619abe5        | no7yfgnh3z7e-0-p4vb3meacxwn      |
| 8c59f7b1-2675-42a9-ae2c-           | overcloud-ComputeDeployedServer- |
| 2de4a066f2a9                       | no7yfgnh3z7e-2-mmmaayxqnf3o      |
+------------------------------------+----------------------------------+

The indices 0, 1, or 2 in the stack_name column correspond to the node order in the heat resource group. Pass the corresponding UUID value from the physical_resource_id column to openstack overcloud node delete command.

After you remove overcloud nodes from the stack, power off these nodes. In a standard deployment, the bare metal services on the director control this function. However, with pre-provisioned nodes, you must either manually shut down these nodes or use the power management control for each physical system. If you do not power off the nodes after removing them from the stack, they might remain operational and reconnect as part of the overcloud environment.

After you power off the removed nodes, reprovision them to a base operating system configuration so that they do not unintentionally join the overcloud in the future

Note

Do not attempt to reuse nodes previously removed from the overcloud without first reprovisioning them with a fresh base operating system. The scale down process only removes the node from the overcloud stack and does not uninstall any packages.

9.13. Removing a pre-provisioned overcloud

To remove an entire overcloud that uses pre-provisioned nodes, see Section 12.5, “Removing the overcloud” for the standard overcloud remove procedure. After you remove the overcloud, power off all nodes and reprovision them to a base operating system configuration.

Note

Do not attempt to reuse nodes previously removed from the overcloud without first reprovisioning them with a fresh base operating system. The removal process only deletes the overcloud stack and does not uninstall any packages.

9.14. Next steps

This concludes the creation of the overcloud using pre-provisioned nodes. For post-creation functions, see Chapter 11, Performing overcloud post-installation tasks.