The group variables originally in ansible/group_vars/ were playbook group variables, due to being adjacent to the playbooks. Typically they provided default values for global variables in the all group, as well as some more specific groups. This has worked fairly well, but results in (at least) a couple of problems. 1. The default variable precedence rules mean that these playbook group variables have a higher precedence than inventory group variables (for a given group). This can make it challenging to override playbook group variables in the inventory in Kayobe configuration. 2. Any playbook run by Kayobe must be in the same directory as the playbook group variables in order to use them. Given that they include variables required for connectivity such as ansible_host and ansible_user, this is quite critical. For Kayobe custom playbooks, we work around this by symlinking to the group_vars directory from the directory containing the custom playbook. This is not an elegant workaround, and has assumptions about the relative paths of the Kayobe configuration and virtual environment in which Kayobe is installed. Story: 2010280 Task: 46233 Change-Id: Ifea5c7e73f6f410f96a7398bfd349d1f631d9fc0
9.9 KiB
Control Plane Service Placement
Note
This is an advanced topic and should only be attempted when familiar with kayobe and OpenStack.
The default configuration in kayobe places all control plane services on a single set of servers described as 'controllers'. In some cases it may be necessary to introduce more than one server role into the control plane, and control which services are placed onto the different server roles.
Configuration
Overcloud Inventory Discovery
If using a seed host to enable discovery of the control plane
services, it is necessary to configure how the discovered hosts map into
kayobe groups. This is done using the
overcloud_group_hosts_map
variable, which maps names of
kayobe groups to a list of the hosts to be added to that group.
This variable will be used during the command
kayobe overcloud inventory discover
. An inventory file will
be generated in ${KAYOBE_CONFIG_PATH}/inventory/overcloud
with discovered hosts added to appropriate kayobe groups based on
overcloud_group_hosts_map
.
Kolla-ansible Inventory Mapping
Once hosts have been discovered and enrolled into the kayobe
inventory, they must be added to the kolla-ansible inventory. This is
done by mapping from top level kayobe groups to top level kolla-ansible
groups using the
kolla_overcloud_inventory_top_level_group_map
variable.
This variable maps from kolla-ansible groups to lists of kayobe groups,
and variables to define for those groups in the kolla-ansible
inventory.
Variables For Custom Server Roles
Certain variables must be defined for hosts in the
overcloud
group. For hosts in the controllers
group, many variables are mapped to other variables with a
controller_
prefix in files under
ansible/inventory/group_vars/controllers/
. This is done in
order that they may be set in a global extra variables file, typically
controllers.yml
, with defaults set in
ansible/inventory/group_vars/all/controllers
. A similar
scheme is used for hosts in the monitoring
group.
Variable | Purpose |
---|---|
ansible_user |
Username with which to access the host via SSH. |
|
Username with which to access the host before
|
|
List of LVM volume groups to configure. See mrlesmithjr.manage-lvm role for format. |
|
List of software RAID arrays. See mrlesmithjr.mdadm role for format. |
|
List of names of networks to which the host is connected. |
sysctl_parameters |
Dict of sysctl parameters to set. |
|
List of users to create. See singleplatform-eng.users role |
If configuring BIOS and RAID via
kayobe overcloud bios raid configure
, the following
variables should also be defined:
Variable | Purpose |
---|---|
|
Dict mapping BIOS configuration options to their required values. See stackhpc.drac role for format. |
|
List of RAID virtual disks to configure. See stackhpc.drac role for format. |
These variables can be defined in inventory host or group variables
files, under
${KAYOBE_CONFIG_PATH}/inventory/host_vars/<host>
or
${KAYOBE_CONFIG_PATH}/inventory/group_vars/<group>
respectively.
Custom Kolla-ansible Inventories
As an advanced option, it is possible to fully customise the content of the kolla-ansible inventory, at various levels. To facilitate this, kayobe breaks the kolla-ansible inventory into three separate sections.
Top level groups define the roles of hosts, e.g.
controller
or compute
, and it is to these
groups that hosts are mapped directly.
Components define groups of services, e.g.
nova
or ironic
, which are mapped to top level
groups.
Services define single containers, e.g.
nova-compute
or ironic-api
, which are mapped
to components.
The default top level inventory is generated from
kolla_overcloud_inventory_top_level_group_map
. Kayobe's
component- and service-level inventory for kolla-ansible is static, and
taken from the kolla-ansible example multinode
inventory.
The complete inventory is generated by concatenating these
inventories.
Each level may be separately overridden by setting the following variables:
Variable | Purpose |
---|---|
|
Overcloud inventory containing a mapping from top level groups to hosts. |
|
Overcloud inventory containing a mapping from components to top level groups. |
|
Overcloud inventory containing a mapping from services to components. |
|
Full overcloud inventory contents. |
Examples
Example 1: Adding Network Hosts
This example walks through the configuration that could be applied to
enable the use of separate hosts for neutron network services and load
balancing. The control plane consists of three controllers,
controller-[0-2]
, and two network hosts,
network-[0-1]
. All file paths are relative to
${KAYOBE_CONFIG_PATH}
.
First, we must make the network group separate from controllers:
[controllers]
# Empty group to provide declaration of controllers group.
[network]
# Empty group to provide declaration of network group.
Then, we must map the hosts to kayobe groups.
overcloud_group_hosts_map:
controllers:
- controller-0
- controller-1
- controller-2
network:
- network-0
- network-1
Next, we must map these groups to kolla-ansible groups.
kolla_overcloud_inventory_top_level_group_map:
control:
groups:
- controllers
network:
groups:
- network
Finally, we create a group variables file for hosts in the network group, providing the necessary variables for a control plane host.
ansible_user: "{{ kayobe_ansible_user }}"
bootstrap_user: "{{ controller_bootstrap_user }}"
lvm_groups: "{{ controller_lvm_groups }}"
mdadm_arrays: "{{ controller_mdadm_arrays }}"
network_interfaces: "{{ controller_network_host_network_interfaces }}"
sysctl_parameters: "{{ controller_sysctl_parameters }}"
users: "{{ controller_users }}"
Here we are using the controller-specific values for some of these variables, but they could equally be different.
Example 2: Overriding the Kolla-ansible Inventory
This example shows how to override one or more sections of the
kolla-ansible inventory. All file paths are relative to
${KAYOBE_CONFIG_PATH}
.
It is typically best to start with an inventory template taken from
the Kayobe source code, and then customize it. The templates can be
found in ansible/roles/kolla-ansible/templates
, e.g.
components template is overcloud-components.j2
.
First, create a file containing the customised inventory section. We'll use the components section in this example.
[nova]
control
[ironic]
{% if kolla_enable_ironic | bool %}
control
{% endif %}
...
Next, we must configure kayobe to use this inventory template.
kolla_overcloud_inventory_custom_components: "{{ lookup('template', kayobe_env_config_path ~ '/kolla/inventory/overcloud-components.j2') }}"
Here we use the template
lookup plugin to render the
Jinja2-formatted inventory template.