kayobe/doc/source/configuration/hosts.rst

.. _configuration-hosts:

==================
Host Configuration
==================

This section covers configuration of hosts. It does not cover configuration or
deployment of containers. Hosts that are configured by Kayobe include:

* Seed hypervisor (``kayobe seed hypervisor host configure``)
* Seed (``kayobe seed host configure``)
* Overcloud (``kayobe overcloud host configure``)

Unless otherwise stated, all host configuration described here is applied to
each of these types of host.

.. seealso:: Ansible tags for limiting the scope of Kayobe commands are
             included under the relevant sections of this page
             (for more information see :ref:`usage-tags`).

Configuration Location
======================

Some host configuration options are set via global variables, and others have a
variable for each type of host. The latter variables are included in the
following files under ``${KAYOBE_CONFIG_PATH}``:

* ``seed-hypervisor.yml``
* ``seed.yml``
* ``compute.yml``
* ``controller.yml``
* ``monitoring.yml``
* ``storage.yml``

Note that any variable may be set on a per-host or per-group basis, by using
inventory host or group variables - these delineations are for convenience.

Paths
=====

Several directories are used by Kayobe on the remote hosts. There is a
hierarchy of variables in ``${KAYOBE_CONFIG_PATH}/globals.yml`` that can be
used to control where these are located.

* ``base_path`` (default ``/opt/kayobe/``) sets the default base path for
  various directories.
* ``config_path`` (default ``{{ base_path }}/etc``) is a path in which to store
  configuration files.
* ``image_cache_path`` (default ``{{ base_path }}/images``) is a path in which
  to cache downloaded or built images.
* ``source_checkout_path`` (default ``{{ base_path }}/src``) is a path into
  which to store clones of source code repositories.
* ``virtualenv_path`` (default ``{{ base_path }}/venvs``) is a path in which to
  create Python virtual environments.

SSH Known Hosts
===============
*tags:*
  | ``ssh-known-host``

While strictly this configuration is applied to the Ansible control host
(``localhost``), it is applied during the ``host configure`` commands.
The ``ansible_host`` of each host is added as an SSH known host. This is
typically the host's IP address on the admin network (``admin_oc_net_name``),
as defined in ``${KAYOBE_CONFIG_PATH}/network-allocation.yml`` (see
:ref:`configuration-network-ip-allocation`).

Kayobe User Bootstrapping
=========================
*tags:*
  | ``kayobe-ansible-user``

Kayobe uses a user account defined by the ``kayobe_ansible_user`` variable (in
``${KAYOBE_CONFIG_PATH}/globals.yml``) for remote SSH access. By default, this
is ``stack``.

Typically, the image used to provision these hosts will not include this user
account, so Kayobe performs a bootstrapping step to create it, as a different
user. In cloud images, there is often a user named after the OS distro, e.g.
``centos`` or ``ubuntu``. This user defaults to the name of the user running
Kayobe, but may be set via the following variables:

* ``seed_hypervisor_bootstrap_user``
* ``seed_bootstrap_user``
* ``compute_bootstrap_user``
* ``controller_bootstrap_user``
* ``monitoring_bootstrap_user``
* ``storage_bootstrap_user``

For example, to set the bootstrap user for controllers to ``centos``:

.. code-block:: yaml
   :caption: ``controllers.yml``

   controller_bootstrap_user: centos

PyPI Mirror and proxy
=====================
*tags:*
  | ``pip``

Kayobe supports configuration of a PyPI mirror and/or proxy, via variables in
``${KAYOBE_CONFIG_PATH}/pip.yml``.
Mirror functionality is enabled by setting the ``pip_local_mirror`` variable to
``true`` and proxy functionality is enabled by setting ``pip_proxy`` variable
to a proxy URL.

Kayobe will generate configuration for:

* ``pip`` to use the mirror and proxy
* ``easy_install`` to use the mirror

for the list of users defined by ``pip_applicable_users`` (default
``kayobe_ansible_user`` and ``root``), in addition to the user used for Kolla
Ansible (``kolla_ansible_user``). The mirror URL is configured via
``pip_index_url``, and ``pip_trusted_hosts`` is a list of 'trusted' hosts, for
which SSL verification will be disabled.

For example, to configure use of the test PyPI mirror at
https://test.pypi.org/simple/:

.. code-block:: yaml
   :caption: ``pip.yml``

   pip_local_mirror: true
   pip_index_url: https://test.pypi.org/simple/

To configure use of the PyPI proxy:

.. code-block:: yaml
   :caption: ``pip.yml``

   pip_proxy: http://your_proxy_server:3128


Kayobe Remote Virtual Environment
=================================
*tags:*
  | ``kayobe-target-venv``

By default, Ansible executes modules remotely using the system python
interpreter, even if the Ansible control process is executed from within a
virtual environment (unless the ``local`` connection plugin is used).
This is not ideal if there are python dependencies that must be installed
with isolation from the system python packages. Ansible can be configured to
use a virtualenv by setting the host variable ``ansible_python_interpreter``
to a path to a python interpreter in an existing virtual environment.

If kayobe detects that ``ansible_python_interpreter`` is set and references a
virtual environment, it will create the virtual environment if it does not
exist. Typically this variable should be set via a group variable in the
inventory for hosts in the ``seed``, ``seed-hypervisor``, and/or ``overcloud``
groups.

The default Kayobe configuration in the ``kayobe-config`` repository sets
``ansible_python_interpreter`` to ``{{ virtualenv_path }}/kayobe/bin/python``
for the ``seed``, ``seed-hypervisor``, and ``overcloud`` groups.

Disk Wiping
===========
*tags:*
  | ``wipe-disks``

Using hosts that may have stale data on their disks could affect the deployment
of the cloud. This is not a configuration option, since it should only be
performed once to avoid losing useful data. It is triggered by passing the
``--wipe-disks`` argument to the ``host configure`` commands.

Users and Groups
================
*tags:*
  | ``users``

Linux user accounts and groups can be configured using the ``users_default``
variable in ``${KAYOBE_CONFIG_PATH}/users.yml``. The format of the list is
that used by the ``users`` variable of the `singleplatform-eng.users
<https://galaxy.ansible.com/singleplatform-eng/users>`__ role.  The following
variables can be used to set the users for specific types of hosts:

* ``seed_hypervisor_users``
* ``seed_users``
* ``compute_users``
* ``controller_users``
* ``monitoring_users``
* ``storage_users``

In the following example, a single user named ``bob`` is created. A password
hash has been generated via ``mkpasswd --method=sha-512``. The user is added to
the ``wheel`` group, and an SSH key is authorised. The SSH public key should be
added to the Kayobe configuration.

.. code-block:: yaml
   :caption: ``users.yml``

   users_default:
    - username: bob
      name: Bob
      password: "$6$wJt9MLWrHlWN8$oXJHbdaslm9guD5EC3Dry1mphuqF9NPeQ43OXk3cXZa2ze/F9FOTxm2KvvDkbdxBDs7ouwdiLTUJ1Ff40.cFU."
      groups:
        - wheel
      append: True
      ssh_key:
        - "{{ lookup('file', kayobe_config_path ~ '/ssh-keys/id_rsa_bob.pub') }}"

Package Repositories (CentOS 7)
===============================
*tags:*
  | ``yum``

Kayobe supports configuration of package repositories via Yum, via variables in
``${KAYOBE_CONFIG_PATH}/yum.yml``.

Configuration of yum.conf
-------------------------

Global configuration of Yum is stored in ``/etc/yum.conf``, and options can be
set via the ``yum_config`` variable. Options are added to the ``[main]``
section of the file. For example, to configure Yum to use a proxy server:

.. code-block:: yaml
   :caption: ``yum.yml``

   yum_config:
     proxy: https://proxy.example.com

CentOS and EPEL Mirrors
-----------------------

CentOS and EPEL mirrors can be enabled by setting ``yum_use_local_mirror`` to
``true``.  CentOS repository mirrors are configured via the following
variables:

* ``yum_centos_mirror_host`` (default ``mirror.centos.org``) is the mirror
  hostname.
* ``yum_centos_mirror_directory`` (default ``centos``) is a directory on the
  mirror in which repositories may be accessed.

EPEL repository mirrors are configured via the following variables:

* ``yum_epel_mirror_host`` (default ``download.fedoraproject.org``) is the
  mirror hostname.
* ``yum_epel_mirror_directory`` (default ``pub/epel``) is a directory on the
  mirror in which repositories may be accessed.

For example, to configure CentOS and EPEL mirrors at mirror.example.com:

.. code-block:: yaml
   :caption: ``yum.yml``

   yum_use_local_mirror: true
   yum_centos_mirror_host: mirror.example.com
   yum_epel_mirror_host: mirror.example.com

Custom Yum Repositories
-----------------------

It is also possible to configure a list of custom Yum repositories via the
``yum_custom_repos`` variable. The format is a dict/map, with repository names
mapping to a dict/map of arguments to pass to the Ansible ``yum_repository``
module.

For example, the following configuration defines a single Yum repository called
``widgets``.

.. code-block:: yaml
   :caption: ``yum.yml``

   yum_custom_repos:
     widgets:
       baseurl: http://example.com/repo
       file: widgets
       gpgkey: http://example.com/gpgkey
       gpgcheck: yes

Disabling EPEL
--------------

It is possible to disable the EPEL Yum repository by setting
``yum_install_epel`` to ``false``.

Package Repositories (CentOS 8)
===============================
*tags:*
  | ``dnf``

Kayobe supports configuration of package repositories via DNF, via variables in
``${KAYOBE_CONFIG_PATH}/dnf.yml``. For backwards compatibility, all variables
in this section starting with ``dnf_`` default to the equivalently named Yum
variable starting with ``yum_``.

Configuration of dnf.conf
-------------------------

Global configuration of DNF is stored in ``/etc/dnf/dnf.conf``, and options can
be set via the ``dnf_config`` variable. Options are added to the ``[main]``
section of the file. For example, to configure DNF to use a proxy server:

.. code-block:: yaml
   :caption: ``dnf.yml``

   dnf_config:
     proxy: https://proxy.example.com

CentOS and EPEL Mirrors
-----------------------

CentOS and EPEL mirrors can be enabled by setting ``dnf_use_local_mirror`` to
``true``.  CentOS repository mirrors are configured via the following
variables:

* ``dnf_centos_mirror_host`` (default ``mirror.centos.org``) is the mirror
  hostname.
* ``dnf_centos_mirror_directory`` (default ``centos``) is a directory on the
  mirror in which repositories may be accessed.

EPEL repository mirrors are configured via the following variables:

* ``dnf_epel_mirror_host`` (default ``download.fedoraproject.org``) is the
  mirror hostname.
* ``dnf_epel_mirror_directory`` (default ``pub/epel``) is a directory on the
  mirror in which repositories may be accessed.

For example, to configure CentOS and EPEL mirrors at mirror.example.com:

.. code-block:: yaml
   :caption: ``dnf.yml``

   dnf_use_local_mirror: true
   dnf_centos_mirror_host: mirror.example.com
   dnf_epel_mirror_host: mirror.example.com

Custom DNF Repositories
-----------------------

It is also possible to configure a list of custom DNF repositories via the
``dnf_custom_repos`` variable. The format is a dict/map, with repository names
mapping to a dict/map of arguments to pass to the Ansible ``yum_repository``
module.

For example, the following configuration defines a single DNF repository called
``widgets``.

.. code-block:: yaml
   :caption: ``dnf.yml``

   dnf_custom_repos:
     widgets:
       baseurl: http://example.com/repo
       file: widgets
       gpgkey: http://example.com/gpgkey
       gpgcheck: yes

Disabling EPEL
--------------

It is possible to disable the EPEL DNF repository by setting
``dnf_install_epel`` to ``false``.

DNF Automatic
-------------

DNF Automatic provides a mechanism for applying regular updates of packages.
DNF Automatic is disabled by default, and may be enabled by setting
``dnf_automatic_enabled`` to ``true``.

.. code-block:: yaml
   :caption: ``dnf.yml``

   dnf_automatic_enabled:  true

By default, only security updates are applied. Updates for all packages may be
installed by setting ``dnf_automatic_upgrade_type`` to ``default``. This may
cause the system to be less predictable as packages are updated without
oversight or testing.

SELinux
=======
*tags:*
  | ``disable-selinux``

SELinux is not supported by Kolla Ansible currently, so it is disabled by
Kayobe. If necessary, Kayobe will reboot systems in order to apply a change to
the SELinux configuration. The timeout for waiting for systems to reboot is
``disable_selinux_reboot_timeout``. Alternatively, the reboot may be avoided by
setting ``disable_selinux_do_reboot`` to ``false``.

Network Configuration
=====================
*tags:*
  | ``network``

Configuration of host networking is covered in depth in
:ref:`configuration-network`.

Sysctls
=======
*tags:*
  | ``sysctl``

Arbitrary ``sysctl`` configuration can be applied to hosts. The variable format
is a dict/map, mapping parameter names to their required values. The following
variables can be used to set ``sysctl`` configuration specific types of hosts:

* ``seed_hypervisor_sysctl_parameters``
* ``seed_sysctl_parameters``
* ``compute_sysctl_parameters``
* ``controller_sysctl_parameters``
* ``monitoring_sysctl_parameters``
* ``storage_sysctl_parameters``

For example, to set the ``net.ipv4.ip_forward`` parameter to ``1`` on controllers:

.. code-block:: yaml
   :caption: ``controllers.yml``

   controller_sysctl_parameters:
     net.ipv4.ip_forward: 1

Disable cloud-init
==================
*tags:*
  | ``disable-cloud-init``

cloud-init is a popular service for performing system bootstrapping. If you are
not using cloud-init, this section can be skipped.

If using the seed's Bifrost service to provision the control plane hosts, the
use of cloud-init may be configured via the ``kolla_bifrost_dib_init_element``
variable.

cloud-init searches for network configuration in order of increasing
precedence; each item overriding the previous.  In some cases, on subsequent
boots cloud-init can automatically reconfigure network interfaces and cause
some issues in network configuration. To disable cloud-init from running after
the initial server bootstrapping, set ``disable_cloud_init`` to ``true`` in
``${KAYOBE_CONFIG_PATH}/overcloud.yml``.

Disable Glean
=============
*tags:*
  | ``disable-glean``

The ``glean`` service can be used to perform system bootstrapping, serving a
similar role to ``cloud-init``. If you are not using ``glean``, this section
can be skipped.

If using the seed's Bifrost service to provision the control plane hosts, the
use of ``glean`` may be configured via the ``kolla_bifrost_dib_init_element``
variable.

After the initial server bootstrapping, the glean service can cause problems as
it attempts to enable all network interfaces, which can lead to timeouts while
booting. To avoid this, the ``glean`` service is disabled. Additionally, any
network interface configuration files generated by ``glean`` and not
overwritten by Kayobe are removed.

Timezone
========
*tags:*
  | ``ntp``

The timezone can be configured via the ``timezone`` variable in
``${KAYOBE_CONFIG_PATH}/ntp.yml``. The value must be a valid Linux
timezone. For example:

.. code-block:: yaml
   :caption: ``ntp.yml``

   timezone: Europe/London

NTP
===
*tags:*
  | ``ntp``

.. note::

   CentOS 8 does not support configuring an NTP daemon. Use :ref:`chrony
   <configuration-hosts-chrony>` instead.

Network Time Protocol (NTP) may be configured via variables in
``${KAYOBE_CONFIG_PATH}/ntp.yml``. The list of NTP servers is
configured via ``ntp_config_server``, and by default the ``pool.ntp.org``
servers are used. A list of restrictions may be added via
``ntp_config_restrict``, and a list of interfaces to listen on via
``ntp_config_listen``. Other options and their default values may be found in
the `resmo.ntp <https://galaxy.ansible.com/resmo/ntp>`__ Ansible role.

.. code-block:: yaml
   :caption: ``ntp.yml``

   ntp_config_server:
     - 1.ubuntu.pool.ntp.org
     - 2.ubuntu.pool.ntp.org

   ntp_config_restrict:
     - '-4 default kod notrap nomodify nopeer noquery'

   ntp_config_listen:
     - eth0

The NTP service may be disabled as follows:

.. code-block:: yaml
   :caption: ``ntp.yml``

   ntp_service_enabled: false

.. _configuration-hosts-chrony:

Chrony
------

Kolla Ansible can deploy a chrony container. This is disabled by default in
Kayobe on CentOS 7 to avoid conflicting with the NTP daemon on the host. On
CentOS 8 Chrony is enabled by default.

To use the containerised chrony daemon and disable the host NTP daemon on
CentOS 7, set the following in ``${KAYOBE_CONFIG_PATH}/kolla.yml``:

.. code-block:: yaml

   kolla_enable_chrony: true

.. _configuration-hosts-mdadm:

Software RAID
=============
*tags:*
  | ``mdadm``

While it is possible to use RAID directly with LVM, some operators may prefer
the userspace tools provided by ``mdadm`` or may have existing software RAID
arrays they want to manage with Kayobe.

Software RAID arrays may be configured via the ``mdadm_arrays`` variable. For
convenience, this is mapped to the following variables:

* ``seed_hypervisor_mdadm_arrays``
* ``seed_mdadm_arrays``
* ``compute_mdadm_arrays``
* ``controller_mdadm_arrays``
* ``monitoring_mdadm_arrays``
* ``storage_mdadm_arrays``

The format of these variables is as defined by the ``mdadm_arrays`` variable of
the `mrlesmithjr.mdadm <https://galaxy.ansible.com/mrlesmithjr/mdadm>`__
Ansible role.

For example, to configure two of the seed's disks as a RAID1 ``mdadm`` array
available as ``/dev/md0``:

.. code-block:: yaml
   :caption: ``seed.yml``

   seed_mdadm_arrays:
     - name: md0
       devices:
         - /dev/sdb
         - /dev/sdc
       level: '1'
       state: present

.. _configuration-hosts-lvm:

LVM
===
*tags:*
  | ``lvm``

Logical Volume Manager (LVM) physical volumes, volume groups, and logical
volumes may be configured via the ``lvm_groups`` variable. For convenience,
this is mapped to the following variables:

* ``seed_hypervisor_lvm_groups``
* ``seed_lvm_groups``
* ``compute_lvm_groups``
* ``controller_lvm_groups``
* ``monitoring_lvm_groups``
* ``storage_lvm_groups``

The format of these variables is as defined by the ``lvm_groups`` variable of
the `mrlesmithjr.manage-lvm
<https://galaxy.ansible.com/mrlesmithjr/manage-lvm>`__ Ansible role.

LVM for libvirt
---------------

LVM is not configured by default on the seed hypervisor. It is possible to
configure LVM to provide storage for a ``libvirt`` storage pool, typically
mounted at ``/var/lib/libvirt/images``.

To use this configuration, set the ``seed_hypervisor_lvm_groups`` variable to
``"{{ seed_hypervisor_lvm_groups_with_data }}"`` and provide a list of disks
via the ``seed_hypervisor_lvm_group_data_disks`` variable.

LVM for Docker
--------------

The default LVM configuration is optimised for the ``devicemapper`` Docker
storage driver, which requires a thin provisioned LVM volume. A second logical
volume is used for storing Docker volume data, mounted at
``/var/lib/docker/volumes``. Both logical volumes are created from a single
``data`` volume group.

To use this configuration, a list of disks must be configured via the following
variables:

* ``seed_lvm_group_data_disks``
* ``compute_lvm_group_data_disks``
* ``controller_lvm_group_data_disks``
* ``monitoring_lvm_group_data_disks``
* ``storage_lvm_group_data_disks``

For example, to configure two of the seed's disks for use by LVM:

.. code-block:: yaml
   :caption: ``seed.yml``

   seed_lvm_group_data_disks:
     - /dev/sdb
     - /dev/sdc

The Docker volumes LVM volume is assigned a size given by the following
variables, with a default value of 75% (of the volume group's capacity):

* ``seed_lvm_group_data_lv_docker_volumes_size``
* ``compute_lvm_group_data_lv_docker_volumes_size``
* ``controller_lvm_group_data_lv_docker_volumes_size``
* ``monitoring_lvm_group_data_lv_docker_volumes_size``
* ``storage_lvm_group_data_lv_docker_volumes_size``

If using a Docker storage driver other than ``devicemapper``, the remaining 25%
of the volume group can be used for Docker volume data. In this case, the LVM
volume's size can be increased to 100%:

.. code-block:: yaml
   :caption: ``controllers.yml``

   controller_lvm_group_data_lv_docker_volumes_size: 100%

If using a Docker storage driver other than ``devicemapper``, it is possible to
avoid using LVM entirely, thus avoiding the requirement for multiple disks. In
this case, set the appropriate ``<host>_lvm_groups`` variable to an empty list:

.. code-block:: yaml
   :caption: ``storage.yml``

   storage_lvm_groups: []

Custom LVM
----------

To define additional logical logical volumes in the default ``data`` volume
group, modify one of the following variables:

* ``seed_lvm_group_data_lvs``
* ``compute_lvm_group_data_lvs``
* ``controller_lvm_group_data_lvs``
* ``monitoring_lvm_group_data_lvs``
* ``storage_lvm_group_data_lvs``

Include the variable ``<host>_lvm_group_data_lv_docker_volumes`` in the list to
include the LVM volume for Docker volume data:

.. code-block:: yaml
   :caption: ``monitoring.yml``

   monitoring_lvm_group_data_lvs:
     - "{{ monitoring_lvm_group_data_lv_docker_volumes }}"
     - lvname: other-vol
       size: 1%
       create: true
       filesystem: ext4
       mount: true
       mntp: /path/to/mount

It is possible to define additional LVM volume groups via the following
variables:

* ``seed_lvm_groups_extra``
* ``compute_lvm_groups_extra``
* ``controller_lvm_groups_extra``
* ``monitoring_lvm_groups_extra``
* ``storage_lvm_groups_extra``

For example:

.. code-block:: yaml
   :caption: ``compute.yml``

   compute_lvm_groups_extra:
     - vgname: other-vg
       disks: /dev/sdb
       create: true
       lvnames:
         - lvname: other-vol
           size: 100%
           create: true
           mount: false

Alternatively, replace the entire volume group list via one of the
``<host>_lvm_groups`` variables to replace the default configuration with a
custom one.

.. code-block:: yaml
   :caption: ``controllers.yml``

   controller_lvm_groups:
     - vgname: only-vg
       disks: /dev/sdb
       create: true
       lvnames:
         - lvname: only-vol
           size: 100%
           create: true
           mount: false

Kolla-Ansible bootstrap-servers
===============================

Kolla Ansible provides some host configuration functionality via the
``bootstrap-servers`` command, which may be leveraged by Kayobe. Due to the
bootstrapping nature of the command, Kayobe uses ``kayobe_ansible_user`` to
execute it, and uses the Kayobe remote Python virtual environment (or the
system Python interpreter if no virtual environment is in use).

See the :kolla-ansible-doc:`Kolla Ansible documentation
<reference/deployment-and-bootstrapping/bootstrap-servers.html>`
for more information on the functions performed by this command, and how to
configure it.

Kolla-Ansible Remote Virtual Environment
========================================
*tags:*
  | ``kolla-ansible``
  | ``kolla-target-venv``

See :ref:`configuration-kolla-ansible-venv` for information about remote Python
virtual environments for Kolla Ansible.

.. _configuration-hosts-docker:

Docker Engine
=============
*tags:*
  | ``docker``

Docker engine configuration is applied by both Kayobe and Kolla Ansible (during
bootstrap-servers).

The ``docker_storage_driver`` variable sets the Docker storage driver, and by
default the ``devicemapper`` driver is used. If using this driver, see
:ref:`configuration-hosts-lvm` for information about configuring LVM for
Docker.

Various options are defined in ``${KAYOBE_CONFIG_PATH}/docker.yml``
for configuring the ``devicemapper`` storage.

A private Docker registry may be configured via ``docker_registry``, with a
Certificate Authority (CA) file configured via ``docker_registry_ca``.

To use one or more Docker Registry mirrors, use the ``docker_registry_mirrors``
variable.

If using an MTU other than 1500, ``docker_daemon_mtu`` can be used to configure
this. This setting does not apply to containers using ``net=host`` (as Kolla
Ansible's containers do), but may be necessary when building images.

Docker's live restore feature can be configured via
``docker_daemon_live_restore``, although it is disabled by default due to
issues observed.