Building Container Images

Firstly, ensure kolla and the container engine of your choice is installed.

Currently supported container engines are docker and podman.

python3 -m pip install kolla
#only one of these is needed:
python3 -m pip install podman
python3 -m pip install docker

Then, the kolla-build command is available for building Docker images.

Building kolla images

In general, images are built like this:

kolla-build

By default, the above command would build all images based on a CentOS Stream image.

The operator can change the base distro with the -b option:

kolla-build -b ubuntu

There are following distros (bases) available for building images:

  • centos

  • debian

  • ubuntu

See the support matrix for information on supported base image distribution versions and supported images on each distribution.

It is possible to build only a subset of images by specifying them on the command line:

kolla-build keystone

In this case, the build script builds all images whose name contains the keystone string, along with their parents.

Multiple names may be specified on the command line:

kolla-build keystone nova

Each string is actually a regular expression so one can do:

kolla-build ^nova-

kolla-build can be configured via an INI file, canonically named kolla-build.conf and placed in /etc/kolla. A custom path to it can be set via the --config-file argument. Most CLI arguments can be set via this config file. Remember to convert the names from hyphenated to underscored. Run kolla-build --help to see all available options.

The set of images to build can be defined as a profile in the profiles section of kolla-build.conf. Then, profile can be specified by --profile CLI argument or profile option in kolla-build.conf.

For example, since Magnum requires Heat, one could add the following profile to profiles section in kolla-build.conf:

[profiles]
magnum = magnum,heat

These images could then be built using command line:

kolla-build --profile magnum

Or putting the following line in the DEFAULT section in kolla-build.conf file:

[DEFAULT]
profile = magnum

The kolla-build uses kolla as default Docker namespace. This is controlled with the -n command line option. To push images to a Dockerhub repository named mykollarepo:

kolla-build -n mykollarepo --push

To push images to a local registry, use the --registry flag:

kolla-build --registry 172.22.2.81:4000 --push

Build OpenStack from source

The locations of OpenStack source code are written in kolla-build.conf. The source’s type supports url, git and local. The location of the local source type can point to either a directory containing the source code or to a tarball of the source. The local source type permits to make the best use of the Docker cache. A source may be disabled by setting enabled to False.

The kolla-build.conf file could look like this:

[horizon]
type = url
location = https://tarballs.openstack.org/horizon/horizon-master.tar.gz

[keystone-base]
type = git
location = https://opendev.org/openstack/keystone
reference = stable/mitaka

[heat-base]
type = local
location = /home/kolla/src/heat

[ironic-base]
type = local
location = /tmp/ironic.tar.gz
enabled = False

Note

Note that the name of the section should exactly match the image name you are trying to change source location for.

If using the local source type, the --locals-base flag can be used to define a path prefix, which you can reference in the config.

[DEFAULTS]
locals_base = /home/kolla/src

[heat-base]
type = local
location = $locals_base/heat

Dockerfile customisation

The kolla-build tool provides a Jinja2-based mechanism which allows operators to customise the Dockerfiles used to generate Kolla images.

This offers a lot of flexibility on how images are built, for example: installing extra packages as part of the build, tweaking settings or installing plugins. Examples of these are described in more detail below.

Note

The Docker file Jinja2 template for each image is found in subdirectories of the docker directory included in the kolla package.

Using a different base image

Base image can be specified using --base-image:

kolla-build --base-image <image-identifier>

The image-identifier accepts any format that Docker accepts when referencing an image.

Generic customisation

Kolla templates are designed such that each Docker file has logical sections represented by Jinja2’s named block section directives. These can be overridden at will by Kolla users. The following is an example of how an operator would modify the setup steps within the Horizon Dockerfile.

First, create a file to contain the customisations, for example: template-overrides.j2. Fill it with the following contents:

{% extends parent_template %}

# Horizon
{% block horizon_ubuntu_source_setup %}
RUN useradd --user-group myuser
{% endblock %}

Then rebuild the horizon image, passing the --template-override argument:

kolla-build --template-override template-overrides.j2 ^horizon$

Note

The above example will replace all contents of the original block. Hence, one may want to copy the original contents of the block before and modify it. Do note it makes the customisations ignore changes in Kolla upstream.

We recommend users use more specific customisation functionalities, such as removing/appending entries for packages. These other customisations are described in the following sections.

Two block series are of particular interest and are safe to override as they are empty by design. The top of each Dockerfile includes <image_name>_header block which can be used for early customisations, such as RHN registration described later. The bottom of each Dockerfile includes <image_name>_footer block which is intended for image-specific modifications. Do note to use the underscored name of the image, i.e., replace dashes with underscores. All leaf Dockerfiles, i.e. those meant for direct consumption, additionally have a footer block which is then guaranteed to exist once at the very end of the image recipe chain.

Packages customisation

Packages installed as part of an image build can be overridden, appended to, and deleted. Taking the Horizon example, the following packages are installed as part of a package install (among others):

  • gettext

  • locales

To add a package to this list, say, iproute, first create a file, for example, template-overrides.j2. In it place the following:

{% extends parent_template %}

# Horizon
{% set horizon_packages_append = ['iproute'] %}

Then rebuild the horizon image, passing the --template-override argument:

kolla-build --template-override template-overrides.j2 ^horizon$

Alternatively template_override can be set in kolla-build.conf.

The append suffix in the above example carries special significance. It indicates the operation taken on the package list. The following is a complete list of operations available:

override

Replace the default packages with a custom list.

append

Add a package to the default list.

remove

Remove a package from the default list.

To remove a package from that list, say locales, one would do:

{% extends parent_template %}

# Horizon
{% set horizon_packages_remove = ['locales'] %}

Python packages build options

The block base_pip_conf in the base Dockerfile can be used to provide the PyPI build customisation options via the standard environment variables like PIP_INDEX_URL, PIP_TRUSTED_HOST, etc. Also here can be provided the standard environment variable UPPER_CONSTRAINTS_FILE used for building the bifrost_deploy container when PyPI upper-constraints needs to be overridden. Also this variable would be used in the kolla-toolbox if provided instead of the defaults.

Plugin functionality

The Dockerfile customisation mechanism is useful for adding/installing plugins to services. An example of this is Neutron’s third party L2 drivers.

For example, to add the networking-cisco plugin to the neutron_server image, one may be tempted to add the following to the template-override file:

Warning

Do NOT do the below. Read on for why.

{% extends parent_template %}

{% block neutron_server_footer %}
RUN git clone https://opendev.org/x/networking-cisco \
    && python3 -m pip --no-cache-dir install networking-cisco
{% endblock %}

Some readers may notice there is one problem with this, however. Assuming nothing else in the Dockerfile changes for a period of time, the above RUN statement will be cached by Docker, meaning new commits added to the Git repository may be missed on subsequent builds. To solve this, the kolla-build tool also supports cloning additional repositories at build time, which will be automatically made available to the build, within an archive named plugins-archive.

To use this, add a section to kolla-build.conf in the following format:

[<image-name>-plugin-<plugin-name>]

Where <image-name> is the hyphenated name of the image that the plugin should be installed into, and <plugin-name> is the chosen plugin identifier.

Continuing with the above example, one could add the following to kolla-build.conf:

[neutron-server-plugin-networking-cisco]
type = git
location = https://opendev.org/x/networking-cisco
reference = master

The build will clone the repository, resulting in the following archive structure:

plugins-archive.tar
|__ plugins
    |__networking-cisco

The template now becomes:

{% block neutron_server_footer %}
ADD plugins-archive /
python3 -m pip --no-cache-dir install /plugins/*
{% endblock %}

Some plugins are installed by default. For images with default plugins, the Dockerfiles already copy the plugins-archive to the image and install available plugins at build time. These default plugins may be disabled by setting enabled to False in the relevant plugin source configuration section in kolla-build.conf.

Neutron plugins

One example of a service with many available plugins is Neutron. The neutron-base image Dockerfile has plugins archive copying and installation enabled already. In the contrib directory of Kolla (as available in the repository, the tarball or the share directory of the installation target), there is a neutron-plugins directory with examples of Neutron plugins definitions. Some of these plugins used to be enabled by default but, due to their release characteristic, have been excluded from the default builds. Please read the included README.rst to learn how to apply them.

Additions functionality

The Dockerfile customisation mechanism is useful for adding/installing additions into images. An example of this is adding your jenkins job build metadata (say, formatted into a jenkins.json file) into the image.

Similarly to the plugins mechanism, the Kolla build tool also supports cloning additional repositories at build time, which will be automatically made available to the build, within an archive named additions-archive. The main difference between plugins-archive and additions-archive is that plugins-archive is automatically copied in many images and processed to install available plugins while additions-archive processing is left solely to the Kolla user.

To use this, add a section to kolla-build.conf in the following format:

[<image>-additions-<additions-name>]

Where <image-name> is the hyphenated name of the image that the additions should be copied into, and <additions-name> is the chosen additions identifier.

For example, one could add the following to kolla-build.conf file:

[neutron-server-additions-jenkins]
type = local
location = /path/to/your/jenkins/data

The build will copy the directory, resulting in the following archive structure:

additions-archive.tar
|__ additions
    |__jenkins

The template becomes now:

{% block neutron_server_footer %}
ADD additions-archive /
RUN cp /additions/jenkins/jenkins.json /jenkins.json
{% endblock %}

Custom repos

Red Hat

Kolla allows the operator to build containers using custom repos. The repos are accepted as a list of comma separated values and can be in the form of .repo, .rpm, or a url. See examples below.

To use current RDO packages (aka Delorean or DLRN), update rpm_setup_config in kolla-build.conf:

rpm_setup_config = https://trunk.rdoproject.org/centos8/current/delorean.repo,https://trunk.rdoproject.org/centos8/delorean-deps.repo

If specifying a .repo file, each .repo file will need to exist in the same directory as the base Dockerfile (kolla/docker/base):

rpm_setup_config = epel.repo,delorean.repo,delorean-deps.repo

Debian / Ubuntu

For Debian based images, additional apt sources may be added to the build as follows:

apt_sources_list = custom.list

Building behind a proxy

We can insert http_proxy settings into the images to fetch packages during build, and then unset them at the end to avoid having them carry through to the environment of the final images. Note, however, it’s not possible to drop the info completely using this method; it will still be visible in the layers of the image.

To set the proxy settings, we can add this to the template’s header block:

ENV http_proxy=https://evil.corp.proxy:80
ENV https_proxy=https://evil.corp.proxy:80

To unset the proxy settings, we can add this to the template’s footer block:

ENV http_proxy=""
ENV https_proxy=""

Besides this configuration options, the script will automatically read these environment variables. If the host system proxy parameters match the ones going to be used, no other input parameters will be needed. These are the variables that will be picked up from the user env:

HTTP_PROXY, http_proxy, HTTPS_PROXY, https_proxy, FTP_PROXY,
ftp_proxy, NO_PROXY, no_proxy

Also these variables could be overwritten using --build-args, which have precedence.

Cross-compiling

It is possible to cross-compile container images in order to, e.g., build aarch64 images on a x86_64 machine.

To build ARM images on x86_64 platform, pass the --base-arch and --platform arguments:

kolla-build --platform linux/arm64 --base-arch aarch64

Note

To make this work on x86_64 platform you can use tools like: qemu-user-static or binfmt.

To make this work on Apple Silicon you can use Docker Desktop or Podman Desktop to build x86_64 or native ARM images.

Known issues

  1. Mirrors are unreliable.

    Some of the mirrors Kolla uses can be unreliable. As a result occasionally some containers fail to build. To rectify build problems, the build tool will automatically attempt three retries of a build operation if the first one fails. The retry count is modified with the --retries option.