Wikipedia said: In much the same way as the trade or vocation of smithing produced the common English surname Smith and the German name Schmidt, the cooper trade is also the origin of German names like Kübler.

There is still demand for high-quality wooden barrels containers, and it is thought that the highest-quality barrels containers are those hand-made by professional coopers kublers.

Kubler

A generic, extendable build orchestrator, in Bash. The default batteries focus on creating and maintaining Docker base images.

Table Of Contents

• Why Should You Care?
• Requirements
  • Kubler
  • Docker
• Installation
  • On Gentoo
  • On macOS
  • Manual Installation
  • Initial Configuration
  • Uninstall
• Tour de Kubler
  • The Basics
  • Every Image needs a Home - Working Dirs and Namespaces
  • Hello Image
  • Anatomy of an Image
  • Understanding the Build Process
  • But Does it Work? - Image Tests
  • Common Build Pitfalls
  • Custom Build Containers
    • Extend Existing Builder
    • New Builder from Scratch
  • Updating Build Containers
  • Pushing Images to a Docker Repository
  • Handling Software that doesn't have an Ebuild (yet ;)
• Other Resources
• Discord

Why Should You Care?

Perhaps:

1. You love Docker but are annoyed by some of the restrictions of it's build command that keep getting into your way. Wouldn't it be nice if you could docker build your images with all docker run args, like -v, at your disposal? Or if your Dockerfile was fully parameterizable?
2. You are a SysAdmin or DevOps engineer who seeks complete governance over the contents of their container images, with full control of the update cycle and the ability to track all software version changes from a centralized vcs repository.
3. You need to manage a lot of Docker base/service images in a sane way and want peace of mind with automated post-build tests.
4. You are a Gentoo user and want to build slim Docker images with a toolset you are familiar with. Not having to wrestle with CrossDev would also be a plus.
5. You are looking for an interactive OS host agnostic Gentoo playground or portable ebuild development environment.
6. You want to create custom root file systems, possibly for different cpu architectures and/or libc implementations (i.e. musl, uclibc, etc) in an isolated and repeatable manner.

Requirements

Kubler

• Bash version 4.2+, using 4.4+ is highly recommended due to bugs in previous versions.

Optional:

• GPG for download verification

# Assuming a working gpg setup just import the Gentoo ebuild repository signing key:
gpg --keyserver keys.gentoo.org --recv-keys E1D6ABB63BFCFB4BA02FDF1CEC590EEAC9189250

• rlwrap for command history

Docker or Podman

• Working Docker or Podman setup
• Git
• jq to parse Docker/Podman json output

Installation

On Gentoo

An ebuild can be found at https://github.com/edannenberg/kubler-overlay/

Add the overlay (see link for instructions) and install as usual:

emerge -av kubler

On macOS

The standard version of Bash in macOS is too old. Easiest way to upgrade to a later version is to use Homebrew. Once Homebrew is installed, update Bash:

    $ brew install bash

This will install an updated version of Bash in /usr/local/bin/. To make it your default shell, you need to edit Advanced Options... in System Preferences. Just right-click your user icon to find the option.

Also, macOS does not load ~/.bashrc by default, but uses ~/.bash_profile, so when following the instructions below, make sure to edit the correct file.

Manual Installation

Kubler has been tested on Gentoo, CoreOS and macOS. It should run on all Linux distributions. Feel free to open an issue or ask on Discord if you run into problems.

1. Clone the repo or download/extract the release archive to a location of your choice, i.e.

    $ cd ~/tools/
    $ curl -L https://github.com/edannenberg/kubler/archive/master.tar.gz | tar xz

2. Optional, add kubler.sh to your path

The recommended way is to add the following at the end of your ~/.bashrc file, don't forget to adjust the Kubler path for each line accordingly:

export PATH="${PATH}:/path/to/kubler/bin"
# optional but highly recommended, adds bash completion support for all kubler commands
source /path/to/kubler/lib/kubler-completion.bash

Note: You will need to open a new shell for this to take effect, if this fails on a Linux SystemD host re-logging might be required instead.

Initial Configuration

Kubler doesn't require any further configuration but you may want to review the main config file located at /etc/kubler.conf. If the file doesn't exist the kubler.conf file in Kubler's root folder is used as a fallback.

All of Kubler's runtime data, like user config overrides, downloads or custom scripts, is kept at a path defined via KUBLER_DATA_DIR. This defaults to ~/.kubler/, which is suitable if user accounts have Docker/Podman access on the host. If you plan to use Docker/Podman/Kubler only with sudo, like on a server, you may want to use /var/lib/kubler, or some other location, as data dir instead.

Managing your KUBLER_DATA_DIR with a VCS tool like Git is supported, a proper .gitignore is added on initialization.

Uninstall

1. Remove any build artifacts and container images created by Kubler:

    $ kubler clean -N

2. Remove Kubler itself:

   • On Gentoo and ebuild install: emerge -C kubler then remove the kubler overlay
   • Manual install: reverse the steps you did during manual installation

3. Delete any namespace dirs and configured KUBLER_DATA_DIR (default is ~/.kubler/) you had in use, this may require su permissions.

Tour de Kubler

The Basics

To get a quick overview of available commands/options:

$ kubler --help

Or to view details for a specific command:

$ kubler build -h

Per default almost all of Kubler's commands will need to be run from a --working-dir, if the option is omitted the current working dir of the executing shell is used. It behaves much like Git in that regard, executing any Kubler command from a sub directory of a valid working dir will also work as expected.

A --working-dir is considered valid if it has a kubler.conf file and either an images/ dir, or one ore more namespace dirs, which are just a collection of images.

Kubler currently ships with Docker and Podman build engines. The rest of this tour will focus on building Docker images, it's worth noting that the build process may be completely different, i.e. it may not involve Gentoo or Docker at all, for other build engines.

If you are not familiar with Gentoo some of it's terms you will encounter may be confusing, a short 101 glossary:

stage3	A tar ball provided by Gentoo which on extraction provides an almost-complete root file system for a Gentoo installation
Portage	Gentoo's default package manager, this is where all the magic happens
emerge	Portage's main executable
ebuild	text file which identifies a specific software package and how Portage should handle it
Portage Tree	Categorized collection of ebuilds, Gentoo ships with ~20k ebuilds
Portage Overlay	Additional ebuild repository maintained by the community/yourself

Every Image needs a Home - Working Dirs and Namespaces

To accommodate different use cases there are three types of working dirs:

multi	The working dir is a collection of one or more namespace dirs
single	The working dir doubles as namespace dir, you can't create a new namespace in it, but you save a directory level
local	Same as multi but --working-dir is equal to KUBLER_DATA_DIR

First switch to a directory where you would like to store your Kubler managed images or namespaces:

$ cd ~/projects

Then use the new command to take care of the boiler plate, choose 'single' when asked for the namespace type:

    $ kubler new namespace mytest
    $ cd mytest/

Although not strictly required it's recommended to install Kubler's example images by running:

$ kubler update

Hello Image

Let's start with a simple task and dockerize [Figlet][], a nifty tool that produces ascii fonts. First create a new image stub:

$ kubler new image mytest/figlet

When asked for the image parent, enter kubler/bash and bt when asked for tests:

»»» Extend an existing Kubler managed image? Fully qualified image id (i.e. kubler/busybox) or scratch
»[?]» Parent Image (scratch): kubler/bash
»»»
»»» Add test template(s)? Possible choices:
»»»   hc  - Add a stub for Docker's HEALTH-CHECK, recommended for images that run daemons
»»»   bt  - Add a stub for a custom build-test.sh script, a good choice if HEALTH-CHECK is not suitable
»»»   yes - Add stubs for both test types
»»»   no  - Fck it, we'll do it live!
»[?]» Tests (hc): bt
»»»
»[✔]» Successfully created new image at projects/mytest/images/figlet

A handy feature when working on a Kubler managed image is the --interactive build arg. As the name suggests it allows us to poke around in a running build container and plan/debug the image build. Let's give it a try:

$ kubler build mytest/figlet -i

This will also build any missing parent images/builders, so the first run may take quite a bit of time. Don't worry, once the local binary package cache and build containers are seeded future runs will be much faster. When everything is ready you are dropped into a new shell:

»[✔]»[kubler/bash]» done.
»»»»»[mytest/figlet]» using: docker / builder: kubler/bob-bash
kubler-bob-bash / #

To search Portage's package db you may use eix, or whatever your preferred method is:

kubler-bob-bash / # eix figlet
* app-misc/figlet
     Available versions:  2.2.5 ~2.2.5-r1
     Homepage:            http://www.figlet.org/
     Description:         program for making large letters out of ordinary text

* dev-php/PEAR-Text_Figlet

As with most package managers, software in Portage is grouped by categories. The category and package name combined form a unique package atom, in our case we want to install app-misc/figlet.

Now manifest the new found knowledge by editing the image's build script:

    kubler-bob-bash / # nano /config/build.sh

Note: The /config folder in the build container is the host mounted image directory at mytest/images/figlet/. Feel free to use your local IDE/editor to edit build.sh instead.

Add the app-misc/figlet package atom to the _packages variable in build.sh:

_packages="app-misc/figlet"

Then start a test run of the first build phase (more on that later), if you are in a hurry you may skip this step:

kubler-bob-bash / # kubler-build-root

Once this finishes exit the interactive builder by hitting crtl+d or typing exit. All that is left to do is building the actual image:

$ kubler build mytest/figlet -nF

The args are short hand for --no-deps and --force-full-image-build, omitting -n would also rebuild all parent images, which can be handy but is just a waste of time in this case.

    »[✘]»[mytest/figlet]» fatal: build-test.sh for image mytest/figlet:20190228 failed with exit signal: 1

Oops, looks like we forgot the image test. Let's fix that by editing the mentioned build-test.sh file:

    #!/usr/bin/env sh
    set -eo pipefail

    # check figlet version string
    figlet -v | grep -A 2 'FIGlet Copyright' || exit 1

Not exactly exhausting but it will do for now. Rebuild the image again but this time only pass -f instead of -F, this too forces an image rebuild but skips the first build phase:

$ kubler build mytest/figlet -nf
»[✔]»[mytest/figlet]» done.
$ docker run -it --rm mytest/figlet figlet foooo

Anatomy of an Image

$ tree images/figlet
images/figlet/
├── Dockerfile            <- generated, never edit this manually
├── Dockerfile.template   <- standard Dockerfile, except it's fully parameterizable
├── PACKAGES.md           <- generated, lists all installed packages with version and use flags
├── README.md             <- optional, image specific documentation written by you
├── build-test.sh         <- optional, if the file exists it activates a post-build test
├── build.conf            <- general image/builder config, sourced on the host
├── build.sh              <- configures the first build phase, only sourced in build containers

The stub files generated with the new command are heavily commented with further details.

Understanding the Build Process

After executing a build command an image dependency graph is generated for the passed target ids by parsing the IMAGE_PARENT and BUILDER vars in the respective build.conf files. You can visualize the graph for any given target ids with the dep-graph command:

$ kubler dep-graph -b kubler/nginx mytest

Once all required data is gathered, each missing, as in not already built, image will go through a two phase build process:

1. The configured builder image is passed to docker run to produce a rootfs.tar file in the image folder

   • mounts current image dir into a fresh build container as /config
   • executes build-root.sh (a generic script provided by Kubler) inside build container
   • build-root.sh reads build.sh from the mounted /config directory
   • if configure_builder() hook is defined in build.sh, execute it
   • package.installed file is generated which is used by depending images as [package.provided][]
   • ROOT env is set to custom path
   • if configure_rootfs_build() hook is defined in build.sh, execute it
   • _packages defined in build.sh are installed via Portage at custom empty root directory
   • if finish_rootfs_build() hook is defined in build.sh, execute it
   • ROOT dir is packaged as rootfs.tar and placed in image dir on the host
   • preserve exact builder state for child images by committing the used build container as a new builder image

The build-root.sh file effectively just uses a feature of Gentoo's package manager that allows us to install any given _packages, with all it's dependencies, at a custom path by setting the ROOT env in the build container. The other piece to the puzzle is Portage's [package.provided][] file which is constantly updated and preserved by committing the build container as a new builder image after each build. Thanks to Docker's shared layers the overhead of this is fairly minimal.

Kubler's default build container names generally start with bob, when a new build container state is committed the current image name gets appended. For example kubler/bob-openssl refers to the container used to build the kubler/openssl image. Any image that has kubler/openssl as IMAGE_PARENT will use kubler/bob-openssl as it's build container.

There are no further assumptions or magic, the hooks in build.sh