Learning AlpineLinux.org, Ansible.com, and Packer.io

For my 2020-1 holiday break, I decided my next personal project would be to rebuild my home infrastructure on containers. I’d been inspired by a recent discovery of LinuxServer.io, where open source projects are built and run as containers. While I have a burning desire to better my Kubernetes practice, not enough of the world delivers software as a Helm chart, so merely running everything as containers would be a good intermediate step. Let me share my first six steps with examples…

A Pre-destined Journey

My previous three employers had provided an Apple laptop and in 2020, I requested to a Dell XPS laptop with Windows. I made the XPS dual-boot Linux (because I knew that Dell line supported Linux via Sputnik project). I had refined my strategic practices over the years to make my work mostly OS independent and portable, adapting everything I could towards open source applications, version controlled work, and containers when possible. The tactical choices I made along the way to support these strategic goals were exemplified by adopting git, dotfiles, KeePassXC.org, and Unison. The ultimate goal was to “containerize” as many applications and services as possible and manage them with Kubernetes, but take smaller, logical steps to arrive there.


When COVID-19 shut-down travel, my 10 year old Linux desktop became my primary workstation. Fortunately, I’d upgraded my monitor earlier that year to a 32" UHD display, so I was set. With a lot of polishing, I migrated all of my day to day work to Linux from Mac. Today, the primary reason I use my Mac is for being on the road, videoconferencing, and Powerpoint.

Update: February 2021 The following month, I found I could no longer use Homebrew’s Linux compiled Python because my older AMD Athelon Phenom II processor doesn’t support the Intel Core 2 instruction set, which was now required. Of course, my OS provides Python as a workaround, but this wasn’t the first time I had encountered a processor compile issue. Exactly a decade later (to the month), I upgraded my desktop to the current AMD Rizen 3000 desktop CPU/motherboard/RAM. COVID-19 has impacted choice and availability in the market. Ironically, I’ve yet to feel much of a difference in performance, but I’ve finally shed IDE and Serial port peripherals to focus on SATA and USB exclusively.

I had personally invested into buying double the memory for a work supplied Intel Skull Canyon i5 NUC (due to Nutanix’s partnership with Intel), but it always struggled to run Nutanix Community Edition with AOS, Prism Central, and Calm plus some workload VMs and furthermore, community support would challenge my productivity. So it was idle, smaller, faster, and quieter than my desktop, and ideal to host this holiday project as a headless server.

Operating System

I considered loading my Linux desktop OS onto the Skull Canyon for familiarity, OS parity, and ease of maintenance (even caching OS updates across them). But my strategic goal was to boot and run containers, that meant the OS should be focused on “boot to Docker” and eventually “boot to Kubernetes” which meant that it should be a minimal Linux OS, not a general purpose desktop focus.

I investigated the current minimal Ubuntu or Debian server variant distributions to remain in the .deb package format Linux family branch. But for years, the majority of the Docker community has settled on using AlpineLinux.org as a base OS image for containers because it is small and secure. The AlpineLinux team’s pragmatic and relentless focus means that the OS minimizes (memory and disk) resource use, performs well, and updates regularly, so I pursued it.

After importing the AlpineLinux.org GPG key, downloading, checking,


for FILE in '' .asc .sha256; do
  wget --continue ${URL}${FILE}

IFS='/' read -ra FILESPEC <<< "${URL}"
gpg --verify ${FILESPEC[-1]}.asc \
  && sha256sum --check ${FILESPEC[-1]}.sha256 \
  && echo -e "\n__VERIFIED: ${FILESPEC[-1]}" \
  && rm -i ${FILESPEC[-1]}.*

…and writing the AlpineLinux Extended 3.12.3 x86-64 ISO (525 MB) to a USB thumb drive, I booted it up on the Skull Canyon. This was important to see how much of the hardware was supported by Alpine and worked “out of the box.” Although the ethernet NIC wasn’t recognized, WiFi did work. So I perused wiki.AlpineLinux.org guides, logged in as root, and began a CLI journey to learning Alpine!

Reproducible Pets

I knew that everything I did by hand was bespoke, so like any good scientist, I kept notes on goals, facilities (research and references), and procedures (mostly shell commands). My blog journey had already ingrained the habits of using Markdown with my choice of IDE with Markdown Preview, but a recent discovery of Typora as a great hybrid Markdown WYSIWYG editor had spoiled me. Now I had a nice visual notebook to log my work, reminiscent of my past work with Project Jyupiter and Mathematica!

Documenting PetOps (manual operations made by hand) was not enough, the goal was to automate my work.

setup-alpine accepts environment variables and command line arguments or an answer file for automation, so I set about to learn just enough of the initial configuration, service (NTP, SSH), network, and disk choices to install the OS to the disk. I created a stub answer file, adapted it with what I knew (hard-coded every value), and worked to repeat the OS install with the answer file. This small loop guided my research, documentation, and answer file development! (Red-green-refactor!)

The generic answer file setup-alpine -c generates is a configuration file with key=“value” and # comment format, compatible with Java.properties or TOML. My editor can lint TOML, so I decided to organize answer files in a directory: alpine/answerfile/${HOSTNAME}.toml.

I SCP’d the answer file and any other supporting notes, scripts, or configuration off the Skull Canyon and kept the results along with my alpine/README.md documentation in Git. All of this work was done on the Skull Canyon while hooked up to a keyboard and monitor. I remember the days of VMK (Video, Mouse, Keyboard) switches and cables to each physical server in the data center, but I was eager to get the Skull Canyon off my desk and work on it remotely.

Here’s my development VM answer file = alpine/answerfile/skull-vm-packer-dev.toml:

# Answer file for setup-alpine script; tested on: cat /etc/alpine-release
# 3.12.3, 3.13.0
HOSTNAMEOPTS="-n skull-vm-packer-dev"

# Search domain, public nameserver
DNSOPTS="-d example.com"

# Contents of /etc/network/interfaces
iface lo inet loopback

auto eth0
iface eth0 inet dhcp
    hostname skull-vm-packer-dev

# If you don't want to use a certain option, then comment it out
# -r: Add a random mirror, -f = detect fastest

SSHDOPTS="-c openssh"
NTPOPTS="-c chrony"

DISKOPTS="-m sys -v /dev/vda"

The next goal was to setup a regular account with sudo access to perform post-OS install work, which became a second script. Now I could remotely work on the Skull Canyon! Having achieved a beach head with remote access to my pet Skull Canyon “production” server, it was time to make a virtual pet server and experiment with it safely while also speeding up development.

I used VirtualBox’s GUI to create a VM from the same Alpine ISO and figured out the basic VM resource configuration for this development instance (memory, storage, bridge network, etc.). Then I learned how to script changing the VM disk using VirtualBox’s CLI VBoxManage, rather than create a new VM from scratch each time in the GUI, to simulate a new OS install and improve development cycle time. I minimized the VM disk size to further speed each cycle.

I tried three experiments to further improve Alpine VM development:

  1. An Alpine Vagrant box is convenient, but it substituted my source ISO with a less vetted source and required re-learning additional Vagrant configuration to manipulate VirtualBox.
  2. Adding VirtualBox Guest Additions to Alpine is a work in progress that seems to require an entire desktop installation, which is unnecessary for a headless server. I missed being able to cut and paste between guest VM and my host Desktop, but this helped keep me honest for state transfer of reproducible scripts and files for my work on the VM guest, not hand-maintained PetOps.
  3. Cloud-init is in the (rolling OS) edge repository, but I wanted to remain on a more fixed, stable OS release.

I was spoiled by the small customizations and optimizations I could make with VBoxManage, which had quickened my development cycle time into a vm-init.sh launch script which handled:

  1. Rip down a running VM
  2. Create a new VM disk
  3. Start a Python HTTP web server (to supply the answer file and bootstrapping scripts)
  4. Start the VM with specific resource configurations and media as needed

I worked to refactor many of the hard-coded values out of vm-init.sh and move them to a DirEnv .env file, which automatically populate my shell’s environment variables when I change into an enabled directory. I configured .gitignore to omit any .env file, so I will not accidentally commit any sensitive information, such as a password. By moving the hard-coded values to environment values, my code is much more portable and stable because it supports different scenarios without change. The environment variables can be further leveraged for Docker, Ansible, and Packer. Finally, I can think about how to organize the environment variables for better re-use and consistency across tools and environments.

Let me share my current DirEnv’s .envrc and .env configuration files, then my vm-init.sh guest maintenance script. You can see I override VM_NAME and MEDIUM_ISO rather than comment them out in the .env file. I’ve also expanded the maintenance script for more use cases and you will see some cloud-init experimental work.

$ cat .envrc
# https://direnv.net/man/direnv-stdlib.1.html
# EDITOR=vi direnv edit . && direnv allow # enable delta

if on_git_branch; then
  echo && git status --short --branch &&
    echo && git fetch --verbose

$ cat .env
# megabytes = 6GB*1024MB/GB (was 4GB)
# sda3 4.3G
# df -hs | grep -e Size -e sda
# CREATEMEDIUM_VARIANT=Variable didn't work with Alpine: too primitive || need VMBox.ext?
VM_DIR=VirtualBox VMs

$ cat vm-init.sh
#!/usr/bin/env bash

# vboxmanage controlvm "${VM_NAME}" reset #poweroff
#__ static and defaults moved to .env
# https://www.virtualbox.org/manual/UserManual.html

function vm_create() {
  # TODO: create new VM props, we are recreating Vagrant :(
  local _image_mount
  local _image_raw
  local _image_vmdk

  echo "|CREATE|"

  VBoxManage unregistervm "${VM_NAME}" --delete || true
  VBoxManage createvm --name "${VM_NAME}" --ostype "${OSTYPE}" --register
  # VBoxManage createhd --filename /VirtualBox/${VM_NAME}/$VM.vdi --size 32768
  VBoxManage storagectl "${VM_NAME}" --name 'IDE' --add ide
  # VBoxManage storageattach "${VM_NAME}" --storagectl "IDE Controller" --port 0 --device 0 --type dvddrive --medium /u01/Software/OL/OracleLinux-R7-U6-Server-x86_64-dvd.iso
  VBoxManage storagectl "${VM_NAME}" --name 'SATA' --add sata --controller IntelAHCI
  #VBoxManage storageattach "${VM_NAME}" --storagectl "SATA Controller" --port 0 --device 0 --type hdd --medium /VirtualBox/$VM/$VM.vdi

  VBoxManage modifyvm "${VM_NAME}" --ioapic on \
    --boot1 disk --boot2 dvd --boot3 none --boot4 none \
    --memory 1024 --vram 33 \
    --bioslogodisplaytime 20 \
    --nic1 bridged --bridgeadapter1 enp5s0 \
    --mouse ps2 --clipboard bidirectional \
    --audio oss --audioout on

  echo -e '\n|CLOUD-INIT|development...'

  sudo umount "${_image_mount}" || true
  rm -f "${_image_raw}" "${_image_vmdk}" || true

  dd if=/dev/zero of="${_image_raw}" \
    bs=1 count=0 seek=2M 2> /dev/null \
    && mkfs.vfat -n CIDATA "${_image_raw}" \
    && (sudo mkdir --parents "${_image_mount}" 2> /dev/null || true) \
    && sudo mount "${_image_raw}" "${_image_mount}" \
    && mount -l | grep "${_image_mount}" \
    && echo \
    && sudo cp cloud-init/meta-data "${_image_mount}/" \
    && echo 'instance-id: skull-vm-1' \
    | sudo tee "${_image_mount}/user-data" \
    && ls -l "${_image_mount}"/*-data \
    && sudo umount "${_image_mount}" \
    && echo -e '\n\thttps://stackoverflow.com/questions/454899/how-to-convert-flat-raw-disk-image-to-vmdk-for-virtualbox-or-vmplayer' \
    && VBoxManage convertfromraw "${_image_raw}" "${_image_vmdk}" --format VMDK \
    && vboxmanage storageattach "${VM_NAME}" --type hdd \
      --medium "${_image_vmdk}" \
      --storagectl "${STORAGEATTACH_STORAGECTL}" \
      --port $((STORAGEATTACH_PORT + 1)) \
    && echo "|ADD|${_image_vmdk}"

if (($(vboxmanage list runningvms | grep --count "${VM_NAME}") > 0)); then
  echo "|POWER|${VM_NAME} ${POWER} ..."
  vboxmanage controlvm "${VM_NAME}" "${POWER}" \
    && sleep 2

if [[ -z ${1} ]]; then

  if vboxmanage showvminfo "${VM_NAME}" > /dev/null; then
    INDEX=$(vboxmanage showvminfo "${VM_NAME}" \
      | grep "(${STORAGEATTACH_PORT}," \
      | awk -F".${CREATEMEDIUM_FORMAT} " '{print $1}' \
      | awk -F"${SUFFIX}" '{print $NF}')

  if (($((INDEX)) >= 1)); then
    echo "|INDEX|Guess: ${INDEX}"
    echo "|DEBUG|INDEX=${INDEX}|"
    echo "|ERROR|${ERROR}|REQUIRED: the numeric index of the current mounted VM Disk for: ${VM_NAME}"
    exit ${ERROR}

  if ((INDEX == 0)); then
    echo "|USAGE|${0} [Integer: new disk index] [VM_NAME]"

if [[ -z ${2} ]]; then
  echo "|WARNING|Using: ${VM_NAME}"

#__ derived:
# echo "|DEBUG|INDEX=${INDEX}|"

if [[ ! -e ${NEW_PATH} ]]; then
  vboxmanage createmedium disk --format "${CREATEMEDIUM_FORMAT}" \
    --filename "${NEW_PATH}" --variant "${CREATEMEDIUM_VARIANT}" \
    --size "${CREATEMEDIUM_SIZE}"
  echo "|EXISTS|${NEW_PATH}"

vboxmanage storageattach "${VM_NAME}" --type hdd --medium "${NEW_PATH}" \
  --storagectl "${STORAGEATTACH_STORAGECTL}" \
  --port "${STORAGEATTACH_PORT}" \
  && echo "|ADD|${NEW_PATH} ..." \
  && if [[ -e ${OLD_PATH} ]]; then
    vboxmanage closemedium --delete disk "${OLD_PATH}" \
      && echo "|DELETE|${OLD_PATH}"
    echo "|GONE|${OLD_PATH}"

# TODO: change boot order?
vboxmanage storageattach "${VM_NAME}" \
  --storagectl IDE --port 1 --device 0 --type dvddrive \
  --medium "${MEDIUM_ISO}" \
  && echo -e "|ADD|CD-ROM@IDE, port 1, slot 0|boot media: ${MEDIUM_ISO}"

# echo -e "|START|${VM_NAME}.\n|SLEEP|${SLEEP}|to allow ^Cancel..."
# sleep "${SLEEP}"

echo && sleep 4 && echo '|START|...' \
  && vboxmanage startvm "${VM_NAME}" --type gui & # headless || gui

vboxmanage showvminfo "${VM_NAME}" --machinereadable \
  > "${VM_NAME}".machinereadable.toml

# sudo systemctl disable apache2
pkill --full "python -m SimpleHTTPServer ${PORT}" \
  && echo -e '\n|TERMINATE|python.http server...'
  python -m SimpleHTTPServer "${PORT}" \
    || python -m http.server "${PORT}"
) \
  && echo -e '\n|START|python.http server...' &

ssh-keygen -f "${HOME}/.ssh/known_hosts" \
  -R 'skull-vm' -R 'skull-vm-dev' || true

if (($(pgrep VBoxXPCOMIPCD) > 0)); then
  echo 'TODO: start GUI via: virtualbox &'

echo "\
 lost: mount setlogo=0 and controlvm run \
 vboxmanage controlvm ${VM_NAME} reset #poweroff \

" && head -n7 init.sh

From the script above, I wanted to call attention to this command:

`vboxmanage showvminfo "${VM_NAME}" --machinereadable "${VM_NAME}".machinereadable.toml`

which allows me to easily find the textual difference between any two VM setups, e.g: diff alpine/*.machinereadable.toml, to understand what I needed to synthesize the same results by building an infrastructure VM. This is a critical differentiation for infrastructure as code versus the standard operator facilities of snapshot and clone in order to reproduce VM results. Furthermore, this leads to the path of building infrastructure VM artifacts in a later section. When we act more like a scientist than a craftsman, we can amplify that knowledge gain across any organization. Of course, I aspire to be a craftsman scientist: the hybrid is more interesting and relatable. The journey to improve one’s skills and while pursuing or improving the state of the art will never end!

I continued development of my bootstrap procedure and scripts by creating an initialization script (init.sh), which would SCP the answer file and a post-OS install script (post.sh) from my desktop, then use them.

Alpine’s default shell is busybox, a reduced form of ash. Fortunately, shellcheck is a nice addition to my IDE and it kept me in good form and syntax as I developed these scripts. I didn’t have to install and test busybox directly on my Desktop OS or install bash on Alpine, which would have required further bootstrapping during the initialization stage.

I re-factored init.sh from using scp to use git and then to use http for file transfer with wget, removing the need to install any packages. You can see in my examples that I have a habit of keeping copious notes in my code. Notes can consist of goals, pseudo code, lots of URLs, #TODOs, and #TOFIX notes. All of the time, I would improve capabilities while moving procedures and references out of my code and into documentation when I found a better method, an alternate facility, or a dead-end. I also moved work between init.sh and post.sh where appropriate to minimize setup before the first reboot to boot the OS from storage.

I arrived at a point where I needed to type three commands to bootstrap the development VM with a fully automated OS install and reboot, leaving one post OS configuration script that would survive the reboot. You will see this pattern repeated and most of the scripts reused in the the VM Artifact section!

KEYMAPOPTS='us us' setup-alpine -q # keep commands in this order!
hostname skull-vm
wget http://desktop:port/init.sh -O - | sh

I learned about the OpenRC init system on Alpine, but did not automate the invocation of post OS configuration on first boot because I was eager to start on the next step: where fully remote automation work would take over.

Let me share my current init.sh and post.sh, these have been re-factored to accommodate bootstrap.sh for Packer in the “VM Artifacts” section.

$ cat init.sh
#!/usr/bin/env sh
# tested on: 3.12.3, 3.13.0 # cat /etc/alpine-release
# packer example from wiki.alpinelinux.org: "ifconfig eth0 up && udhcpc -i eth0<enter><wait5>",
#--- Diagnostics:
# cat /etc/network/interfaces; echo; cat /etc/apk/repositories
# ip link && echo && ip route list && echo && ip addr show
# mount | grep md


# shellcheck disable=SC2039
echo -e "${_artifact_ip}\t${_artifact_host}" >> /etc/hosts \
  && wget "${_artifact_url}post.sh" -O /etc/post.sh \
  && echo "sh /etc/post.sh" >> /etc/motd \
  && wget "${_artifact_url}id_rsa.pub" -O /etc/id_rsa.pub \
  && wget "${_artifact_url}answerfile/$(hostname).toml" \
  && wget "${_artifact_url}bootstrap.sh" -O - | sh

$ cat post.sh
#!/usr/bin/env sh

_remote_user='ansible' # defined in ansible/.env, still some hardcodes

apk_repo_add() {
  _repos='community' # community testing'

  # shellcheck disable=SC2015
  for _repo in ${_repos}; do

    if [ ! -e ${_repositories}.original ]; then
      cp ${_repositories} ${_repositories}.original

    grep '#' ${_repositories}.original \
      | grep ${_repo} \
      | head -n 1 \
      | tr -d '#' >> ${_repositories} \
      && echo "${_repositories}: enabled ${_repo}" >> /etc/motd

remote_user_group_add() {
  addgroup -g ${_remote_group_id} ${_remote_group} || true
  # apk update &&
  # apk add bash direnv libuser &&
  #   mkdir /etc/default &&
  #   touch /etc/default/useradd /etc/login.defs &&
  #   echo -e "/bin/bash\n" | lchsh root &&
  #   echo 'dotenv' > .envrc &&
  #   direnv hook bash > .profile &&
  #   wget ${_artifact_url}.env &&
  #   direnv allow &&

  adduser -D -h /home/${_remote_user} \
    -u ${_remote_user_id} \
    -G ${_remote_group} \
    ${_remote_user} || true

  addgroup ${_remote_user} wheel || true

  su ${_remote_user} -c \
    "mkdir /home/${_remote_user}/.ssh || true &&
      cat /etc/id_rsa.pub \
      >> /home/${_remote_user}/.ssh/authorized_keys &&
      chmod go-rwx /home/${_remote_user}/.ssh/authorized_keys"
case "${1}" in
    echo "__TOFIX: /usr/bin/env bash ${0} stage2" \
      && /usr/bin/env bash "${0}" stage2

    delgroup ${_remote_user} ${_remote_group} \
      && deluser --remove-home ${_remote_user} \
      && /usr/bin/env bash "${0}" stage2

    # APK Dependencies
    # Ansible: openssh-client python
    # docker{-compose},cloud-init@community repo

    if [ "$(hostname)" = 'skull-vm-dev' ] \
      && [ "$(grep -c '3.13.' /etc/alpine-release)" -gt 0 ]; then
      apk add cloud-init util-linux \
        && setup-cloud-init
      # shellcheck disable=SC2039
      # shellcheck disable=SC2015
      echo -e "\n$0: done!" >> /etc/motd \
        && tail -n 1 /etc/motd \
        && rm -f .env || true

    # shellcheck disable=SC2016
    apk add openssh-client python3 \
      && echo 'TODO: echo "${_remote_root_password}" | chpasswd' \
      && cat /etc/alpine-release >> /etc/motd \
      && mv /etc/profile.d/color_prompt /etc/profile.d/color_prompt.sh

    # https://wiki.syslinux.org/wiki/index.php?title=Config#GLOBAL_DIRECTIVES_-_SECONDARY
    echo 'TIMEOUT 15' >> /boot/extlinux.conf

    echo && df -hT | grep -e Size -e da
    #rm -f /etc/post.sh /etc/id_rsa.pub || true

My OS install was reproducible in development on a VM, so I worked to use these procedures to reinstall on my “production” hardware. I had to understand disk configuration (software RAID-1) and create conditional configuration steps for production with a related production answer file. With each production reinstall, more work was captured in git, and I finally had a reproducible pet in production!

Configuration Management

It was time to re-acquaint myself with Ansible, I had not used it for years! It seemed that a few major changes had happened and another major change was about to land in 10.x, so I attempted to adopt newer techniques when possible.

I designated my Desktop to be an Ansible controller with brew install ansible ansible-lint – it was easy to reference both ansible-doc and web site documentation to develop a playbook and leverage roles and collections from Ansible Galaxy.

I created a simple Ansible YAML inventory of my production server and development VMs, then worked inside the playbook to configure each. I would SSH to a server only to troubleshoot issues and then re-run the playbook from the controller. As I got more comfortable with Ansible, I relied less on creating new shell and command tasks and tried to leverage built-in and community roles, loops, and conditionals.

I created these shell functions to speed my development cycle: the following will check the syntax of a playbook (if the ${PLAYBOOK} variable isn’t defined, it defaults to playbook.yml) and then run the playbook with variables converted from the .env file.

After I populate the .env file with variable keys and values, I type lint-playbook and we’ve started another development lifecycle test and run. I’ve previously used a file watcher to trigger runs, but I found a new system, didn’t get it working, and didn’t need it enough to justify further effort because !! is simple enough to re-run the previous command!

$ cat .env

function dotenv2keypair() {
  local _delimiter=' '
  local _hold
  local _key_value_pair

  while read -r _key_value_pair; do
  done < .env

  echo "${_hold:0:-1}"

alias lint-playbook='ansible-lint ${PLAYBOOK:-playbook.yml}
  && ansible-playbook $_ --extra-vars "$(dotenv2keypair)"'

$ head playbook.yml

- name: "Web host: monolithic"
  hosts: dev
  remote_user: "{{ _remote_user }}"

    _destination_filespec: "{{ _remote_home }}/{{ _remote_user }}/"

  - name: "Install package(s)"
    become: true
      name: bash, git, jq, rsync, tree
      # Bootstrap: assume community repo enabled during alpine/post.sh, unsure of dependency
      state: present
      # update_cache: yes
      # state: latest

There are three community Alpine specific Ansible modules, I used apk and discovered many Alpine tested Galaxy roles. Like all communities, each collection and role varies in quality, scope, assumptions, and maintenance. I also contributed to the AlpineLinux wiki a little bit!

VM Artifacts

Packer is another fabulous HashiCorp open-source DevOps tool I had not used in years! I mentioned Vagrant earlier, also by HashiCorp, as an experiment. When coupled with the open source QEMU project, Packer can build customized VM images.

After finding an Alpine wiki guide, some other Packer QEMU OS examples, and following the QEMU Builder documentation, I made a Packer qemu.json file which performs:

  1. Pull in enviroment variables
  2. Use the QEMU builder:
    1. Boot the AlpineLinux ISO
    2. Provide a web server for file transfer
    3. Perform the bootstrapping steps to leverage my init.sh installation script
  3. Provision: Wait for reboot and use SSH to perform my post.sh post-deployment script
  4. Post-process:
    1. Create a new VM image artifact in QCOW2 format
    2. Compress, checksum, create an artifact manifest, upload to artifact repository or upload to AHV Image Service in Prism Central.

Writing JSON from scratch, even when cutting and pasting in examples, is easy to make mistakes: a good JSON lint package for my IDE was a continual life-saver to keep valid syntax. Here is a snapshot of the current file, I’ve nearly refactored out all of the hard-coded values:

  "variables": {
    "boot_key_interval": "{{env `boot_key_interval`}}",
    "boot_wait": "{{env `boot_wait`}}",
    "disk_size": "{{env `disk_size`}}",
    "format": "{{env `format`}}",
    "iso_url": "{{env `iso_url`}}",
    "iso_checksum": "{{env `iso_checksum`}}",
    "remote_user": "{{env `remote_user`}}",
    "remote_password": "{{env `remote_password`}}",
    "ssh_timeout": "{{env `ssh_timeout`}}",
    "vm_name": "{{env `vm_name`}}"

  "builders": [
      "type": "qemu",
      "accelerator": "kvm",
      "format": "{{user `format`}}",
      "http_directory": "../",
      "output_directory": "packer_cache/bin",
      "iso_url": "{{user `iso_url`}}",
      "iso_checksum": "{{user `iso_checksum`}}",
      "vm_name": "{{user `vm_name`}}.{{user `format`}}",
      "disk_interface": "virtio",
      "disk_size": "{{user `disk_size`}}",
      "net_device": "virtio-net",
      "ssh_timeout": "{{user `ssh_timeout`}}",
      "ssh_username": "{{user `remote_user`}}",
      "ssh_password": "{{user `remote_password`}}",
      "ssh_certificate_file": "~/.ssh/id_rsa.pem",
      "shutdown_command": "echo 'packer' | sudo -S poweroff",
      "boot_wait": "{{user `boot_wait`}}",
      "boot_key_interval": "{{user `boot_key_interval`}}",
      "boot_command": [
        "{{user `remote_user`}}<enter><wait>",
        "passwd<enter><wait1>{{user `remote_password`}}<enter>{{user `remote_password`}}<enter>",
        "KEYMAPOPTS='us us' setup-alpine -q ; hostname {{user `vm_name`}}<enter>",
        "wget http://{{ .HTTPIP }}:{{ .HTTPPort }}/id_rsa.pub -O /etc/id_rsa.pub<enter>",
        "wget http://{{ .HTTPIP }}:{{ .HTTPPort }}/answerfile/{{user `vm_name`}}.toml<enter><wait>",
        "wget http://{{ .HTTPIP }}:{{ .HTTPPort }}/bootstrap.sh -O - | sh<enter>"

  "provisioners": [
      "type": "shell",
      "script": "../post.sh"

  "post-processors": [
      "type": "manifest"
      "type": "shell-local",
      "script": "post-processor.sh"

Now I have reproducible, lightly customized VM images ready for Configuration Management and further development. Cloud-init will be a welcome replacement for the Packer+QEMU work (it has entered into the Alpine community repo since I started this project), but the entire exercise has refreshed my knowledge with these great tools, leverage my previous work, and form a flexible foundation for moving forward. Another plus is that it was trivial to switch to a new release of Alpine Linux (3.13.0), use the smaller and faster 43 MB Virtual image ISO (rather than the 525MB 3.12.3 extended ISO I started with), and my post-process custom image compressed to 63MB! I plan to use that QCOW2 image with Nutanix AHV and convert it to VMDK for VirtualBox development.


My holiday project has spilled over into all of January, but I’ve enjoyed it immensely, sometimes waking up to work on it or not able to fall asleep without sorting out my next experiment for a feature or blocker! I’ve used it to upskill my shell scripting and programming knowledge, refresh my Ansible and Packer skills, push VirtualBox and QEMU forward, drive towards QCOW2 and AHV, all while reaffirming my love of Linux with Alpine.

I hope you’ve seen some of my development best practices and strategies, I try to set small goals that achieve strategic wins. Because this was a personal project, I lavished a lot of time on yak shaving which helped me learn and improve my skills and procedures, while refactoring and reorganizing my work to reduce tech debt and toil. I made short experiments that were dead-ends or that were too challenging for me to be productive, so I would research an alternative and try a quick experiment with it for a win. When I found the alternative lacking, it justified turning back to rededicate time to pursue a discarded path. I tried to be agile, not loyal, to any facility.

  • It is worth investing time on making your development cycle (change, build, test, report) easy and fast:

    • Use file format Linting inside your editor, because it:
      • cuts down on mistakes as I learned configuration syntax for AlpineLinux, Ansible, and Packer and my shell scripts for different shells
      • prevents finding mistakes later during build and test, every context switch out of the editor costs you time and focus and fixing mistakes later is more resource costly
      • can be even better to use specific language (Ansible, Packer, Vagrant) editor linters, but YAML, TOML, and JSON are fundamental substitutes you should already have to fall back on or seek to establish for any new facility configuration!
    • Try to make as simple as possible, a build and test change procedure:
      • Ideal for a 1-click or a simple CLI command:
        • make the command less characters by using a shell alias, function, or script to cut down on human error and parameters
        • even more ideal to setup a file watcher or IDE plugin to build and test automatically (0-click)
        • many of these facilities can be re-used later in CI/CD build and testing
        • make the procedure environment variable driven with DirEnv or .env file reuse to further cut down on human error
      • Always use facility linting as a second check before execution:
        • work to reduce and eventually remove any exceptions and warnings
    • Optimize the entire cycle to be minimal on resources, therefore faster to synthesize an ephemeral instance or provision a test
      • the smaller the time (and distance) between your change and a build+test result, the better = agility and #fastfeedback is important for developer flow
    • All of the above supports red-green-refactor, a development cycle pattern to keep your work in an optimal working state!
  • Make your work reproducable:

    • Script your PetOps, save your work in git, share your work from git!
    • Virtualization is cheap: I used two providers over the course of this project to work on virtual machine guests for development, make mistakes and learn on them: #failfastfixfast!
  • Don’t over-rotate on custom work for a single facility if you can avoid it:

    • Sprint to establish a beach head with pet ops and hard coded work
      • Relentlessly re-factor your work to use variables and scripted procedures
      • Refactor scripts to facilities
    • Each stage helps you understand the full system and reorganize information appropriately while reducing tech debt
      • I loved that my post.sh was reused by Packer, that my vm-init.sh arrived at the same place as Packer using HTTP for basic file transfer without adding more facilities to the OS,
      • I tried to leverage environment variables in all facilities (Docker, Vagrant, Ansible, shell scripts) to remove hard-coded values and increase re-use across facilities.

Finally, you haven’t seen the services I’m working on, that’ll be for the future. There is so much bootstrapping, research, and discovery involved with each that I’m still evaluating alternatives based upon support and documentation, when I’m not helping improve the docs and filing bugs as well!

Postscript: Patterns (TODO)

  • Development pattern: Red -> Green -> Refactor (repeat)
  • My research and development process over time was Alpine -> Ansible -> VirtualBox -> Packer -> QEMU, but my work process is focused on either:
    • Alpine -> Packer -> QEMU -> VirtualBox and Artifact pipeline
    • VirtualBox -> Ansible for container service development

2022-06-07 Update

  • I’ve continually deepened my AlpineLinux understanding since starting this project, which has been a boost to related embedded Linux experiments for firewall, phone, tablet, and IoT controller for home automation.
  • Converging my Markdown notes with my code in a private git repository has been a great boost in productivity because my hobby work is ad-hoc. Without Git and Markdown, I would forget where I left off due to months of lapses between work sessions to create reproducable procedures, code, and artifacts. I manage Markdown documentation interchangeably by leveraging Atom.io, Typora.io, and since the beginning of 2022, with Logseq.com (more on this later).
  • I explored the early Alpine cloud-init and VirtualBox integrations, but those experiments stopped when I didn’t see enough progress. I’ve remained focused on Packer to build golden VM images for my needs.
    • Alpine has deprecated sudo in release 3.16.0. Alpine has improved setup-alpine scripts and I’ve had to make conditional adjustments to Packer and my supporting initial boot and first reboot scripts.
    • I’ve contributed documentation improvements to the opendoas project.
    • I’ve gotten my hosting provider to add recent Alpine images and I am working to rebuild my web host on it, refactoring from my [/build_LAMP_stack_with_Phing/] custom built webplatform LAMP stack.
    • Making .env documented and reproducable (prompting for secrets) has been an interesting exercise, I’m leveraging two partial solutions now: createnv and