Author: admin yep

In the seventh part of this tutorial series on developing PHP on Docker we will setup a CI
(Continuous Integration) pipeline to run code quality tools and tests on Github Actions and Gitlab
Pipelines.

All code samples are publicly available in my
Docker PHP Tutorial repository on github.
You find the branch for this tutorial at
part-7-ci-pipeline-docker-php-gitlab-github.

Published parts of the Docker PHP Tutorial

If you want to follow along, please subscribe to the RSS feed
or via email
to get automatic notifications when the next part comes out 🙂

Table of contents

Introduction

CI is short for Continuous Integration and to me mostly means running the code quality
tools and tests of a codebase in an isolated environment (preferably automatically). This is
particularly important when working in a team, because the CI system acts as the final
gatekeeper before features or bugfixes are merged into the main branch.

I initially learned about CI systems when I stubbed my toes into the open source water. Back in the
day I used Travis CI for my own projects and replaced it
with Github Actions at some point. At ABOUT YOU we started
out with a self-hosted Jenkins server and then moved on to
Gitlab CI as a fully
managed solution (though we use custom runners).

Recommended reading

This tutorial builds on top of the previous parts. I’ll do my best to cross-reference the
corresponding articles when necessary, but I would still recommend to do some upfront reading on:

And as a nice-to-know:

Approach

In this tutorial I’m going to explain how to make our existing docker setup work with Github Actions
and Gitlab CI/CD Pipelines. As I’m a big fan of a
“progressive enhancement” approach, we will ensure that all necessary steps can be performed
locally through make. This has the additional benefit of keeping a single source of truth (the
Makefile) which will come in handy when we set up the CI system on two different providers
(Github and Gitlab).

The general process will look very similar to the one for local development:

• build the docker setup
• start the docker setup
• run the qa tools
• run the tests

You can see the final results in the CI setup section, including the concrete yml
files and links to the repositories, see

On a code level, we will treat CI as an environment, configured through the env variable ENV. So
far we only used ENV=local and we will extend that to also use ENV=ci. The necessary changes
are explained after the concrete CI setup instructions in the sections

Try it yourself

To get a feeling for what’s going on, you can start by
executing the local CI run:

This should give you a similar output as presented in the Execution example.

git checkout part-7-ci-pipeline-docker-php-gitlab-github

# Initialize make
make make-init

# Execute the local CI run
bash .local-ci.sh

CI setup

General CI notes

Initialize make for CI

As a very first step we need to “configure” the codebase to operate for the ci environment.
This is done through the make-init target as explained later in more detail in the
Makefile changes section via

make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=12345678"

$ make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=12345678"
Created a local .make/.env file

ENV=ci ensures that we

TAG=latest is just a simplification for now because we don’t do anything with the images yet.
In an upcoming tutorial we will push them to a container registry for later usage in production
deployments and then set the TAG to something more meaningful (like the build number).

EXECUTE_IN_CONTAINER=true forces every make command that uses a
RUN_IN_*_CONTAINER setup
to run in a container. This is important, because the Gitlab runner will actually run in a
docker container itself. However, this would cause any affected target to omit the
$(DOCKER_COMPOSER) exec prefix.

GPG_PASSWORD=12345678 is the password for the secret gpg key as mentioned in
Add a password-protected secret gpg key.

wait-for-service.sh

I’ll explain the “container is up and running but the underlying service is not” problem
for the mysql service and how we can solve it with a health check later in this article at
Adding a health check for mysql.
On purpose, we don’t want docker compose to take care of the waiting because we can make
“better use of the waiting time” and will instead implement it ourselves with a simple bash
script located at .docker/scripts/wait-for-service.sh:

#!/bin/bash

name=$1
max=$2
interval=$3

[ -z "$1" ] && echo "Usage example: bash wait-for-service.sh mysql 5 1"
[ -z "$2" ] && max=30
[ -z "$3" ] && interval=1

echo "Waiting for service '$name' to become healthy, checking every $interval second(s) for max. $max times"

while true; do 
  ((i++))
  echo "[$i/$max] ..."; 
  status=$(docker inspect --format "" "$(docker ps --filter name="$name" -q)")
  if echo "$status" | grep -q '"healthy"'; then 
   echo "SUCCESS";
   break
  fi
  if [ $i == $max ]; then 
    echo "FAIL"; 
    exit 1
  fi 
  sleep $interval; 
done

This script waits for a docker $service to become “healthy” by
checking the .State.Health.Status info
of the docker inspect command.

CAUTION: The script uses $(docker ps --filter name="$name" -q) to determine the id of the
container, i.e. it will “match” all running containers against the $name – this would fail if
there is more than one matching container! I.e. you must ensure that $name is specific
enough to identify one single container uniquely.

The script will check up to $max times
in a interval of $interval seconds. See these
answers on the
“How do I write a retry logic in script to keep retrying to run it up to 5 times?” question for
the implementation of the retry logic. To check the health of the mysql service for 5
times with 1 seconds between each try, it can be called via

bash wait-for-service.sh mysql 5 1

Output

$ bash wait-for-service.sh mysql 5 1
Waiting for service 'mysql' to become healthy, checking every 1 second(s) for max. 5 times
[1/5] ...
[2/5] ...
[3/5] ...
[4/5] ...
[5/5] ...
FAIL

# OR

$ bash wait-for-service.sh mysql 5 1
Waiting for service 'mysql' to become healthy, checking every 1 second(s) for max. 5 times
[1/5] ...
[2/5] ...
SUCCESS

The problem of “container dependencies” isn’t new and there are already some existing solutions
out there, e.g.

But unfortunately all of them operate by checking the availability of a host:port combination
and in the case of mysql that didn’t help, because the container was up, the port was reachable
but the mysql service in the container was not.

Setup for a “local” CI run

As mentioned under Approach, we want to be able to perform all necessary steps
locally and I created a corresponding script at .local-ci.sh:

#!/bin/bash
# fail on any error 
# @see https://stackoverflow.com/a/3474556/413531
set -e

make docker-down ENV=ci || true

start_total=$(date +%s)

# STORE GPG KEY
cp secret-protected.gpg.example secret.gpg

# DEBUG
docker version
docker compose version
cat /etc/*-release || true

# SETUP DOCKER
make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=12345678"
start_docker_build=$(date +%s)
make docker-build
end_docker_build=$(date +%s)
mkdir -p .build && chmod 777 .build

# START DOCKER
start_docker_up=$(date +%s)
make docker-up
end_docker_up=$(date +%s)
make gpg-init
make secret-decrypt-with-password

# QA
start_qa=$(date +%s)
make qa || FAILED=true
end_qa=$(date +%s)

# WAIT FOR CONTAINERS
start_wait_for_containers=$(date +%s)
bash .docker/scripts/wait-for-service.sh mysql 30 1
end_wait_for_containers=$(date +%s)

# TEST
start_test=$(date +%s)
make test || FAILED=true
end_test=$(date +%s)

end_total=$(date +%s)

# RUNTIMES
echo "Build docker:        " `expr $end_docker_build - $start_docker_build`
echo "Start docker:        " `expr $end_docker_up - $start_docker_up  `
echo "QA:                  " `expr $end_qa - $start_qa`
echo "Wait for containers: " `expr $end_wait_for_containers - $start_wait_for_containers`
echo "Tests:               " `expr $end_test - $start_test`
echo "---------------------"
echo "Total:               " `expr $end_total - $start_total`

# CLEANUP
# reset the default make variables
make make-init
make docker-down ENV=ci || true

# EVALUATE RESULTS
if [ "$FAILED" == "true" ]; then echo "FAILED"; exit 1; fi

echo "SUCCESS"

Run details

• as a preparation step, we first ensure that no outdated ci containers are running (this is
  only necessary locally, because runners on a remote CI system will start “from scratch”)

  make docker-down ENV=ci || true

• we take some time measurements to understand how long certain parts take via

  start_total=$(date +%s)

  to store the current timestamp

• we need the secret gpg key in order to decrypt the secrets and simply copy the
  password-protected example key
  (in the actual CI systems the key will be configured as a secret value that is injected in
  the run)

  # STORE GPG KEY
  cp secret-protected.gpg.example secret.gpg

• I like printing some debugging info in order to understand which exact circumstances
  we’re dealing with (tbh, this is mostly relevant when setting the CI system up or making
  modifications to it)

  # DEBUG
  docker version
  docker compose version
  cat /etc/*-release || true

• for the docker setup, we start with
  initializing the environment for ci

  # SETUP DOCKER
  make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=12345678"

  then build the docker setup

  make docker-build

  and finally add a .build/ directory to
  collect the build artifacts

  mkdir -p .build && chmod 777 .build

• then, the docker setup is started

  # START DOCKER
  make docker-up

  and gpg is initialized so that
  the secrets can be decrypted

  make gpg-init
  make secret-decrypt-with-password

  We don’t need to pass a GPG_PASSWORD to secret-decrypt-with-password because we have set
  it up in the previous step as a default value via make-init

• once the application container is running, the qa tools are run by invoking the
  qa make target

  # QA
  make qa || FAILED=true

  The || FAILED=true part makes sure that the script will not be terminated if the checks fail.
  Instead, the fact that a failure happened is “recorded” in the FAILED variable so that we
  can evaluate it at the end. We don’t want the script to stop here because we want the
  following steps to be executed as well (e.g. the tests).

• to mitigate the
  “mysql is not ready” problem, we will now apply the
  wait-for-service.sh script

  # WAIT FOR CONTAINERS
  bash .docker/scripts/wait-for-service.sh mysql 30 1

• once mysql is ready, we can execute the tests via the
  test make target and
  apply the same || FAILED=true workaround as for the qa tools

  # TEST
  make test || FAILED=true

• finally, all the timers are printed

  # RUNTIMES
  echo "Build docker:        " `expr $end_docker_build - $start_docker_build`
  echo "Start docker:        " `expr $end_docker_up - $start_docker_up  `
  echo "QA:                  " `expr $end_qa - $start_qa`
  echo "Wait for containers: " `expr $end_wait_for_containers - $start_wait_for_containers`
  echo "Tests:               " `expr $end_test - $start_test`
  echo "---------------------"
  echo "Total:               " `expr $end_total - $start_total`

• we clean up the resources (this is only necessary when running locally, because the runner of
  a CI system would be shut down anyways)

  # CLEANUP
  make make-init
  make docker-down ENV=ci || true

• and finally evaluate if any error occurred when running the qa tools or the tests

  # EVALUATE RESULTS
  if [ "$FAILED" == "true" ]; then echo "FAILED"; exit 1; fi
  
  echo "SUCCESS"

Execution example

Executing the script via

bash .local-ci.sh

yields the following (shortened) output:

$ bash .local-ci.sh
Container dofroscra_ci-redis-1  Stopping
# Stopping all other `ci` containers ...
# ...

Client:
 Cloud integration: v1.0.22
 Version:           20.10.13
# Print more debugging info ...
# ...

Created a local .make/.env file
ENV=ci TAG=latest DOCKER_REGISTRY=docker.io DOCKER_NAMESPACE=dofroscra APP_USER_NAME=application APP_GROUP_NAME=application docker compose -p dofroscra_ci --env-file ./.docker/.env -f ./.docker/docker-compose/docker-compose-php-base.yml build php-base
#1 [internal] load build definition from Dockerfile
# Output from building the docker containers 
# ...

ENV=ci TAG=latest DOCKER_REGISTRY=docker.io DOCKER_NAMESPACE=dofroscra APP_USER_NAME=application APP_GROUP_NAME=application docker compose -p dofroscra_ci --env-file ./.docker/.env -f ./.docker/docker-compose/docker-compose.local.ci.yml -f ./.docker/docker-compose/docker-compose.ci.yml up -d
Network dofroscra_ci_network  Creating
# Starting all `ci` containers ...
# ...

"C:/Program Files/Git/mingw64/bin/make" -s gpg-import GPG_KEY_FILES="secret.gpg"
gpg: directory '/home/application/.gnupg' created
gpg: keybox '/home/application/.gnupg/pubring.kbx' created
gpg: /home/application/.gnupg/trustdb.gpg: trustdb created
gpg: key D7A860BBB91B60C7: public key "Alice Doe protected <[email protected]>" imported
# Output of importing the secret and public gpg keys
# ...

"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="reveal -f -p 12345678"
git-secret: done. 1 of 1 files are revealed.
"C:/Program Files/Git/mingw64/bin/make" -j 8 -k --no-print-directory --output-sync=target qa-exec NO_PROGRESS=true
phplint                             done   took 4s
phpcs                               done   took 4s
phpstan                             done   took 8s
composer-require-checker            done   took 8s
Waiting for service 'mysql' to become healthy, checking every 1 second(s) for max. 30 times
[1/30] ...
SUCCESS
PHPUnit 9.5.19 #StandWithUkraine

........                                                            8 / 8 (100%)

Time: 00:03.077, Memory: 28.00 MB

OK (8 tests, 15 assertions)
Build docker:         12
Start docker:         2
QA:                   9
Wait for containers:  3
Tests:                5
---------------------
Total:                46
Created a local .make/.env file

Container dofroscra_ci-application-1  Stopping
Container dofroscra_ci-mysql-1  Stopping
# Stopping all other `ci` containers ...
# ...

SUCCESS

Setup for Github Actions

If you are completely new to Github Actions, I recommend to start with the
official Quickstart Guide for GitHub Actions
and the
Understanding GitHub Actions
article. In short:

• Github Actions are based on so called Workflows
  • Workflows are yaml files that live in the special .github/workflows directory in the
    repository
• a Workflow can contain multiple Jobs
• each Job consists of a series of Steps
• each Step needs a run: element that represents a command that is executed by a new shell

The Workflow file

Github Actions are triggered automatically based on the files in the .github/workflows directory.
I have added the file .github/workflows/ci.yml with the following content:

name: CI build and test

on:
  # automatically run for pull request and for pushes to branch "part-7-ci-pipeline-docker-php-gitlab-github"
  # @see https://stackoverflow.com/a/58142412/413531
  push:
    branches:
      - part-7-ci-pipeline-docker-php-gitlab-github
  pull_request: {}
  # enable to trigger the action manually
  # @see https://github.blog/changelog/2020-07-06-github-actions-manual-triggers-with-workflow_dispatch/
  # CAUTION: there is a known bug that makes the "button to trigger the run" not show up
  # @see https://github.community/t/workflow-dispatch-workflow-not-showing-in-actions-tab/130088/29
  workflow_dispatch: {}

jobs:
  build:

    runs-on: ubuntu-latest

    steps:
      - uses: actions/[email protected]

      - name: start timer
        run: |
          echo "START_TOTAL=$(date +%s)" > $GITHUB_ENV

      - name: STORE GPG KEY
        run: |
          # Note: make sure to wrap the secret in double quotes (")
          echo "$" > ./secret.gpg

      - name: SETUP TOOLS
        run : |
          DOCKER_CONFIG=${DOCKER_CONFIG:-$HOME/.docker}
          # install docker compose
          # @see https://docs.docker.com/compose/cli-command/#install-on-linux
          # @see https://github.com/docker/compose/issues/8630#issuecomment-1073166114
          mkdir -p $DOCKER_CONFIG/cli-plugins 
          curl -sSL https://github.com/docker/compose/releases/download/v2.2.3/docker-compose-linux-$(uname -m) -o $DOCKER_CONFIG/cli-plugins/docker-compose
          chmod +x $DOCKER_CONFIG/cli-plugins/docker-compose

      - name: DEBUG
        run: |
          docker compose version
          docker --version
          cat /etc/*-release

      - name: SETUP DOCKER
        run: |
          make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=$"
          make docker-build
          mkdir .build && chmod 777 .build

      - name: START DOCKER
        run: |
          make docker-up
          make gpg-init
          make secret-decrypt-with-password

      - name: QA
        run: |
          # Run the tests and qa tools but only store the error instead of failing immediately
          # @see https://stackoverflow.com/a/59200738/413531
          make qa || echo "FAILED=qa" >> $GITHUB_ENV

      - name: WAIT FOR CONTAINERS
        run: |
          # We need to wait until mysql is available.
          bash .docker/scripts/wait-for-service.sh mysql 30 1 

      - name: TEST
        run: |
          make test || echo "FAILED=test $FAILED" >> $GITHUB_ENV

      - name: RUNTIMES
        run: |
          echo `expr $(date +%s) - $START_TOTAL`

      - name: EVALUATE
        run: |
          # Check if $FAILED is NOT empty
          if [ ! -z "$FAILED" ]; then echo "Failed at $FAILED" && exit 1; fi

      - name: upload build artifacts
        uses: actions/[email protected]
        with:
          name: build-artifacts
          path: ./.build

The steps are essentially the same as explained before at
Run details for the local run. Some additional notes:

• I want the Action to be triggered automatically only when I
  push to branch part-7-ci-pipeline-docker-php-gitlab-github
  OR when a pull request is created (via pull_request). In addition, I want to be able to
  trigger the Action manually on any branch
  (via workflow_dispatch).

  on:
  push:
    branches:
      - part-7-ci-pipeline-docker-php-gitlab-github
  pull_request: {}
  workflow_dispatch: {}

  For a real project, I would let the action only run automatically on long-living branches like
  main or develop. The manual trigger is helpful if you just want to test your current work
  without putting it up for review. CAUTION: There is a
  known issue that “hides” the “Trigger workflow” button to trigger the action manually.

• a new shell is started for each run: instruction, thus we must store our timer in the “global”
  environment variable $GITHUB_ENV

    - name: start timer
    run: |
      echo "START_TOTAL=$(date +%s)" > $GITHUB_ENV 

  This will be the only timer we use, because the job uses multiple steps that are timed
  automatically – so we don’t need to take timestamps manually:
  

• the gpg key is configured as an
  encrypted secret named
  GPG_KEY and is stored in ./secret.gpg. The value is the content of the
  secret-protected.gpg.example file

    - name: STORE GPG KEY
      run: |
        echo "$" > ./secret.gpg

  Secrets are configured in the Github repository under Settings > Secrets > Actions at

  https://github.com/$user/$repository/settings/secrets/actions
  
  e.g.
  
  https://github.com/paslandau/docker-php-tutorial/settings/secrets/actions

  

• the ubuntu-latest image doesn’t contain the docker compose plugin, thus we need to
  install it manually

    - name: SETUP TOOLS
    run : |
      DOCKER_CONFIG=${DOCKER_CONFIG:-$HOME/.docker}
      mkdir -p $DOCKER_CONFIG/cli-plugins 
      curl -sSL https://github.com/docker/compose/releases/download/v2.2.3/docker-compose-linux-$(uname -m) -o $DOCKER_CONFIG/cli-plugins/docker-compose
      chmod +x $DOCKER_CONFIG/cli-plugins/docker-compose

• for the make initialization we need the second secret named GPG_PASSWORD – which is
  configured as 12345678 in our case, see
  Add a password-protected secret gpg key

    - name: SETUP DOCKER
      run: |
        make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=$"

• because the runner will be shutdown after the run, we need to move the build artifacts to a
  permanent location, using the
  actions/[email protected] action

    - name: upload build artifacts
      uses: actions/[email protected]
      with:
        name: build-artifacts
        path: ./.build

  You can
  download the artifacts in the Run overview UI
  

Setup for Gitlab Pipelines

If you are completely new to Gitlab Pipelines, I recommend to start with the
official Get started with GitLab CI/CD guide. In
short:

• the core concept of Gitlab Pipelines is the Pipeline
  • it is defined in the yaml file .gitlab-ci.yml that lives in the root of the repository
• a Pipeline can contain multiple Stages
• each Stage consists of a series of Jobs
• each Job contains a script section
• the script section consists of a series of shell commands

The .gitlab-ci.yml pipeline file

Gitlab Pipelines are triggered automatically based on a .gitlab-ci.yml file located at the
root of the repository. It has the following content:

stages:
  - build_test

QA and Tests:
  stage: build_test

  rules:
    # automatically run for pull request and for pushes to branch "part-7-ci-pipeline-docker-php-gitlab-github"
    - if: '($CI_PIPELINE_SOURCE == "merge_request_event" || $CI_COMMIT_BRANCH == "part-7-ci-pipeline-docker-php-gitlab-github")'

  # see https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#use-docker-in-docker
  image: docker:20.10.12

  services:
    - name: docker:dind

  script:
    - start_total=$(date +%s)

    ## STORE GPG KEY
    - cp $GPG_KEY_FILE ./secret.gpg

    ## SETUP TOOLS
    - start_install_tools=$(date +%s)
    # "curl" is required to download docker compose
    - apk add --no-cache make bash curl
    # install docker compose
    # @see https://docs.docker.com/compose/cli-command/#install-on-linux
    - mkdir -p ~/.docker/cli-plugins/
    - curl -sSL https://github.com/docker/compose/releases/download/v2.2.3/docker-compose-linux-x86_64 -o ~/.docker/cli-plugins/docker-compose
    - chmod +x ~/.docker/cli-plugins/docker-compose
    - end_install_tools=$(date +%s)

    ## DEBUG
    - docker version
    - docker compose version
    # show linux distro info
    - cat /etc/*-release

    ## SETUP DOCKER
    # Pass default values to the make-init command - otherwise we would have to pass those as arguments to every make call
    - make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=$GPG_PASSWORD"
    - start_docker_build=$(date +%s)
    - make docker-build
    - end_docker_build=$(date +%s)
    - mkdir .build && chmod 777 .build

    ## START DOCKER
    - start_docker_up=$(date +%s)
    - make docker-up
    - end_docker_up=$(date +%s)
    - make gpg-init
    - make secret-decrypt-with-password

    ## QA
    # Run the tests and qa tools but only store the error instead of failing immediately
    # @see https://stackoverflow.com/a/59200738/413531
    - start_qa=$(date +%s)
    - make qa ENV=ci || FAILED=true
    - end_qa=$(date +%s)

    ## WAIT FOR CONTAINERS
    # We need to wait until mysql is available.
    - start_wait_for_containers=$(date +%s)
    - bash .docker/scripts/wait-for-service.sh mysql 30 1
    - end_wait_for_containers=$(date +%s)

    ## TEST
    - start_test=$(date +%s)
    - make test ENV=ci || FAILED=true
    - end_test=$(date +%s)

    - end_total=$(date +%s)

    # RUNTIMES
    - echo "Tools:" `expr $end_install_tools - $start_install_tools`
    - echo "Build docker:" `expr $end_docker_build - $start_docker_build`
    - echo "Start docker:" `expr $end_docker_up - $start_docker_up  `
    - echo "QA:" `expr $end_qa - $start_qa`
    - echo "Wait for containers:" `expr $end_wait_for_containers - $start_wait_for_containers`
    - echo "Tests:" `expr $end_test - $start_test`
    - echo "Total:" `expr $end_total - $start_total`

    # EVALUATE RESULTS
    # Use if-else constructs in Gitlab pipelines
    # @see https://stackoverflow.com/a/55464100/413531
    - if [ "$FAILED" == "true" ]; then exit 1; fi

  # Save the build artifact, e.g. the JUNIT report.xml file, so we can download it later
  # @see https://docs.gitlab.com/ee/ci/pipelines/job_artifacts.html
  artifacts:
    when: always
    paths:
      # the quotes are required
      # @see https://stackoverflow.com/questions/38009869/how-to-specify-wildcard-artifacts-subdirectories-in-gitlab-ci-yml#comment101411265_38055730
      - ".build/*"
    expire_in: 1 week

The steps are essentially the same as explained before under
Run details for the local run. Some additional notes:

• we start by defining the stages of the pipeline – though that’s currently just one (build_test)

  stages:
  - build_test

• then we define the job QA and Tests and assign it to the build_test stage

  QA and Tests:
  stage: build_test

• I want the Pipeline to be triggered automatically only when I
  push to branch part-7-ci-pipeline-docker-php-gitlab-github
  OR when a pull request is created
  Triggering the Pipeline manually on any branch is possible by default.

  rules:
  - if: '($CI_PIPELINE_SOURCE == "merge_request_event" || $CI_COMMIT_BRANCH == "part-7-ci-pipeline-docker-php-gitlab-github")'

• since we want to build and run docker images, we need to use a docker base image and activate the
  docker:dind service. See Use Docker to build Docker images: Use Docker-in-Docker

  image: docker:20.10.12
  
  services:
  - name: docker:dind

• we store the secret gpg key as a secret file (using the
  “file” type) in the
  CI/CD variables configuration of the Gitlab repository
  and move it to ./secret.gpg in order to decrypt the secrets later

  ## STORE GPG KEY
  - cp $GPG_KEY_FILE ./secret.gpg

  Secrets can be configured under Settings > CI/CD > Variables at

  https://gitlab.com/$project/$repository/-/settings/ci_cd
  
  e.g.
  
  https://gitlab.com/docker-php-tutorial/docker-php-tutorial/-/settings/ci_cd

  

• the docker base image doesn’t come with all required tools, thus we need to install the
  missing ones (make, bash, curl and docker compose)

    ## SETUP TOOLS
    - apk add --no-cache make bash curl
    - mkdir -p ~/.docker/cli-plugins/
    - curl -sSL https://github.com/docker/compose/releases/download/v2.2.3/docker-compose-linux-x86_64 -o ~/.docker/cli-plugins/docker-compose
    - chmod +x ~/.docker/cli-plugins/docker-compose

• for the initialization of make we use the $GPG_PASSWORD variable that we defined in the
  CI/CD settings

  ## SETUP DOCKER
  - make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=$GPG_PASSWORD"

  Note: I have marked the variable as “masked”
  so it won’t show up in any logs

• finally, we store the job artifacts

  artifacts:
  when: always
  paths:
    - ".build/*"
  expire_in: 1 week 

  They can be accessed in the Pipeline overview UI
  

Performance

Performance isn’t an issue right now, because the CI runs take only about ~1 min (Github Actions)
and ~2 min (Gitlab Pipelines), but that’s mostly because we only ship a super minimal
application and those times will go up when things get more complex. For the local setup I
used all 8 cores of my laptop. The time breakdown is roughly as follows:

Step	Gitlab	Github	local without cache	local with cached images	local with cached images + layers
SETUP TOOLS	1	0	0	0	0
SETUP DOCKER	33	17	39	39	5
START DOCKER	17	11	34	2	1
QA	17	5	10	13	1
WAIT FOR CONTAINERS	5	5	3	2	13
TESTS	3	1	3	6	3
total (excl. runner startup)	78	43	97	70	36
total (incl. runner startup)	139	54	97	70	36

Times taken from

Optimizing the performance is out of scope for this tutorial, but I’ll at least document my
current findings.

The caching problem on CI

A good chunk of time is usually spent on building the docker images. We did our best to optimize
the process by leveraging the layer cache and using cache mounts
(see section Build stage ci in the php-base image).
But those steps are futile on CI systems, because the corresponding runners will start “from
scratch” for every CI run – i.e. there is no local cache that they could use. In
consequence, the full docker setup is also built “from scratch” on every run.

There are ways to mitigate that e.g.

But: None of that worked for me out-of-the-box 🙁 We will take a closer look in an upcoming
tutorial. Some reading material that I found valuable so far:

Docker changes

As a first step we need to decide which containers are required and how to provide the
codebase.

Since our goal is running the qa tools and tests, we only need the application php container. The
tests also need a database and a queue, i.e. the mysql and redis containers are required as
well – whereas nginx, php-fpm and php-worker are not required. We’ll handle that through
dedicated docker compose configuration files that only contain the necessary services. This is
explained in more detail in section Compose file updates.

In our local setup, we have shared the codebase between the host system and docker – mainly
because we wanted our changes to be reflected immediately in docker. This isn’t necessary for the
CI use case. In fact we want our CI images as close as possible to our production images – and
those should “contain everything to run independently”. I.e. the codebase should live in the
image – not on the host system. This will be explained in section
Use the whole codebase as build context.

Compose file updates

We will not only have some differences between the CI docker setup and the local docker setup
(=different containers), but also in the configuration of the individual services. To accommodate
for that, we will use the following docker compose config files in the
.docker/docker-compose/ directory:

• docker-compose.local.ci.yml:
  • holds configuration that is valid for local and ci, trying to keep the config files
    DRY
• docker-compose.ci.yml:
  • holds configuration that is only valid for ci
• docker-compose.local.yml:
  • holds configuration that is only valid for local

When using docker compose we then need to make sure to include only the required files, e.g. for
ci:

docker compose -f docker-compose.local.ci.yml -f docker-compose.ci.yml

I’ll explain the logic for that later in
section ENV based docker compose config. In short:

When comparing ci with local, for ci

• we don’t need to share the codebase with the host system

  application:
    volumes:
    - ${APP_CODE_PATH_HOST?}:${APP_CODE_PATH_CONTAINER?}

• we don’t need persistent volumes for the redis and mysql data

  mysql:
    volumes:
      - mysql:/var/lib/mysql
  
  redis:
    volumes:
      - redis:/data

• we don’t need to share ports with the host system

  application:
    ports:
      - "${APPLICATION_SSH_HOST_PORT:-2222}:22"
  
  redis:
    ports:
      - "${REDIS_HOST_PORT:-6379}:6379"

• we don’t need any settings for local dev tools like xdebug or strace

  application:
    environment:
      - PHP_IDE_CONFIG=${PHP_IDE_CONFIG?}
    cap_add:
      - "SYS_PTRACE"
    security_opt:
      - "seccomp=unconfined"
    extra_hosts:
      - host.docker.internal:host-gateway  

So all of those config values will only live in the docker-compose.local.yml file. In fact, there
are only two things that ci needs that local doesn’t:

• a volume mount to share only the secret gpg key from the host with the application container

  application:
    volumes:
      - ${APP_CODE_PATH_HOST?}/secret.gpg:${APP_CODE_PATH_CONTAINER?}/secret.gpg:ro

  This
  is required to decrypt the secrets:

  […] the private key has to be named secret.gpg and put in the root of the codebase,
  so that the import can be be simplified with make targets

• a volume mount to share a .build folder for build artifacts with the application container

  application:
    volumes:
      - ${APP_CODE_PATH_HOST?}/.build:${APP_CODE_PATH_CONTAINER?}/.build

  This will be used to collect any files we want to retain from a build (e.g. code coverage
  information, log files, etc.)

Adding a health check for mysql

When running the tests for the first time on a CI system, I noticed some weird errors related to the
database:

1) Tests\Feature\App\Http\Controllers\HomeControllerTest::test___invoke with data set "default" (array(), '    <li><a href="?dispatch=fo...></li>')
PDOException: SQLSTATE[HY000] [2002] Connection refused

As it turned out, the mysql container itself was up and running – but the mysql process
within the container was not yet ready to accept connections. Locally, this hasn’t been a problem,
because we usually would not run the tests “immediately” after starting the containers – but on CI
this is the case.

Fortunately, docker compose has us covered here and provides a
healtcheck configuration option:

healthcheck declares a check that’s run to determine whether or not containers for this service are “healthy”.

Since this healthcheck is also “valid” for local, I defined it in the combined
docker-compose.local.ci.yml file:

  mysql:
    healthcheck:
      # Only mark the service as healthy if mysql is ready to accept connections
      # Check every 2 seconds for 30 times, each check has a timeout of 1s
      test: mysqladmin ping -h 127.0.0.1 -u $$MYSQL_USER --password=$$MYSQL_PASSWORD
      timeout: 1s
      retries: 30
      interval: 2s

The script in test was taken
from SO: Docker-compose check if mysql connection is ready.

When starting the docker setup, docker ps will now add a health info to the STATUS:

$ make docker-up

$ docker ps
CONTAINER ID   IMAGE                            STATUS                           NAMES
b509eb2f99c0   dofroscra/application-ci:latest  Up 1 seconds                     dofroscra_ci-application-1
503e52fd9e68   mysql:8.0.28                     Up 1 seconds (health: starting)  dofroscra_ci-mysql-1

# a couple of seconds later

$ docker ps
CONTAINER ID   IMAGE                            STATUS                   NAMES
b509eb2f99c0   dofroscra/application-ci:latest  Up 13 seconds            dofroscra_ci-application-1
503e52fd9e68   mysql:8.0.28                     Up 13 seconds (healthy)  dofroscra_ci-mysql-1

Note the (health: starting) and (healthy) infos for the mysql service.

We can also get this info from docker inspect (used by our
wait-for-service.sh script) via:

$ docker inspect --format "" dofroscra_ci-mysql-1
"healthy"

FYI: We could also use the
depends_on property with a
condition: service_healthy on the application container so that docker compose would
only start the container once the mysql service is healthy:

application:
  depends_on:
    mysql: 
      condition: service_healthy

However, this would “block” the make docker-up until mysql is actually up and running. In
our case this is not desirable, because we can do “other stuff” in the meantime (namely: run the
qa checks, because they don’t require a database) and thus save a couple of seconds on each CI
run.

Build target: ci

We’ve already introduced build targets in
Environments and build targets
and how to “choose”
them through make with the ENV variable defined in a shared .make/.env file.
Short recap:

• in the Dockerfile of a service, define the ENV as a build stage. E.g. in
  .docker/images/php/base/Dockerfile

  FROM base as ci
  # ...

So to enable the new ci environment, we need to modify the Dockerfiles for the php-base and
the application image.

Build stage ci in the php-base image

Use the whole codebase as build context

As mentioned in section Docker changes we want to “bake” the codebase into
the ci image of the php-base container. Thus, we must change the context property in
.docker/docker-compose/docker-compose-php-base.yml to not only use the .docker/ directory
but instead the whole codebase. I.e. “dont use ../ but ../../“:

# File: .docker/docker-compose/docker-compose-php-base.yml

  php-base:
    build:
      # pass the full codebase to docker for building the image
      context: ../../

Build the dependencies

The composer dependencies must be set up in the image as well, so we introduce a new stage
stage in .docker/images/php/base/Dockerfile. The most trivial solution would look like this:

• copy the whole codebase
• run composer install

FROM base as ci

COPY . /codebase

RUN composer install --no-scripts --no-plugins --no-progress -o

However, this approach has some downsides:

• if any file in the codebase changes, the COPY . /codebase layer will be invalidated. I.e.
  docker could not use the layer cache
  which also means that every layer afterwards cannot use the cache as well. In consequence the
  composer install would run every time – even when the composer.json file doesn’t change.
• composer itself uses a cache for
  storing dependencies locally so it doesn’t have to download dependencies that haven’t changed.
  But since we run composer install in Docker, this cache would be “thrown away” every time
  a build finishes. To mitigate that, we can use
  --mount=type=cache
  to define a directory that docker will re-use between builds:

  Contents of the cache directories persists between builder invocations without invalidating
  the instruction cache.

Keeping those points in mind, we end up with the following instructions:

# File: .docker/images/php/base/Dockerfile
# ...

FROM base as ci

# By only copying the composer files required to run composer install
# the layer will be cached and only invalidated when the composer dependencies are changed
COPY ./composer.json /dependencies/
COPY ./composer.lock /dependencies/

# use a cache mount to cache the composer dependencies
# this is essentially a cache that lives in Docker BuildKit (i.e. has nothing to do with the host system) 
RUN --mount=type=cache,target=/tmp/.composer \
    cd /dependencies && \
    # COMPOSER_HOME=/tmp/.composer sets the home directory of composer that
    # also controls where composer looks for the cache 
    # so we don't have to download dependencies again (if they are cached)
    COMPOSER_HOME=/tmp/.composer composer install --no-scripts --no-plugins --no-progress -o 

# copy the full codebase
COPY . /codebase

RUN mv /dependencies/vendor /codebase/vendor && \
    cd /codebase && \
    # remove files we don't require in the image to keep the image size small
    rm -rf .docker/ && \
    # we need a git repository for git-secret to work (can be an empty one)
    git init

FYI: The COPY . /codebase step doesn’t actually copy “everything in the repository”, because we
have also introduced a .dockerignore file to exclude some files from being included in the
build context – see section .dockerignore.

Some notes on the final RUN step:

• rm -rf .docker/ doesn’t really save “that much” in the current setup – please take it more
  as an example to remove any files that shouldn’t end up in the final image (e.g. “tests in a
  production image”)
• the git init part is required because we need to decrypt the secrets later – and
  git-secret requires a git repository (which can be empty). We can’t decrypt the secrets
  during the build, because we do not want decrypted secret files to end up in the image.

When tested locally, the difference between the trivial solution and the one that makes use of
layer caching is ~35 seconds, see the results in the Performance section.

Create the final image

As a final step, we will rename the current stage to codebase and copy the “build
artifact” from that stage into our final ci build stage:

FROM base as codebase

# build the composer dependencies and clean up the copied files
# ...

FROM base as ci

COPY --from=codebase --chown=$APP_USER_NAME:$APP_GROUP_NAME /codebase $APP_CODE_PATH

Why are we not just using the previous stage directly as ci?

Because using multistage-builds
is a
good practice to keep the final layers of an image to a minimum:
Everything that “happened” in the previous codebase stage will be “forgotten”, i.e. not
exported as layers.

That does not only save us some layers, but also allows us to get rid of
files like the .docker/ directory. We needed that directory in the build context because
some files where required in other parts of the Dockerfile (e.g. the php ini files), so we
can’t exclude it via .dockerignore. But we can remove it in the codebase stage – so it will NOT
be copied over and thus not end up in the final image. If we wouldn’t have the codebase stage,
the folder would be part of the layer created when COPYing all the files from the build context
and removing it via rm -rf .docker/ would have no effect on the image size.

Currently, that doesn’t really matter, because the building step is super simple (just a
composer install) – but in a growing and more complex codebase you can easily
save a couple MB.

To be concrete, the multistage build has 31 layers and the final layer containing the
codebase has a size of 65.1MB.

$ docker image history -H dofroscra/application-ci
IMAGE          CREATED          CREATED BY                                      SIZE      COMMENT
d778c2ee8d5e   17 minutes ago   COPY /codebase /var/www/app # buildkit          65.1MB    buildkit.dockerfile.v0
                                                                                ^^^^^^
<missing>      17 minutes ago   WORKDIR /var/www/app                            0B        buildkit.dockerfile.v0
<missing>      17 minutes ago   COPY /usr/bin/composer /usr/local/bin/compos…   2.36MB    buildkit.dockerfile.v0
<missing>      17 minutes ago   COPY ./.docker/images/php/base/.bashrc /root…   395B      buildkit.dockerfile.v0
<missing>      17 minutes ago   COPY ./.docker/images/php/base/.bashrc /home…   395B      buildkit.dockerfile.v0
<missing>      17 minutes ago   COPY ./.docker/images/php/base/conf.d/zz-app…   196B      buildkit.dockerfile.v0
<missing>      17 minutes ago   COPY ./.docker/images/php/base/conf.d/zz-app…   378B      buildkit.dockerfile.v0
<missing>      17 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   1.28kB    buildkit.dockerfile.v0
<missing>      17 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   41MB      buildkit.dockerfile.v0
<missing>      18 minutes ago   ADD https://php.hernandev.com/key/php-alpine…   451B      buildkit.dockerfile.v0
<missing>      18 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   62.1MB    buildkit.dockerfile.v0
<missing>      18 minutes ago   ADD https://gitsecret.jfrog.io/artifactory/a…   450B      buildkit.dockerfile.v0
<missing>      18 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   4.74kB    buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV ENV=ci                                      0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV ALPINE_VERSION=3.15                         0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV TARGET_PHP_VERSION=8.1                      0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV APP_CODE_PATH=/var/www/app                  0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV APP_GROUP_NAME=application                  0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV APP_USER_NAME=application                   0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV APP_GROUP_ID=10001                          0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ENV APP_USER_ID=10000                           0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG ENV                                         0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG ALPINE_VERSION                              0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG TARGET_PHP_VERSION                          0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG APP_CODE_PATH                               0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG APP_GROUP_NAME                              0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG APP_USER_NAME                               0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG APP_GROUP_ID                                0B        buildkit.dockerfile.v0
<missing>      18 minutes ago   ARG APP_USER_ID                                 0B        buildkit.dockerfile.v0
<missing>      2 days ago       /bin/sh -c #(nop)  CMD ["/bin/sh"]              0B
<missing>      2 days ago       /bin/sh -c #(nop) ADD file:5d673d25da3a14ce1…   5.57MB

The non-multistage build has 32 layers and the final layer(s) containing the
codebase have a combined size of 65.15MB (60.3MB + 4.85MB).

$ docker image history -H dofroscra/application-ci
IMAGE          CREATED          CREATED BY                                      SIZE      COMMENT
94ba50438c9a   2 minutes ago    RUN /bin/sh -c COMPOSER_HOME=/tmp/.composer …   60.3MB    buildkit.dockerfile.v0
<missing>      2 minutes ago    COPY . /var/www/app # buildkit                  4.85MB    buildkit.dockerfile.v0
                                                                                ^^^^^^
<missing>      31 minutes ago   WORKDIR /var/www/app                            0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   COPY /usr/bin/composer /usr/local/bin/compos…   2.36MB    buildkit.dockerfile.v0
<missing>      31 minutes ago   COPY ./.docker/images/php/base/.bashrc /root…   395B      buildkit.dockerfile.v0
<missing>      31 minutes ago   COPY ./.docker/images/php/base/.bashrc /home…   395B      buildkit.dockerfile.v0
<missing>      31 minutes ago   COPY ./.docker/images/php/base/conf.d/zz-app…   196B      buildkit.dockerfile.v0
<missing>      31 minutes ago   COPY ./.docker/images/php/base/conf.d/zz-app…   378B      buildkit.dockerfile.v0
<missing>      31 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   1.28kB    buildkit.dockerfile.v0
<missing>      31 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   41MB      buildkit.dockerfile.v0
<missing>      31 minutes ago   ADD https://php.hernandev.com/key/php-alpine…   451B      buildkit.dockerfile.v0
<missing>      31 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   62.1MB    buildkit.dockerfile.v0
<missing>      31 minutes ago   ADD https://gitsecret.jfrog.io/artifactory/a…   450B      buildkit.dockerfile.v0
<missing>      31 minutes ago   RUN |8 APP_USER_ID=10000 APP_GROUP_ID=10001 …   4.74kB    buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV ENV=ci                                      0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV ALPINE_VERSION=3.15                         0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV TARGET_PHP_VERSION=8.1                      0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV APP_CODE_PATH=/var/www/app                  0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV APP_GROUP_NAME=application                  0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV APP_USER_NAME=application                   0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV APP_GROUP_ID=10001                          0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ENV APP_USER_ID=10000                           0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG ENV                                         0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG ALPINE_VERSION                              0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG TARGET_PHP_VERSION                          0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG APP_CODE_PATH                               0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG APP_GROUP_NAME                              0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG APP_USER_NAME                               0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG APP_GROUP_ID                                0B        buildkit.dockerfile.v0
<missing>      31 minutes ago   ARG APP_USER_ID                                 0B        buildkit.dockerfile.v0
<missing>      2 days ago       /bin/sh -c #(nop)  CMD ["/bin/sh"]              0B
<missing>      2 days ago       /bin/sh -c #(nop) ADD file:5d673d25da3a14ce1…   5.57MB

Again: It is expected that the differences aren’t big, because the only size savings come from
the .docker/ directory with a size of ~70kb.

$ du -hd 0 .docker
73K     .docker

Finally, we are also using the --chown option of the RUN instruction
to ensure that the files have the correct permissions.

Build stage ci in the application image

There is actually “nothing” to be done here. We don’t need SSH any longer because it is only
required for the SSH Configuration of PhpStorm.
So the build stage is simply “empty”:

ARG BASE_IMAGE
FROM ${BASE_IMAGE} as base

FROM base as ci

FROM base as local
# ...

Though there is one thing to keep in mind: In the local image we used sshd as the entrypoint,
i.e. we had a long running process that would keep the container running. To keep the
ci application container running, we must

.dockerignore

The .dockerignore file
is located in the root of the repository and ensures that certain files are kept out of the
Docker build context. This will

• speed up the build (because less files need to be transmitted to the docker daemon)
• keep images smaller (because irrelevant files are kept out of the image)

The syntax is quite similar to the .gitignore file – in fact I’ve found it to be quite often
the case that the contents of the .gitignore file are a subset of the .dockerignore file. This
makes kinda sense, because you typically wouldn’t want files that are excluded from the
repository to end up in a docker image (e.g. unencrypted secret files). This has also been
noticed by others, see e.g.

but to my knowledge there is currently (2022-04-24) no way to “keep the two files in sync”.

In our case, the content of the .dockerignore file looks like this:

# gitignore
!.env.example
.env
.idea
.phpunit.result.cache
vendor/
secret.gpg
.gitsecret/keys/random_seed
.gitsecret/keys/pubring.kbx~
!*.secret
passwords.txt
.build

# additionally ignored files
.git

Makefile changes

Initialize the shared variables

We have introduced the concept of shared variables via .make/.env
previously. It allows us to define variables in one place (=single source
of truth) that are then used as “defaults” so we don’t have to define them explicitly when
invoking certain make targets (like make docker-build). We’ll make use of this concept by
setting the environment to civiaENV=ci and thus making sure that all docker commands use
ci “automatically” as well.

In addition, I made a small modification by introducing a second file at .make/variables.env
that is also included in the main Makefile and holds the “default” shared variables. Those
are neither “secret” nor are they likely to be be changed for environment adjustments. The file
is NOT ignored by .gitignore and is basically just the previous .make/.env.example file without
the environment specific variables:

# File .make/variables.env

DOCKER_REGISTRY=docker.io
DOCKER_NAMESPACE=dofroscra
APP_USER_NAME=application
APP_GROUP_NAME=application

The .make/.env file is still .gitignored and can be initialized with the make-init
target using the ENVS variable:

make make-init ENVS="ENV=ci SOME_OTHER_DEFAULT_VARIABLE=foo"

which would create a .make/.env file with the content

ENV=ci
SOME_OTHER_DEFAULT_VARIABLE=foo

If necessary, we could also override variables defined in the .make/variables.env file,
because the .make/.env is included last in the Makefile:

# File: Makefile
# ...

# include the default variables
include .make/variables.env
# include the local variables
-include .make/.env

The default value for ENVS is ENV=local TAG=latest to retain the same default behavior as
before when ENVS is omitted. The corresponding make-init target is defined in the main
Makefile and now looks like this:

ENVS?=ENV=local TAG=latest
.PHONY: make-init
make-init: ## Initializes the local .makefile/.env file with ENV variables for make. Use via ENVS="KEY_1=value1 KEY_2=value2"
    @$(if $(ENVS),,$(error ENVS is undefined))
    @rm  -f .make/.env
    for variable in $(ENVS); do \
      echo $$variable | tee -a .make/.env > /dev/null 2>&1; \
    done
    @echo  "Created a local .make/.env file" 

ENV based docker compose config

As mentioned in section Compose file updates we need to select the
“correct” docker compose configuration files based on the ENV value. This is done in
.make/02-00-docker.mk:

# File .make/02-00-docker.mk

# ...

DOCKER_COMPOSE_DIR:=...
DOCKER_COMPOSE_COMMAND:=...

DOCKER_COMPOSE_FILE_LOCAL_CI:=$(DOCKER_COMPOSE_DIR)/docker-compose.local.ci.yml
DOCKER_COMPOSE_FILE_CI:=$(DOCKER_COMPOSE_DIR)/docker-compose.ci.yml
DOCKER_COMPOSE_FILE_LOCAL:=$(DOCKER_COMPOSE_DIR)/docker-compose.local.yml

# we need to "assemble" the correct combination of docker-compose.yml config files
ifeq ($(ENV),ci)
    DOCKER_COMPOSE_FILES:=-f $(DOCKER_COMPOSE_FILE_LOCAL_CI) -f $(DOCKER_COMPOSE_FILE_CI)
else ifeq ($(ENV),local)
    DOCKER_COMPOSE_FILES:=-f $(DOCKER_COMPOSE_FILE_LOCAL_CI) -f $(DOCKER_COMPOSE_FILE_LOCAL)
endif

DOCKER_COMPOSE:=$(DOCKER_COMPOSE_COMMAND) $(DOCKER_COMPOSE_FILES)

When we now take a look at a full recipe when using ENV=ci with a docker target (e.g.
docker-up), we can see that the correct files are chosen, e.g.

$ make docker-up ENV=ci -n
ENV=ci TAG=latest DOCKER_REGISTRY=docker.io DOCKER_NAMESPACE=dofroscra APP_USER_NAME=application APP_GROUP_NAME=application docker compose -p dofroscra_ci --env-file ./.docker/.env -f ./.docker/docker-compose/docker-compose.local.ci.yml -f ./.docker/docker-compose/docker-compose.ci.yml up -d

# =>
# -f ./.docker/docker-compose/docker-compose.local.ci.yml 
# -f ./.docker/docker-compose/docker-compose.ci.yml

Codebase changes

Add a test for encrypted files

We’ve introduced git-secret in the previous tutorial
Use git-secret to encrypt secrets in the repository
and used it to store the file passwords.txt encrypted in the codebase. To make sure that the
decryption works as expected on the CI systems, I’ve added a test at
tests/Feature/EncryptionTest.php to check if the file exists and if the content is correct.

class EncryptionTest extends TestCase
{
    public function test_ensure_that_the_secret_passwords_file_was_decrypted()
    {
        $pathToSecretFile = __DIR__."/../../passwords.txt";

        $this->assertFileExists($pathToSecretFile);

        $expected = "my_secret_password\n";
        $actual   = file_get_contents($pathToSecretFile);

        $this->assertEquals($expected, $actual);
    }
}

Of course this doesn’t make sense in a “real world scenario”, because the secret value would now
be exposed in a test – but it suffices for now as proof of a working secret decryption.

Add a password-protected secret gpg key

I’ve mentioned in
Scenario: Decrypt file
that it is also possible to use a password-protected secret gpg key for
an additional layer of security. I have created such a key and stored it in the repository at
secret-protected.gpg.example (in a “real world scenario” I wouldn’t do that – but since this
is a public tutorial I want you to be able to follow along completely). The password for that
key is 12345678.

The corresponding public key is located at .dev/gpg-keys/alice-protected-public.gpg and
belongs to the email address [email protected]. I’ve
added this address and
re-encrypted the secrets afterwards via

make gpg-init
make secret-add-user EMAIL="[email protected]"
make secret-encrypt

When I now import the secret-protected.gpg.example key, I can decrypt the secrets, though I
cannot use the usual secret-decrypt target but must instead use secret-decrypt-with-password

make secret-decrypt-with-password GPG_PASSWORD=12345678

or store the GPG_PASSWORD in the .make/.env file when it is initialized for CI

make make-init ENVS="ENV=ci TAG=latest EXECUTE_IN_CONTAINER=true GPG_PASSWORD=12345678"
make secret-decrypt-with-password

Create a JUnit report from PhpUnit

I’ve added the
--log-junit option
to the phpunit configuration of the test make target in order to create an XML report in the
.build/ directory in the .make/01-02-application-qa.mk file:

# File: .make/01-02-application-qa.mk
# ...

PHPUNIT_CMD=php vendor/bin/phpunit
PHPUNIT_ARGS= -c phpunit.xml --log-junit .build/report.xml

I.e. each run of the tests will now create a
Junit XML report at
.build/report.xml. The file is used as an example of a build artifact, i.e.
“something that we would like to keep” from a CI run.

Wrapping up

Congratulations, you made it! If some things are not completely clear by now, don’t hesitate to
leave a comment. You should now have a working CI pipeline for Github (via Github Actions)
and/or Gitlab (via Gitlab pipelines) that runs automatically on each push.

In the next part of this tutorial, we will use terraform to create an infrastructure for
production deployments on GCP and deploy the docker containers there.

Please subscribe to the RSS feed or via email to get automatic
notifications when this next part comes out 🙂

---

Wanna stay in touch?

Since you ended up on this blog, chances are pretty high that you’re into Software Development
(probably PHP, Laravel, Docker or Google Big Query) and I’m a big fan of feedback and networking.

So – if you’d like to stay in touch, feel free to shoot me an email with a couple of words about yourself and/or
connect with me on
LinkedIn or
Twitter
or simply subscribe to my RSS feed
or go the crazy route and subscribe via mail
and don’t forget to leave a comment 🙂

Comments

In the sixth part of this tutorial series on developing PHP on Docker we will setup git-secret
to store secrets directly in the repository. Everything will be handled through Docker and
added as make targets for a convenient workflow.

FYI:
This tutorial is a precursor to the next a part
Create a CI pipeline for dockerized PHP Apps
because dealing with secrets is an important aspect when setting up a CI system (and later when
deploying to production) – but I feel it’s complex enough to warrant its own article.

All code samples are publicly available in my
Docker PHP Tutorial repository on github.
You find the branch with the final result of this tutorial at
part-6-git-secret-encrypt-repository-docker.

Published parts of the Docker PHP Tutorial

If you want to follow along, please subscribe to the RSS feed
or via email
to get automatic notifications when the next part comes out 🙂

Table of contents

Introduction

Dealing with secrets (passwords, tokens, key files, etc.) is close to “naming things”
when it comes to hard problems in software engineering. Some things to consider:

• security is paramount – but high security often goes hand in hand with high inconvenience
  • and if things get too complicated, people look for shortcuts…
• in a team, sharing certain secret values is often mandatory
  • so now we need to think about secure ways to distribute and update secrets across multiple
    people
• concrete secret values often depend on the environment
  • inherently tricky to “test” or even “review”, because those values are “by definition”
    different on “your machine” than on “production”

In fact, entire products have been build around dealing with secrets, e.g.
HashiCorp Vault,
AWS Secrets Manager or the
GCP Secret Manager. Introducing those in a project comes
with a certain overhead as it’s yet another service that needs to be integrated and
maintained. Maybe it is the exactly right decision for your use-case – maybe it’s overkill.
By the end of this article you’ll at least be aware of an alternative with a lower barrier to entry.
See also the Pros and cons section in the end for an overview.

Even though it’s
generally not advised to store secrets in a repository,
I’ll propose exactly that in this tutorial:

• identify files that contain secret values
• make sure they are added to .gitignore
• encrypt them via git-secret
• commit the encrypted files to the repository

In the end, we will be able to call

make secret-decrypt

to reveal secrets in the codebase, make modifications to them if necessary and then run

make secret-encrypt

to encrypt them again so that they can be committed (and pushed to the remote repository). To
see it in action, check out branch
part-6-git-secret-encrypt-repository-docker
and run the following commands:

# checkout the branch
git checkout part-6-git-secret-encrypt-repository-docker

# build and start the docker setup
make make-init
make docker-build
make docker-up

# "create" the secret key - the file "secret.gpg.example" would usually NOT live in the repo!
cp secret.gpg.example secret.gpg

# initialize gpg
make gpg-init

# ensure that the decrypted secret file does not exist
ls passwords.txt

# decrypt the secret file
make secret-decrypt

# show the content of the secret file
cat passwords.txt

Tooling

We will set up gpg and git-secret in the php base image, so that the tools become available in
all other containers. Please refer to
Docker from scratch for PHP 8.1 Applications in 2022
for an in-depth explanation of the docker images.

Please note, that there is a caveat when using git-secret in a folder that is shared between
the host system and a docker container. I’ll explain that in more detail (including a workaround)
in section
The git-secret directory and the gpg-agent socket.

gpg

gpg is short for The GNU Privacy Guard and is an open source implementation
of the OpenPGP standard. In short, it allows us to create a personal key file pair
(similar to SSH keys) with a private secret key and a public
key that can be shared with other parties whose messages you want to decrypt.

gpg installation

To install it, we can simply run apk add gnupg and thus update
.docker/images/php/base/Dockerfile accordingly

# File: .docker/images/php/base/Dockerfile

RUN apk add --update --no-cache \
        bash \
        gnupg \
        make \
#...

gpg usage

I’ll only cover the strictly necessary gpg commands here. Please refer to
the “Using GPG” section in the git-secret docu
and/or How to generate PGP keys with GPG
for further information.

Create GPG key pair

We need gpg to create the gpg key pair via

name="Pascal Landau"
email="[email protected]"
gpg --batch --gen-key <<EOF
Key-Type: 1
Key-Length: 2048
Subkey-Type: 1
Subkey-Length: 2048
Name-Real: $name
Name-Email: $email
Expire-Date: 0
%no-protection
EOF

The %no-protection will create a key without password, see
also this gist to “Creating gpg keys non-interactively”.

Output:

$ name="Pascal Landau"
$ email="[email protected]"
$ gpg --batch --gen-key <<EOF
> Key-Type: 1
> Key-Length: 2048
> Subkey-Type: 1
> Subkey-Length: 2048
> Name-Real: $name
> Name-Email: $email
> Expire-Date: 0
> %no-protection
> EOF
gpg: key E1E734E00B611C26 marked as ultimately trusted
gpg: revocation certificate stored as '/root/.gnupg/opengpg-revocs.d/74082D81525723F5BF5B2099E1E734E00B611C26.rev'

You could also run gpg --gen-key without the --batch flag to be guided interactively through the
process.

Export, list and import private GPG keys

The private key can be exported via

email="[email protected]"
path="secret.gpg"
gpg --output "$path" --armor --export-secret-key "$email"

This secret key must never be shared!

It looks like this:

-----BEGIN PGP PRIVATE KEY BLOCK-----

lQOYBF7VVBwBCADo9un+SySu/InHSkPDpFVKuZXg/s4BbZmqFtYjvUUSoRAeSejv
G21nwttQGut+F+GdpDJL6W4pmLS31Kxpt6LCAxhID+PRYiJQ4k3inJfeUx7Ws339
XDPO3Rys+CmnZchcEgnbOfQlEqo51DMj6mRF2Ra/6svh7lqhrixGx1BaKn6VlHkC
...
ncIcHxNZt7eK644nWDn7j52HsRi+wcWsZ9mjkUgZLtyMPJNB5qlKQ18QgVdEAhuZ
xT3SieoBPd+tZikhu3BqyIifmLnxOJOjOIhbQrgFiblvzU1iOUOTOcSIB+7A
=YmRm
-----END PGP PRIVATE KEY BLOCK-----

All secret keys can be listed via

gpg --list-secret-keys

Output:

$ gpg --list-secret-keys
/root/.gnupg/pubring.kbx
------------------------
sec   rsa2048 2022-03-27 [SCEA]
      74082D81525723F5BF5B2099E1E734E00B611C26
uid           [ultimate] Pascal Landau <[email protected]>
ssb   rsa2048 2022-03-27 [SEA]

You can import the private key via

path="secret.gpg"
gpg --import "$path"

and get the following output:

$ path="secret.gpg"
$ gpg --import "$path"
gpg: key E1E734E00B611C26: "Pascal Landau <[email protected]>" not changed
gpg: key E1E734E00B611C26: secret key imported
gpg: Total number processed: 1
gpg:              unchanged: 1
gpg:       secret keys read: 1
gpg:  secret keys unchanged: 1

Caution: If the secret key requires a password, you would now be prompted for it. We can
circumvent the prompt by using --batch --yes --pinentry-mode loopback:

path="secret.gpg"
gpg --import --batch --yes --pinentry-mode loopback "$path"

See also Using Command-Line Passphrase Input for GPG.
In doing so, we don’t need to provide the password just yet – but we must pass it later when we
attempt to decrypt files.

Export, list and import public GPG keys

The public key can be exported to public.gpg via

email="[email protected]"
path="public.gpg"
gpg --armor --export "$email" > "$path"

It looks like this:

-----BEGIN PGP PUBLIC KEY BLOCK-----

mQENBF7VVBwBCADo9un+SySu/InHSkPDpFVKuZXg/s4BbZmqFtYjvUUSoRAeSejv
G21nwttQGut+F+GdpDJL6W4pmLS31Kxpt6LCAxhID+PRYiJQ4k3inJfeUx7Ws339
...
3LLbK7Qxz0cV12K7B+n2ei466QAYXo03a7WlsPWn0JTFCsHoCOphjaVsncIcHxNZ
t7eK644nWDn7j52HsRi+wcWsZ9mjkUgZLtyMPJNB5qlKQ18QgVdEAhuZxT3SieoB
Pd+tZikhu3BqyIifmLnxOJOjOIhbQrgFiblvzU1iOUOTOcSIB+7A
=g0hF
-----END PGP PUBLIC KEY BLOCK-----

List all public keys via

gpg --list-keys

Output:

$ gpg --list-keys
/root/.gnupg/pubring.kbx
------------------------
pub   rsa2048 2022-03-27 [SCEA]
      74082D81525723F5BF5B2099E1E734E00B611C26
uid           [ultimate] Pascal Landau <[email protected]>
sub   rsa2048 2022-03-27 [SEA]

The public key can be imported in the same way as private keys via

path="public.gpg"
gpg --import "$path"

Example:

$ gpg --import /var/www/app/public.gpg
gpg: key E1E734E00B611C26: "Pascal Landau <[email protected]>" not changed
gpg: Total number processed: 1
gpg:              unchanged: 1

git-secret

The official website of git-secret is already doing a great job of
introducing the tool. In short, it allows us to declare certain files as “secrets” and encrypt
them via gpg – using the keys of all trusted parties. The encrypted file can then by stored
safely directly in the git repository and decrypted if required.

In this tutorial I’m using git-secret v0.4.0

$ git secret --version
0.4.0

git-secret installation

The installation instructions for Alpine read as
follows:

sh -c "echo 'https://gitsecret.jfrog.io/artifactory/git-secret-apk/all/main'" >> /etc/apk/repositories
wget -O /etc/apk/keys/git-secret-apk.rsa.pub 'https://gitsecret.jfrog.io/artifactory/api/security/keypair/public/repositories/git-secret-apk'
apk add --update --no-cache git-secret

We update the .docker/images/php/base/Dockerfile accordingly:

# File: .docker/images/php/base/Dockerfile

# install git-secret
# @see https://git-secret.io/installation#alpine
ADD https://gitsecret.jfrog.io/artifactory/api/security/keypair/public/repositories/git-secret-apk /etc/apk/keys/git-secret-apk.rsa.pub

RUN echo "https://gitsecret.jfrog.io/artifactory/git-secret-apk/all/main" >> /etc/apk/repositories  && \
    apk add --update --no-cache \
        bash \
        git-secret \
        gnupg \
        make \
#...

git-secret usage

Initialize git-secret

git-secret is initialized via the following command run in the root of the git repository

git secret init

$ git secret init
git-secret: init created: '/var/www/app/.gitsecret/'

We only need to do this once, because we’ll commit the folder to git later. It contains the
following files:

$ git status | grep ".gitsecret"
        new file:   .gitsecret/keys/pubring.kbx
        new file:   .gitsecret/keys/pubring.kbx~
        new file:   .gitsecret/keys/trustdb.gpg
        new file:   .gitsecret/paths/mapping.cfg

The pubring.kbx~ file (with the trailing tilde ~) is only a temporary file and can safely be
git-ignored. See also
Can’t find any docs about keyring.kbx~ file.

The git-secret directory and the gpg-agent socket

To use git-secret in a directory that is shared between the host system and docker, we need to
also run the following commands:

tee .gitsecret/keys/S.gpg-agent <<EOF
%Assuan%
socket=/tmp/S.gpg-agent
EOF

tee .gitsecret/keys/S.gpg-agent.ssh <<EOF
%Assuan%
socket=/tmp/S.gpg-agent.ssh
EOF

tee .gitsecret/keys/gpg-agent.conf <<EOF
extra-socket /tmp/S.gpg-agent.extra
browser-socket /tmp/S.gpg-agent.browser
EOF

This is necessary because there is an issue when git-secret is used in a setup where the
codebase is shared between the host system and a docker container.
I’ve explained the details in the Github issue
“gpg: can’t connect to the agent: IPC connect call failed” error in docker alpine on shared volume.

In short:

• gpg uses a gpg-agent to perform its tasks and the two tools communicate through sockets
  that are created in the --home-directory of the gpg-agent
• the agent is started implicitly through a gpg command used by git-secret, using the
  .gitsecret/keys directories as a --home-directory
• because the location of the --home-directory is shared with the host system, the socket
  creation fails (potentially only an issue for Docker Desktop, see the related discussion in
  Github issue Support for sharing unix sockets)

The corresponding error messages are

gpg: can't connect to the agent: IPC connect call failed

gpg-agent: error binding socket to '/var/www/app/.gitsecret/keys/S.gpg-agent': I/O error

The workaround for this problem can be found in
this thread: Configure gpg to use different
locations for the sockets by
placing additional gpg configuration files in the .gitsecret/keys directory:

S.gpg-agent

%Assuan%
socket=/tmp/S.gpg-agent

S.gpg-agent.ssh

%Assuan%
socket=/tmp/S.gpg-agent.ssh

gpg-agent.conf

extra-socket /tmp/S.gpg-agent.extra
browser-socket /tmp/S.gpg-agent.browser

Adding, listing and removing users

To add a new user, you must first import its public gpg key. Then
run:

email="[email protected]"
git secret tell "$email"

In this case, the user [email protected] will now be able to decrypt the secrets.

To show the users run

git secret whoknows

$ git secret whoknows
[email protected]

To remove a user, run

email="[email protected]"
git secret killperson "$email"

FYI: This command was renamed to removeperson in git-secret >= 0.5.0

$ git secret killperson [email protected]
git-secret: removed keys.
git-secret: now [[email protected]] do not have an access to the repository.
git-secret: make sure to hide the existing secrets again.

User [email protected] will no longer be able to decrypt the secrets.

Caution: The secrets need to be re-encrypted after removing a user!

Reminder: Rotate the encrypted secrets

Please be aware that not only your secrets are stored in git, but who had access as well. I.e.
even if you remove a user and re-encrypt the secrets, that user would still be able to decrypt
the secrets of a previous commit (when the user was still added). In consequence, you need
to rotate the encrypted secrets themselves as well after removing a user.

But isn’t that a great flaw in the system, making it a bad idea to use git-secret in general?

In my opinion: No.

If the removed user had access to the secrets at any point in time (no
matter where they have been stored), he could very well have just created a local copy or simply
“written them down”. In terms of security there is really no “added downside” due to git-secret.
It just makes it very clear that you must rotate the secrets ¯\_(ツ)_/¯

See also this
lengthy discussion on git-secret on Hacker News.

Adding, listing and removing files for encryption

Run git secret add [filenames...] for files you want to encrypt. Example:

git secret add .env

If .env is not added in .gitignore, git-secret will display a warning and add it
automatically.

git-secret: these files are not in .gitignore: .env
git-secret: auto adding them to .env
git-secret: 1 item(s) added.

Otherwise, the file is added with no warning.

$ git secret add .env
git-secret: 1 item(s) added.

You only need to add files once. They are then stored in .gitsecret/paths/mapping.cfg:

$ cat .gitsecret/paths/mapping.cfg
.env:505070fc20233cb426eac6a3414399d0f466710c993198b1088e897fdfbbb2d5

You can also show the added files via

git secret list

$ git secret list
.env

Caution: The files are not yet encrypted!

If you want to remove a file from being encrypted, run

git secret remove .env

Output

$ git secret remove .env
git-secret: removed from index.
git-secret: ensure that files: [.env] are now not ignored.

Encrypt files

To actually encrypt the files, run:

git secret hide

Output:

$ git secret hide
git-secret: done. 1 of 1 files are hidden.

The encrypted (binary) file is stored at $filename.secret, i.e. .env.secret in this case:

$ cat .env.secret
�☺♀♥�H~�B�Ӯ☺�"��▼♂F�►���l�Cs��S�@MHWs��e������{♣♫↕↓�L� ↕s�1�J$◄♥�;���dž֕�Za�����\u�ٲ& ¶��V�► ���6��
;<�d:��}ҨD%.�;��&��G����vWW�]>���߶��▲;D�+Rs�S→�Y!&J��۪8���ٔF��→f����*��$♠���&RC�8▼♂�☻z h��Z0M�T>

The encrypted files are de-cryptable for all users that have been added via git secret tell.
That also means that you need to run this command again whenever a new user is added.

Decrypting files

You can decrypt files via

git secret reveal

Output:

$ git secret reveal
File '/var/www/app/.env' exists. Overwrite? (y/N) y
git-secret: done. 1 of 1 files are revealed.

• the files are decrypted and will overwrite the current, unencrypted files (if they already exist)
  • use the -f option to force the overwrite and run non-interactively
• if you only want to check the content of an encrypted file, you can use
  git secret cat $filename (e.g. git secret cat .env)

In case the secret gpg key is password protected, you must pass the password
via the -p option. E.g. for password 123456

git secret reveal -p 123456

Show changes between encrypted and decrypted files

One problem that comes with encrypted files: You can’t review them during a code review in a
remote tool. So in order to understand what changes have been made, it is helpful to
show the changes between the encrypted and the decrypted files. This can be done via

git secret changes

Output:

$ echo "foo" >> .env
$ git secret changes
git-secret: changes in /var/www/app/.env:
--- /dev/fd/63
+++ /var/www/app/.env
@@ -34,3 +34,4 @@
 MAIL_ENCRYPTION=null
 MAIL_FROM_ADDRESS=null
 MAIL_FROM_NAME="${APP_NAME}"
+foo

Note the +foo at the bottom of the output. It was added in the first line via
echo "foo"> >> .env.

Makefile adjustments

Since I won’t be able to remember all the commands for git-secret and gpg, I’ve added them to
the Makefile at .make/01-00-application-setup.mk:

# File: .make/01-00-application-setup.mk

#...

# gpg

DEFAULT_SECRET_GPG_KEY?=secret.gpg
DEFAULT_PUBLIC_GPG_KEYS?=.dev/gpg-keys/*

.PHONY: gpg
gpg: ## Run gpg commands. Specify the command e.g. via ARGS="--list-keys"
    $(EXECUTE_IN_APPLICATION_CONTAINER) gpg $(ARGS)

.PHONY: gpg-export-public-key
gpg-export-public-key: ## Export a gpg public key e.g. via EMAIL="[email protected]" PATH=".dev/gpg-keys/john-public.gpg"
    @$(if $(PATH),,$(error PATH is undefined))
    @$(if $(EMAIL),,$(error EMAIL is undefined))
    "$(MAKE)" -s gpg ARGS="gpg --armor --export $(EMAIL) > $(PATH)"

.PHONY: gpg-export-private-key
gpg-export-private-key: ## Export a gpg private key e.g. via EMAIL="[email protected]" PATH="secret.gpg"
    @$(if $(PATH),,$(error PATH is undefined))
    @$(if $(EMAIL),,$(error EMAIL is undefined))
    "$(MAKE)" -s gpg ARGS="--output $(PATH) --armor --export-secret-key $(EMAIL)"

.PHONY: gpg-import
gpg-import: ## Import a gpg key file e.g. via GPG_KEY_FILES="/path/to/file /path/to/file2"
    @$(if $(GPG_KEY_FILES),,$(error GPG_KEY_FILES is undefined))
    "$(MAKE)" -s gpg ARGS="--import --batch --yes --pinentry-mode loopback $(GPG_KEY_FILES)"

.PHONY: gpg-import-default-secret-key
gpg-import-default-secret-key: ## Import the default secret key
    "$(MAKE)" -s gpg-import GPG_KEY_FILES="$(DEFAULT_SECRET_GPG_KEY)"

.PHONY: gpg-import-default-public-keys
gpg-import-default-public-keys: ## Import the default public keys
    "$(MAKE)" -s gpg-import GPG_KEY_FILES="$(DEFAULT_PUBLIC_GPG_KEYS)" 

.PHONY: gpg-init
gpg-init: gpg-import-default-secret-key gpg-import-default-public-keys ## Initialize gpg in the container, i.e. import all public and private keys

# git-secret

.PHONY: git-secret
git-secret: ## Run git-secret commands. Specify the command e.g. via ARGS="hide"
    $(EXECUTE_IN_APPLICATION_CONTAINER) git-secret $(ARGS)

.PHONY: secret-init
secret-init: ## Initialize git-secret in the repository via `git-secret init`
    "$(MAKE)" -s git-secret ARGS="init"

.PHONY: secret-init-gpg-socket-config
secret-init-gpg-socket-config: ## Initialize the config files to change the gpg socket locations
    echo "%Assuan%" > .gitsecret/keys/S.gpg-agent
    echo "socket=/tmp/S.gpg-agent" >> .gitsecret/keys/S.gpg-agent
    echo "%Assuan%" > .gitsecret/keys/S.gpg-agent.ssh
    echo "socket=/tmp/S.gpg-agent.ssh" >> .gitsecret/keys/S.gpg-agent.ssh
    echo "extra-socket /tmp/S.gpg-agent.extra" > .gitsecret/keys/gpg-agent.conf
    echo "browser-socket /tmp/S.gpg-agent.browser" >> .gitsecret/keys/gpg-agent.conf

.PHONY: secret-encrypt
secret-encrypt: ## Decrypt secret files via `git-secret hide`
    "$(MAKE)" -s git-secret ARGS="hide"

.PHONY: secret-decrypt
secret-decrypt: ## Decrypt secret files via `git-secret reveal -f`
    "$(MAKE)" -s git-secret ARGS="reveal -f" 

.PHONY: secret-decrypt-with-password
secret-decrypt-with-password: ## Decrypt secret files using a password for gpg via `git-secret reveal -f -p $(GPG_PASSWORD)`
    @$(if $(GPG_PASSWORD),,$(error GPG_PASSWORD is undefined))
    "$(MAKE)" -s git-secret ARGS="reveal -f -p $(GPG_PASSWORD)" 

.PHONY: secret-add
secret-add: ## Add a file to git secret via `git-secret add $FILE`
    @$(if $(FILE),,$(error FILE is undefined))
    "$(MAKE)" -s git-secret ARGS="add $(FILE)"

.PHONY: secret-cat
secret-cat: ## Show the contents of file to git secret via `git-secret cat $FILE`
    @$(if $(FILE),,$(error FILE is undefined))
    "$(MAKE)" -s git-secret ARGS="cat $(FILE)"

.PHONY: secret-list
secret-list: ## List all files added to git secret `git-secret list`
    "$(MAKE)" -s git-secret ARGS="list"

.PHONY: secret-remove
secret-remove: ## Remove a file from git secret via `git-secret remove $FILE`
    @$(if $(FILE),,$(error FILE is undefined))
    "$(MAKE)" -s git-secret ARGS="remove $(FILE)"

.PHONY: secret-add-user
secret-add-user: ## Remove a user from git secret via `git-secret tell $EMAIL`
    @$(if $(EMAIL),,$(error EMAIL is undefined))
    "$(MAKE)" -s git-secret ARGS="tell $(EMAIL)"

.PHONY: secret-show-users
secret-show-users: ## Show all users that have access to git secret via `git-secret whoknows`
    "$(MAKE)" -s git-secret ARGS="whoknows"

.PHONY: secret-remove-user
secret-remove-user: ## Remove a user from git secret via `git-secret killperson $EMAIL`
    @$(if $(EMAIL),,$(error EMAIL is undefined))
    "$(MAKE)" -s git-secret ARGS="killperson $(EMAIL)"

.PHONY: secret-diff
secret-diff: ## Show the diff between the content of encrypted and decrypted files via `git-secret changes`
    "$(MAKE)" -s git-secret ARGS="changes"

Workflow

Working with git-secret is pretty straight forward:

• initialize git-secret
• add all users
• add all secret files and make sure they are ignored via .gitignore
• encrypt the files
• commit the encrypted files like “any other file”
• if any changes were made by other team members to the files:
  • => decrypt to get the most up-to-date ones
• if any modifications are required from your side:
  • => make the changes to the decrypted files and then re-encrypt them again

But: The devil is in the details. The Process challenges section explains
some of the pitfalls that we have encountered and the Scenarios section gives some
concrete examples for common scenarios.

Process challenges

From a process perspective we’ve encountered some challenges that I’d like to mention – including
how we deal with them.

Updating secrets

When updating secrets you must ensure to always decrypt the files first in order to avoid
using “stale” files that you might still have locally. I usually check out the latest main
branch and run git secret reveal to have the most up-to-date versions of the secret files. You
could also use a post-merge git hook to do
this automatically, but I personally don’t want to risk overwriting my local secret files by
accident.

Code reviews and merge conflicts

Since the encrypted files cannot be diffed meaningfully, the code reviews become more difficult
when secrets are involved. We use Gitlab for reviews and I usually first check the diff of
the .gitsecret/paths/mapping.cfg file to see “which files have changed” directly in the UI.

In addition, I will

• checkout the main branch
• decrypt the secrets via git secret reveal -f
• checkout the feature-branch
• run git secret changes to see the differences between the decrypted files from main and the
  encrypted files from feature-branch

Things get even more complicated when multiple team members need to modify secret files at the same
time on different branches, as the encrypted files cannot be compared – i.e. git cannot be smart
about delta updates.
The only way around this is coordinating the pull requests, i.e. merge the first, update the
secrets of the second and then merge the second.

Fortunately, this has only happened very rarely so far.

Local git-secret and gpg setup

Currently, all developers in our team have git-secret installed locally (instead of using it
through docker) and use their own gpg keys.

This means more onboarding overhead, because

• a new dev must
  • install git-secret locally (*)
  • install and setup gpg locally (*)
  • create a gpg key pair
• the public key must be added by every other team member (*)
• the user of the key must be added via git secret tell
• the secrets must be re-encrypted

And for offboarding

• the public key must be removed by every other team member (*)
• the user of the key must be removed via git secret killperson
• the secrets must be re-encrypted

Plus, we need to ensure that the git-secret and gpg versions are kept up-to-date for everyone to
not run into any compatibility issues.

As an alternative, I’m currently leaning more towards handling everything through docker (as
presented in this tutorial). All steps marked with (*) are then obsolete, i.e. there is no need
to setup git-secret and gpg locally.

But the approach also comes with some downsides, because

• the secret key and all public keys have to be imported every time the container is started
• each dev needs to put his private gpg key “in the codebase” (ignored by .gitignore) so it
  can be shared with docker and imported by gpg (in docker). The alternative would be using
  a single secret key that is shared within the team – which feels very wrong 😛

To make this a little more convenient, we put the public gpg keys of every dev in the
repository under .dev/gpg-keys/ and the private key has to be named secret.gpg and put
in the root of the codebase.

In this setup, secret.gpg must also be added to the.gitignore file.

# File: .gitignore
#...
vendor/
secret.gpg

The import can now be be simplified with make targets:

# gpg

DEFAULT_SECRET_GPG_KEY?=secret.gpg
DEFAULT_PUBLIC_GPG_KEYS?=.dev/gpg-keys/*

.PHONY: gpg
gpg: ## Run gpg commands. Specify the command e.g. via ARGS="--list-keys"
    $(EXECUTE_IN_APPLICATION_CONTAINER) gpg $(ARGS)

.PHONY: gpg-import
gpg-import: ## Import a gpg key file e.g. via GPG_KEY_FILES="/path/to/file /path/to/file2"
    @$(if $(GPG_KEY_FILES),,$(error GPG_KEY_FILES is undefined))
    "$(MAKE)" -s gpg ARGS="--import --batch --yes --pinentry-mode loopback $(GPG_KEY_FILES)"

.PHONY: gpg-import-default-secret-key
gpg-import-default-secret-key: ## Import the default secret key
    "$(MAKE)" -s gpg-import GPG_KEY_FILES="$(DEFAULT_SECRET_GPG_KEY)"

.PHONY: gpg-import-default-public-keys
gpg-import-default-public-keys: ## Import the default public keys
    "$(MAKE)" -s gpg-import GPG_KEY_FILES="$(DEFAULT_PUBLIC_GPG_KEYS)" 

.PHONY: gpg-init
gpg-init: gpg-import-default-secret-key gpg-import-default-public-keys ## Initialize gpg in the container, i.e. import all public and private keys

“Everything” can now be handled via

make gpg-init

that needs to be run one single time after a container has been started.

Scenarios

The scenarios assume the following preconditions:

• You have checked out branch part-6-git-secret-encrypt-repository-docker

  git checkout part-6-git-secret-encrypt-repository-docker

  and no running docker containers

  make docker-down

• You have deleted the existing git-secret folder, the keys in .dev/gpg-keys, the
  secret.gpg key and the passwords.* files

  rm -rf .gitsecret/ .dev/gpg-keys/* secret.gpg passwords.*

Initial setup of gpg keys

Unfortunately, I didn’t find a way to create and export gpg keys through make and docker. You
need to either run the commands interactively OR pass a string with newlines to it. Both things are
horribly complicated with make and docker. Thus, you need to log into the application
container and run the commands in there directly. Not great – but this needs to be done only
once when a new developer is onboarded anyways.

FYI: I usually log into containers via
Easy container access via din .bashrc helper.

The secret key is exported to secret.gpg and the public key to .dev/gpg-keys/alice-public.gpg.

# start the docker setup
make docker-up

# log into the container ('winpty' is only required on Windows)
winpty docker exec -ti dofroscra_local-application-1 bash

# export key pair
name="Alice Doe"
email="[email protected]"
gpg --batch --gen-key <<EOF
Key-Type: 1
Key-Length: 2048
Subkey-Type: 1
Subkey-Length: 2048
Name-Real: $name
Name-Email: $email
Expire-Date: 0
%no-protection
EOF

# export the private key
gpg --output secret.gpg --armor --export-secret-key $email

# export the public key
gpg --armor --export $email > .dev/gpg-keys/alice-public.gpg

$ make docker-up
ENV=local TAG=latest DOCKER_REGISTRY=docker.io DOCKER_NAMESPACE=dofroscra APP_USER_NAME=application APP_GROUP_NAME=application docker compose -p dofroscra_local --env-file ./.docker/.env -f ./.docker/docker-compose/docker-compose.yml -f ./.docker/docker-compose/docker-compose.local.yml up -d
Container dofroscra_local-application-1  Created
...
Container dofroscra_local-application-1  Started
$ docker ps
CONTAINER ID   IMAGE                                COMMAND                  CREATED          STATUS          PORTS                NAMES
...
95f740607586   dofroscra/application-local:latest   "/usr/sbin/sshd -D"      21 minutes ago   Up 21 minutes   0.0.0.0:2222->22/tcp dofroscra_local-application-1

$ winpty docker exec -ti dofroscra_local-application-1 bash
root:/var/www/app# name="Alice Doe"
root:/var/www/app# email="[email protected]"
gpg --batch --gen-key <<EOF
Key-Type: 1
Key-Length: 2048
Subkey-Type: 1
Subkey-Length: 2048
Name-Real: $name
Name-Email: $email
Expire-Date: 0
%no-protection
EOF
root:/var/www/app# gpg --batch --gen-key <<EOF
> Key-Type: 1
> Key-Length: 2048
> Subkey-Type: 1
> Subkey-Length: 2048
> Name-Real: $name
> Name-Email: $email
> Expire-Date: 0
> %no-protection
> EOF
gpg: directory '/root/.gnupg' created
gpg: keybox '/root/.gnupg/pubring.kbx' created
gpg: /root/.gnupg/trustdb.gpg: trustdb created
gpg: key BBBE654440E720C1 marked as ultimately trusted
gpg: directory '/root/.gnupg/openpgp-revocs.d' created
gpg: revocation certificate stored as '/root/.gnupg/openpgp-revocs.d/225C736E0E70AC222C072B70BBBE654440E720C1.rev'

root:/var/www/app# gpg --output secret.gpg --armor --export-secret-key $email
root:/var/www/app# head secret.gpg
-----BEGIN PGP PRIVATE KEY BLOCK-----
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root:/var/www/app# gpg --armor --export $email > .dev/gpg-keys/alice-public.gpg
root:/var/www/app# head .dev/gpg-keys/alice-public.gpg
-----BEGIN PGP PUBLIC KEY BLOCK-----
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That’s it. We now have a new secret and private key for [email protected] and have exported it to
secret.gpg resp. .dev/gpg-keys/alice-public.gpg (and thus shared it with the host system).
The remaining commands can now be run outside of the application container directly on the
host system.

Initial setup of git-secret

Let’s say we want to introduce git-secret “from scratch” to a new codebase. Then you would run
the following commands:

Initialize git-secret

make secret-init

$ make secret-init
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="init";
git-secret: init created: '/var/www/app/.gitsecret/'

Apply the gpg fix for shared directories

See The git-secret directory and the gpg-agent socket.

$ make secret-init-gpg-socket-config

$ make secret-init-gpg-socket-config
echo "%Assuan%" > .gitsecret/keys/S.gpg-agent
echo "socket=/tmp/S.gpg-agent" >> .gitsecret/keys/S.gpg-agent
echo "%Assuan%" > .gitsecret/keys/S.gpg-agent.ssh
echo "socket=/tmp/S.gpg-agent.ssh" >> .gitsecret/keys/S.gpg-agent.ssh
echo "extra-socket /tmp/S.gpg-agent.extra" > .gitsecret/keys/gpg-agent.conf
echo "browser-socket /tmp/S.gpg-agent.browser" >> .gitsecret/keys/gpg-agent.conf

Initialize gpg after container startup

After restarting the containers, we need to initialize gpg, i.e. import all public keys from
.dev/gpg-keys/* and the private key from secret.gpg. Otherwise we will not be able to en-
and decrypt the files.

make gpg-init

$ make gpg-init
"C:/Program Files/Git/mingw64/bin/make" -s gpg-import GPG_KEY_FILES="secret.gpg"
gpg: directory '/home/application/.gnupg' created
gpg: keybox '/home/application/.gnupg/pubring.kbx' created
gpg: /home/application/.gnupg/trustdb.gpg: trustdb created
gpg: key BBBE654440E720C1: public key "Alice Doe <[email protected]>" imported
gpg: key BBBE654440E720C1: secret key imported
gpg: Total number processed: 1
gpg:               imported: 1
gpg:       secret keys read: 1
gpg:   secret keys imported: 1
"C:/Program Files/Git/mingw64/bin/make" -s gpg-import GPG_KEY_FILES=".dev/gpg-keys/*"
gpg: key BBBE654440E720C1: "Alice Doe <[email protected]>" not changed
gpg: Total number processed: 1
gpg:              unchanged: 1

Adding (new) team members

Let’s start by adding our own user to git-secret

make secret-add-user EMAIL="[email protected]"

$ make secret-add-user EMAIL="[email protected]"
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="tell [email protected]"
git-secret: done. [email protected] added as user(s) who know the secret.

And verify that it worked via

make secret-show-users

$ make secret-show-users
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="whoknows"
[email protected]

Adding and encrypting files

Let’s add a new encrypted file secret_password.txt.

Create the file

echo "my_new_secret_password" > secret_password.txt

Add it to .gitignore

echo "secret_password.txt" >> .gitignore

Add it to git-secret

make secret-add FILE="secret_password.txt"

$ make secret-add FILE="secret_password.txt"
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="add secret_password.txt"
git-secret: 1 item(s) added.

Encrypt all files

make secret-encrypt

$ make secret-encrypt
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="hide"
git-secret: done. 1 of 1 files are hidden.

$ ls secret_password.txt.secret
secret_password.txt.secret

Decrypt files

Let’s first remove the “plain” secret_password.txt file

rm secret_password.txt

$ rm secret_password.txt

$ ls secret_password.txt
ls: cannot access 'secret_password.txt': No such file or directory

and then decrypt the encrypted one.

make secret-decrypt

$ make secret-decrypt
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="reveal -f"
git-secret: done. 1 of 1 files are revealed.

$ cat secret_password.txt
my_new_secret_password

Caution: If the secret gpg key is password protected (e.g. 123456), run

make secret-decrypt-with-password GPG_PASSWORD=123456

You could also add the GPG_PASSWORD variable to the
.make/.env
file as a local default value so that you wouldn’t have to specify the value every time and
could then simply run

make secret-decrypt-with-password

without passing GPG_PASSWORD

Removing files

Remove the secret_password.txt file we added previously:

make secret-remove FILE="secret_password.txt"

$ make secret-remove FILE="secret_password.txt"
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="remove secret_password.txt"
git-secret: removed from index.
git-secret: ensure that files: [secret_password.txt] are now not ignored.

Caution: this will neither remove the secret_password.txt file nor
the secret_password.txt.secret file automatically”

$ ls -l | grep secret_password.txt
-rw-r--r-- 1 Pascal 197121     19 Mar 31 14:03 secret_password.txt
-rw-r--r-- 1 Pascal 197121    358 Mar 31 14:02 secret_password.txt.secret

But even though the encrypted secret_password.txt.secret file still exists, it will not be
decrypted:

$ make secret-decrypt
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="reveal -f"
git-secret: done. 0 of 0 files are revealed.

Removing team members

Removing a team member can be done via

make secret-remove-user EMAIL="[email protected]"

$ make secret-remove-user EMAIL="[email protected]"
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="killperson [email protected]"
git-secret: removed keys.
git-secret: now [[email protected]] do not have an access to the repository.
git-secret: make sure to hide the existing secrets again.

If there are any users left, we must make sure to re-encrypt the secrets via

make secret-encrypt

Otherwise (if no more users are left) git-secret would simply error out

$ make secret-decrypt
"C:/Program Files/Git/mingw64/bin/make" -s git-secret ARGS="reveal -f"
git-secret: abort: no public keys for users found. run 'git secret tell [email protected]'.
make[1]: *** [.make/01-00-application-setup.mk:57: git-secret] Error 1
make: *** [.make/01-00-application-setup.mk:69: secret-decrypt] Error 2

Caution: Please keep in mind to
rotate the secrets themselves as well!

Pros and cons

Pro

• very low barrier to entry:
  • no third party service required
  • easy to integrate in existing codebases, because the secrets are located directly in
    the codebase
  • everything can be handled through docker (no additional local software necessary)
• once set up, it is very easy/convenient to use and can be integrated in a team workflow
• changes to secrets can be reviewed before they are merged
  • this leads to less fuck-ups on deployments
• “everything” is in the repository, which brings a lot of familiar benefits like
  • version control
  • a single git pull is the only thing you need to get everything (=> good dev experience)

Cons

• some overhead during onboarding and offboarding
• the secret key must be put in the root of the repository at ./secret.gpg
• no fine grained permissions for different secrets, e.g. the mysql password on production and
  staging can not be treated differently
  • if somebody can decrypt secrets, ALL of them are exposed
• if the a secret key ever gets leaked, all secrets are compromised
  • => can be mitigated (to a degree) by using a passphrase on the secret key
  • => this is kinda true for any other system that stores secrets as well BUT third parties
    could probably implement additional measures like multi factor authentication
• secrets are versioned alongside the users that have access, i.e. even if a user is removed at
  some point, he can still decrypt a previous version of the encrypted secrets

Wrapping up

Congratulations, you made it! If some things are not completely clear by now, don’t hesitate to
leave a comment. You are now able to encrypt and decrypt secret files so that they can be stored
directly in the git repository.

In the next part of this tutorial, we will
set up a CI pipeline for dockerized PHP Apps on Github and Gitlab
that decrypts all necessary secrets and then runs our tests and qa tools.

Please subscribe to the RSS feed or via email to get automatic
notifications when this next part comes out 🙂

---

Wanna stay in touch?

Since you ended up on this blog, chances are pretty high that you’re into Software Development
(probably PHP, Laravel, Docker or Google Big Query) and I’m a big fan of feedback and networking.

So – if you’d like to stay in touch, feel free to shoot me an email with a couple of words about yourself and/or
connect with me on
LinkedIn or
Twitter
or simply subscribe to my RSS feed
or go the crazy route and subscribe via mail
and don’t forget to leave a comment 🙂

Comments

Modern web frameworks teach you to take one group of related concepts and split it across multiple places throughout your codebase. Laravel is a robust framework with a big community behind it. Usually it’s standard structure is enough for most starting projects.

Building scalable applications, instead, requires a different approach. Have you ever heard from a client to work on controllers or review the models folder? Probably never – they ask you to work on invoicing, clients management or users. These concept groups are called domains.

Let’s make a practical exercise applying Domain Driven Design. Our goal is to create a boilerplate to be used universally as base of any Laravel project. Take advantage of the framework power at the same time we meet complex business requirements.

Prerequisites

Understanding of Domain Driven Design and some basic concepts:

We are going to use a fresh Laravel 9 installation for this guide, take a look on how to create your first Laravel project. To run Laravel locally a PHP setup is also required.

Keep in mind

We must keep in mind some important points planning the architecture of our software:

• Clean-code design plays a key role in building highly scalable applications.
• Follow unified business language that everyone in the company (not only developers) will understand and that will be used in our business/product development process.
• Decoupling the application from the framework can be exhausting and pointless. We want to use the power of the framework having the code as much decoupled as possible.
• Carefully choose your third-party services, otherwise, they might cause operational failure.

Architecture layers

There are several ways in which the Laravel framework can be organized to serve as a template for large-scale projects. We will focus on the app (aka src) folder while keeping the framework features almost intact .

Initially, Laravel is structure looks as below:

With modified codebase structure, we are able to follow Domain Driven Design  within our Laravel project which will support the future growth of our software. We also will be ready for the upcoming framework upgrades. We want it to be easy to upgrade to the next versions.

In first place, we should create a folder for each DDD layer:

• app/Domain
• app/Application
• app/Infrastructure
• app/UserInterface

Domain

Since this layer is where abstractions are made, the design of interfaces are included in the domain layer. It will also contain aggregates, value objects (VOs), data transfer objects (DTOs), domain events, entities, models, etc…

The only exception would be anything related to eloquent models. Eloquent makes very easy to interact with databases, tables and rows but the reality is that it’s not a DDD model. It’s an ambiguous definition of the concept of model with implementation of database connection. Does it mean that we can not use Eloquent? Yes we can, it can be used as repository implementation (infrastructure layer). We do have a significant advantage with this approach: we are no longer dependent on Laravel’s method names and we can use some naming that reflects the language of the domain.

Actually we have nothing in domain layer so we will keep it empty.

Application

Application layer provides the required base to use and manipulate the domain in a user-friendly way. It is where business process flows are handled, commands are executed and reactions to domain events are coded.

Actually we have nothing in application layer so we will keep it empty.

Infrastructure

Infrastructure layer is responsible for communication with external websites, access to services such as database (persistence), messaging systems and email services.

We are going to treat Laravel as a third-party service for our application. So all the framework files are going to be grouped inside the infrastructure folder.

What does it imply:

Note: make sure to update namespaces when moving files.

The final result look as following:

User Interface (UI)

User interface layer is the part where interaction with external world happens. The responsible of displaying information to the user and accepting new data. It could be implemented for web, console or any presentation technology.

Actually we have nothing in user interface layer so we will keep it empty.

Binding interfaces

One last thing that our architecture is lacking: the connection of concrete implementations with interfaces within our domain, e.g. repository interfaces.

For each module on the domain layer, we need a matching module in the infrastructure layer which takes responsibility for what the domain layer cannot afford to care about.

We recommend using EventServiceProvider.php to make these bindings:

Here you can define the abstract interface and the concrete implementation. It will be kind of class wiring configuration.

Bonus

As a small bonus, we’ve included shared domain VOs for basic types.

That classes provide an abstraction and shared methods for the final VO definition. An example of usage:

<?php namespace App\Domain\Post\ValueObject;

use App\Domain\Shared\ValueObject\StringValueObject;

class Message extends StringValueObject
{
}

Note: constructor, getters and additional shared methods can be included in the parent StringValueObject.

Conclusion

Note that so far nothing has changed in the way we use Laravel. We still have our Kernels, Providers, Exception Handlers, Rules, Mails and more inside the app folder.

Implementing Domain-Driven Design is always going to be a challenge no matter what framework we use, there is no unique way of defining things. Almost everything depends on the specific project you’re working on and it probably makes sense to apply a different structure or architecture in other cases.

Domain Drive Design is a continuous process that must be carried out according to specific needs that can be adapted over time. Also it’s a trade off: investing time on having a perfect structure or creating a starting base and improving with the time.

Credits

Official GitHub: https://github.com/hibit-dev/laravel9-ddd