"9" branch (Drupal 9, default branch, this branch)
- About
- Quickstart
- HTTPS quickstart
- Initial installation on Docker
- Incremental deployment (updating) on Docker
- ./scripts/docker-compose.sh instead of docker-compose
- Environment types
- Local and environment-specific settings
- Power up/power down the Docker environment
- Uninstalling the Docker environment
- Prescribed development process
- Starter data
- Private and public files
- Database schema versions and project versions
- hook_update_N() and configuration
- Patches
- Development design patterns
- Removing modules
- Running automated tests
- Security and maintenance updates
- Theming
- Acquia
- Login links for Acquia environments
- Pantheon (experimental)
- Linting
- Getting a local version of the database
- Logging
- Logging emails during development
- Troubleshooting
A starterkit to build a Drupal 8 or 9 project.
- The master branch of Drupal 8
- The 9 branch of Drupal 9
- Docker-based: only Docker is required to build a development environment.
- One-click install: a full installation should be available simply by running
./scripts/deploy.sh
. - Dependencies and generated files are not included in the repo: modules, libraries, etc. should not be included in the repo. If you need to generate a directory with all dependencies and generated files, run
./scripts/create-build.sh
. - Sass is not used: to keep the workflow simple, this project does not use SASS; we encourage direct modification of CSS. See the section "No SASS", below.
- The code lives on GitHub.
- The issue queue is on GitHub.
- If you fork of copy this directory for your own project, enter other environments here (production, stage, secondary git origins).
Step 1: Install Docker (nothing else required)
Step 2:
cd ~/Desktop && git clone https://github.com/dcycle/starterkit-drupal8site.git
cd ~/Desktop/starterkit-drupal8site && ./scripts/deploy.sh
Step 3: Click on the login link at the end of the command line output and enjoy a fully installed Drupal 8 or 9 environment (depending on the branch used -- see above).
You can SSH into your container by running:
./scripts/ssh.sh
cd ~/Desktop/starterkit-drupal8site && ./scripts/https-deploy.sh
See the article Local development using Docker and HTTPS, Dcycle Blog, Oct. 27, 2018 for details on how this works.
To install or update the code, install Docker and run ./scripts/deploy.sh
, which will give you a login link to a local development site.
Updating your local installation is the same command as the installation command:
./scripts/deploy.sh
This will bring in new features and keep your existing data.
When you run ./scripts/deploy.sh, you can specify one of these environment types. If you do not specify an environment type, the "dev" environment type is used.
dev (default): used for local development.
- Opcache is turned off: PHP parses your code at every call, insteading of using optimization cache.
- Volumes are shared between your host and your container, meaning changes to your code on your host machine are used by your container.
build: used to build an image, potentially for storage on a Docker registry.
- Opcache is turned on because "build" mode is not designed to do development.
- Code is copied to the container, so your container will have everything it needs to run.
The docker-compose
command will, by default, use docker-compose.yml
, but in our setup, docker-compose.yml
by itself is not valid, because we have several environment types (dev, build, ...). Using ./scripts/docker-compose.sh
will find the right environment type based on the contents of the unversioned .env
file. For example, if the environment type is "dev", we will use, automatically, docker-compose.yml
but also docker-compose.dev.yml
as described in ./scripts/dev/env.txt.
If you need environment-specific settings, or need to track sensitiv api keys, you can create a file named:
./drupal/settings/local-settings/unversioned.php
Use ./drupal/settings/local-settings/unversioned.example.php as an example.
To shut down your containers but keep your data for next time:
./scripts/docker-compose.sh down
To power up your containers:
./scripts/docker-compose.sh up -d
To shut down your containers and destroy your data:
./scripts/destroy.sh
If your project is the only project using the local https via the [Nginx Proxy], you might want to destroy the nginx-proxy container
docker kill nginx-proxy
docker rm nginx-proxy
You might also want to remove, from /etc/hosts, the line which contains your local development domain (use sudo vi /etc/hosts
to edit that file); and also remove the certificate for your domain from ~/.docker-compose-certs
.
The development cycle is as follows:
- Download the latest code and run
./scripts/deploy.sh
. - Perform your local development on a branch
my-feature
. - Export your config changes to code using
./scripts/export-config.sh
. - Merge the latest version of master and run
./scripts/deploy.sh
. - Push to Github and open a pull request.
- Self-review or have a colleague review code in the pull request.
- Make sure automated tests and linting is passing using
./scripts/test.sh
- Use Github's GUI to Squash and merge your branch for clean git history.
- Use Github's GUI to delete your branch.
- Delete your branch locally and switch back to master.
- Pull the latest version of master.
Developers should also read this entire ./README.md file and add to it any information which may be useful for other developers.
Our approach is to include everything necessary for development within our codebase, meaning that, if you use this project as a basis for your own projects, and follow the same approach, developers will almost never need to obtain a staging database to do work.
Here is how this works:
- In
./drupal/starter-data
, you will find a starter database and starter files such as images. - When you run
./scripts/deploy.sh
, the code will use the starter data to populate the environment (see also the section "Database schema versions and project versions", below). - Developers who create, for example, a new content type xyz and a new view at, for example, /listing, will run
./scripts/export-config.sh
to export the content type and view.
The above exports the configuration into code at ./drupal/config
, but not in the starter database. This does not matter, because any time ./scripts/deploy.sh
is run, configuration in code will be imported into the running database.
However the next developer to run ./scripts/deploy.sh
on a new environment, or a testbot, will not have any dummy (starter) data associated with this new configuration. The following steps can be added to a development workflow to remedy this:
- Developers can now create a few dummy nodes of type xyz, along with images in image fields.
- They will then run
./scripts/update-starter-data.sh
which will take this new data and make it part of the codebase. - This new data will now be available to new developers (and existing developers if they run
./scripts/destroy.sh
, then./scripts/deploy.sh
again). - This new data will be available to automated test code at
./tests/browser-tests/test01.js
(which is called by./scripts/end-to-end-tests.sh
in the continuous integration process). - You will also be able to test for accessibility of your new code, with dummy data, at
./scripts/a11y-tests.sh
(also called during the continuous integration process).
With this approach, functionality and configuration is deeply integrated with dummy data in the same codebase.
./docker-compose.yml defines two volumes, drupal-files
(the public files) and drupal-private-files
, the private files.
The public files volume maps to /var/www/html/sites/default/files
on the container; the private files volume maps to /drupal-private-files
on the container. The latter is also set in ./drupal/settings/local-settings/versioned.php.
This means that both public files and private files will persist beyond the life of the drupal container.
This project (Drupal 8/9 Starterkit) decouples module versions from the project code. Concretely, in ./Dockerfile-drupal-base, for example, we download, via composer, modules such as the latest version (not a specific version) of Webform.
In addition, because Drupal code requires a database to do anything useful, and because we want the benefits of version control on our code, we include, as stated above in the "Starter data" section, in ./drupal/starter-data/
, a minimal database and files which work with our code.
The UUID defined in ./drupal/config/system.site.yml is (and needs to be) identical to the UUID in the starter database.
Because, as mentioned, module versions are decoupled from project code, new versions of modules might implement hook_update_N() to modify the database. So concretely, if you create an instance of this project, it will use the starter database, then update it using the update hooks of the latest versions of all modules, which might be (and probably are) different from the versions used to create the starter database.
If you would like to update the starter database dump based on the current (post-hook_update_N()-)database, you can do so by running ./scripts/update-starter-data.sh (which updates the starter database and also the starter files such as images, based on what is in the running environment).
We include, in ./drupal/config, minimal configuration which works with our code. There are two ways in this config can be updated:
- If module update hooks update the configuration, the code is updated as part of the deployment process (see "hook_update_N() and configuration", below) but only for newly installed sites.
- Developers can run ./scripts/export-config-sh to update the configuration based on new features such as fields, content types, newly-installed modules...
Combined, the above approach allows you to check out any version of the starterkit code, and run the deployment script, ending up with a useful website.
When an update hook modifies the database, our process works seamlessly for developers, even if the starter database is in a different schema than what is expected from the modules: ./scripts/deploy.sh
imports the outdated started database, then updates it using drush updb
.
Except in the following case:
Drupal modules sometimes use update hooks to modify configuration. For more details see hook_update_N(), a powerful and dangerous tool to use sparingly, January 29, 2021, Dcycle blog.
Because of #3110362 If an update hook modifies configuration, then old configuration is imported, the changes made by the update hook are forever lost., and our decoupling of code from and module versions, we will, as part of our deployment script, update the configuration in code if update hooks update the configuration in the database.
Here is an example of how it works:
Let's say our starter database has been created with Webform 5.23. Its default "default_page_base_path" config is "form" without a leading slash:
drush config:get webform.settings settings.default_page_base_path
# 'webform.settings:settings.default_page_base_path': form
Running ./scripts/deploy.sh -- on a CI server or on a new installation or after having called ./scripts/destroy.sh -- will do the following: import the starter database normally, then run the ./drupal/scripts/update-config-in-code-if-updb-modifies-config-in-db.sh script (which on the container exists at /scripts/update-config-in-code-if-updb-modifies-config-in-db.sh):
- import the configuration
- run hook updb
- then export the configuration
Simply running ./scripts/deploy.sh thus has the potential to actually modify your configuration in code. This is necessary to prevent configuration from becoming outdated, which can lead to #3110362 If an update hook modifies configuration, then old configuration is imported, the changes made by the update hook are forever lost. on production.
If you need to apply a patch, make sure it is published on Drupal.org in an issue, and apply it in the Dockerfiles. See ./Dockerfile for an example.
If you are looking at the code for the first time, here are a few design patterns and approaches which explain some of the choices which were made during the development process:
We are not committing code such as downloaded modules, keeping the codebase rather small.
We do not use SASS in this project to avoid complexity. If you want to implement SASS support, consider the following questions:
- Generated CSS should be created in the build phase by
./scripts/create-build.sh
. Because we only allow Docker as a dependency of this project, you might consider a technique such as the one outlined in "Compass Sass to CSS using Docker", Feb. 9, 2016, Dcycle blog. - You should not commit generated CSS to this git repo. You might need to
.gitignore
certain files, or make sure they reside on the container, but not your local computer. - Do you need a real-time compiler directly embedded in your container?
./scripts/deploy.sh is designed to use Docker to set up everything you need to have a complete environment.
We make use of traits to allow classes to implement common code and the Singleton design pattern (see ./drupal/custom-modules/my_custom_module/src/App.php and ./drupal/custom-modules/my_custom_module/my_cystom_module.module for usage).
During development, you might decide that a module needs to be disabled and replaced by some other module or approach.
Your pull request should not remove the module from the Dockerfile. The reason is: the production environment needs to have the module available in order to disable it.
If you like, you might put a comment in your Dockerfile stating when the module can be deleted, perhaps 3 months from now, something like:
# On 2021-10-19, remove block_class
RUN docker-compose require \
...
drupal/block_class \
...
Developers are encouraged to run ./scripts/test.sh
on their machine. If you have Docker installed, this will work without any further configuration.
This lints code (that is, it checks code for stylistic errors) and runs automated unit PHPUnit tests. If you get linting errors you feel do not apply, see a workaround in the "Linting" section, below.
We are not using Drupal's automated testing framework, rather:
- PHPUnit is used to test the code;
- Any Drupal-specific code is wrapped in methods
./drupal/custom-modules/my_custom_module/src/traits/Environment.php
, then mocked in the unit tests. See./drupal/custom-modules/my_custom_module/test/AppTest.php
for an example. - Running tests is done using
./scripts/test.sh
(which includes linting) or./scripts/unit.sh
(which does not include linting), requiring only Docker and no additional setup.
Let's use an example:
On Dec. 4th, 2020, this comment was submitted to a core issue, along with a patch. The test failed with 21 failures. Here is how I reproduced the failure locally for debugging purposes:
-
From the issue comment, click on the failing test, in this case PHP 7.3 & MySQL 5.7 27,916 pass, 21 fail.
-
In the resulting page, note what you'd like to debug. In this example, we can the class
Drupal\Tests\locale\Functional\LocalePluralFormatTest
seems to fail. -
Make sure you have a running local environment; this can be done using
./scripts/deploy 9
. -
Make sure the patch in question is applied to the Drupal codebase on the container (note that a current limitation of this approach is that we are using the latest stable code, not the latest HEAD):
./scripts/shell.sh
curl -O https://www.drupal.org/files/issues/2020-12-04/2273889-42-core-plural-index.diff patch -p1 < 2273889-42-core-plural-index.diff exit
-
Now run the failing test locally:
./scripts/test-class.sh "Drupal\Tests\locale\Functional\LocalePluralFormatTest"
Voilà, you will now get the same results locally as you do on Drupal's testbot, making it easier to work on patches.
Because our git repo does not contain actual Drupal or module code, it downloads these each time ./scripts/deploy.sh is run, so simply running ./scripts/deploy.sh will update everything to the latest version.
We are using the CDN Drupal Bootstrap subtheme.
This project provides some integration with Acquia Cloud.
- We need to install the same version of PHP as can be found on the containers by typing
./scripts/docker-compose.sh exec drupal /bin/bash -c 'php -v'
. - Make sure you are periodically calling "drush cron" from the command line using this technique on your Acquia site.
Acquia suggests this technique for setting up cron.
drush @MY-ACQUIA-ACCOUNT -dv -l http://my-site-on-acquia.example.com cron &>> /var/log/sites/${AH_SITE_NAME}/logs/$(hostname -s)/drush-cron.log
Run ./scripts/create-build.sh
./scripts/acquia/uli.sh
If you are using Pantheon as a host, please refer to ./scripts/pantheon/README.md.
If you are deploying to a Docker host, please refer to ./scripts/docker-host/README.md.
We use a PHP linter and other linters in ./scripts/lint.sh. If your code is not commented correctly or structured correctly, Circle CI continuous integration will fail. See the linter documentation on how to ignore certain blocks of code, for example t()
is ignored in ./drupal-server/custom-modules/my_custom_module/src/traits/Utilities.php.
// @codingStandardsIgnoreStart
...
// @codingStandardsIgnoreEnd
If you have a .sql file with the database database, you can run:
./scripts/docker-compose.sh exec drupal /bin/bash -c 'drush sqlc' < /path/to/db.sql
You can set up the Stage File Proxy module to fetch files from the live server instead of importing the files.
We are using syslog instead of dblog for speed and ease of use. This means you will not have access to log messages through the administrative interface.
To access logs you can then use
tail \
-f ./do-not-commit/log/drupal.log \
Or the Mac OS X Console application.
We are using a dummy email server with a full GUI, Mailhog, as a destination for our emails during development. This is described in Debug outgoing emails with Mailhog, a dummy mailserver with a GUI, March 14, 2019, Dcycle Blog.
Here is how it works:
When you create or update your environment ./scripts/deploy.sh, or at any time by running ./scripts/uli.sh, you will see something like:
=> Drupal: http://0.0.0.0:32783/...
=> Dummy email client: http://0.0.0.0:32781
The ports will be different each time, but using the above ports as an example, you can:
(1) visit http://0.0.0.0:32783/user/password (2) enter "admin" (3) you should see "Further instructions have been sent to your email address." (4) to view the actual email in a GUI, visit (in this example) http://0.0.0.0:32781
We construct our docker-compose environment based on an environment type (see above), therefore the docker-compose.yml
file, on its own, is invalid and will produce:
ERROR: The Compose file is invalid because ...
The solution is to use, for example, ./scripts/docker-compose.sh ps
instead of docker-compose ps
.
Make sure you completely delete your environment using:
./scripts/docker-compose.sh down -v
Make sure you have the latest stable version of your OS and of Docker. If you're really in a bind, you can do a factory reset of Docker (which will kill your local data). Make sure Docker has 6Gb of RAM. (On Mac OS, for example, this can be done using Docker > Preferences > Advanced, then restarting Docker).
Rerun ./scripts/deploy.sh
, and reimport the stage database if you need it (see "Getting a local version of the database", above).
If you run docker network rm starterkit_drupal8site_default
and you get "ERROR: network starterkit_drupal8site_default id ... has active endpoints", you might need to disconnect the Drupal 7 site from the Drupal 8 network first. This should be done in the migration process, but it might not have worked. If such is the case, type:
docker network disconnect starterkit_drupal8site_default $(./scripts/docker-compose.sh ps -q database)
Docker is meant to mimic the Acquia environment relatively well, but there can be certain differences, for instance:
- if you
use path/to/class
instead ofpath/To/Class
, it will work in Docker (local) and in automated tests, but fail on Acquia.
Docker will, by default, used cached versions of each step of the build process for images. For example:
FROM ubuntu
RUN apt-get install something
If you know "something" has changed, you might want to run:
./scripts/docker-compose.sh build --no-cache
./scripts/deploy.sh
Run ./scripts/deploy.sh
, which will restart rsyslog.