Our Reference Baseline #

Adopting the baseline on a project enhances operational efficiency. It ensures quality, consistency, and clear communication for customers; reduces decision fatigue, boosts productivity, and fosters collaboration for our team; and supports scalability, compliance, and continuous improvement for Axelerant. This comprehensive approach addresses current challenges and prepares us for future growth. Read more…

At this time, this baseline is standardized for our projects in CDM but will eventually also apply to our managed projects.

Git Workflow and Health #

The exact Git workflow necessarily changes from one project to another. That said, it is possible and we must aim for certain elements of consistency based on our (industry) collective experience. Many of these are defined so because they are obvious to the developer community in general.

Read more details here.

Documentation #

Documentation is one of the most underrated activities a developer can take up. We remind and encourage the team to keep the documentation current by applying process gates. This may look like a check applied when reviewing a PR, when approving a task, or before marking a task as “done”.

Keep in mind the audience for the documentation when writing. The audience could be you, 2 weeks later. It could be another developer on the team learning the system or being onboarded to the project. Keep the documentation readable and get to the point quickly.

Technical documentation must be located in the project repository where it is easy for the developer to refer to it without leaving the IDE. Integration with the IDP allows us to render and read the documentation using a browser. Because the documentation is written in plain text files, markdown is the optimal format for this which is both easy to learn and easy to read.

Code Hosting #

To keep things simple, we will always host our projects on one of our GitHub organizations that are a part of our GitHub Enterprise account. We get benefits such as advanced access control and 50,000 action minutes per month. Exceptions are possible but we will try to avoid them as much as possible.

Local Development with DDEV #

DDEV is our tool of choice for setting up local development environments across projects. You need Docker to run it and you can provide the Docker engine using the official Docker Desktop, Lima, or Orbstack. Follow the installation instructions to set up DDEV on your machine.

It is very likely that the project you’re working on already has DDEV configured. In such cases, you simply have to clone the project and run ddev start in the project directory. Follow these instructions if the project does not have DDEV configured.

Application Cache #

With CMSes like Drupal, we always use an application cache like Redis or Memcached. With Drupal’s built-in dynamic caching abilities, there is no reason to not use an external cache. Given the size of our typical project, the slight increase in complexity is worth it.

We default to Redis unless there is a good reason to use Memcached. Often the reason is that the hosting provider may only support Memcached. Given the recent Redis announcement, we may change this recommendation based on the hosting landscape and performance considerations.

Database #

The default database we pick is MariaDB. Like with the application cache, we may switch to MySQL depending on the hosting provider. In all cases, we try to match the versions on the production instance as closely as possible.

As of this writing, the LTS version for MariaDB is 10.11 and for MySQL is 8.0. Try to pick this version for both local development and production unless there is a specific need. Both Drupal 10 and 11 can use these versions.

On some hosts, you also have to set the transaction isolation level. Follow the instructions here.

File Storage for user content #

Drupal supports public and private file schemes to store user-uploaded content. The public file scheme is the default mechanism and is suitable unless there is a specific requirement. Both of these schemes are stored in the local filesystem (e.g., public files are saved in sites/default/files). Hosts that support Drupal provide a mechanism to make these locations writable.

In the future, we will provide infrastructure support to keep these files on an object store such as S3 by default. For now, use modules like stage_file_proxy to access these files in development environments.

Local checks #

We run local checks on our changes before committing our changes. For Drupal projects, we have a convenient wrapper called drupal-quality-checker that uses GrumPHP to run various checks on commits. Installation is a question of installing the composer package and copying the configuration files.

Code Quality checks #

We use tools such as PHPCS, PHPStan, eslint, stylelint, and others depending on the language. It is always a good idea to run these tools before making a commit and we do that automatically through tools such as drupal-quality-checker and husky. Include such tools in your project at all times and have them run on git commits. Additionally, these tools must also run in CI jobs.

Debugging #

Step debuggers are extremely important tools to have ready during development. For PHP, the debugger of choice is XDebug. There is an excellent integration available with DDEV and IDE configuration is not very complex. Documentation for XDebug with DDEV has all the setup instructions to enable XDebug and configure IDEs to listen to it.

IDP Integration #

Axelerant IDP is the portal to view all technical information relevant to a project. Our IDP is built with Backstage and we add projects to the IDP using a catalog-info.yaml file. This file defines some of the attributes such as the name, Jira project ID, team, and IDs for external tools like NewRelic and Sentry.

apiVersion: backstage.io/v1alpha1
kind: Component
metadata:
  name: contrib-tracker
  description: Axelerant Contribution Tracker
  title: Contrib Tracker
  annotations:
    github.com/project-slug: contrib-tracker/backend
    sentry.io/project-slug: contrib-tracker
    newrelic.com/dashboard-guid: <GUID>
    jira/project-key: CONT
    backstage.io/techdocs-ref: dir:.
spec:
  type: website
  owner: platform_engineering
  lifecycle: production
  system: public website

Once you add this file to the default branch in your repository, manually import the project using the “Register an existing component” button in the IDP.

Continuous Integration (CI) #

Our CI tool of choice is GitHub Actions. We rely on automation to run repeated tasks reliably and as we use GitHub for hosting our code, it is a simple decision to use GitHub Actions we have included as part of our Enterprise subscription.

A typical CI pipeline begins with linting and code-style checks. This is followed by tests arranged in the order of quickest tests first followed by longer ones (this is so that we get quick feedback in case of a failure). Finally, the pipeline deploys to a feature branch or a present environment for manual verification.

Since a CI pipeline is considerably more complex, an example describes the process best. Our go-to example is the pipeline of Contrib Tracker. Use that as your base and change the specifics as required.

Updates Tracking #

As we are on GitHub, we use Dependabot for tracking updates. We typically track updates for all the dependencies we use (typically PHP, JavaScript, and GitHub Actions itself). We schedule alerts on a weekly basis unless there is a specific need. We also group related dependencies so that we are not flooded by pull requests and associated CI runs.

As before, use the contrib tracker project as an example of a Dependabot configuration file and change it to suit your needs.

Logging #

We can’t fix what we can’t see. Logging is the most fundamental aspect of observability in how our applications run so that we fix a problem, even proactively. For this, we rely on something more sophisticated than simply logging into the database or syslog. Our tool of choice for this purpose is Sentry.

Any application that you build must be configured to write logs to Sentry. This is fairly straightforward as Sentry has SDKs in a number of languages and frameworks. For Drupal, use the Raven module to pull in the integration and also configure what should be logged.

Read more details here.

Functional Testing #

We automate our tests using Cypress. We focus more on functional testing rather than unit testing because this is more of a fit for a typical project at Axelerant. Unit testing has its place and we should write unit tests but we recognize that writing tests introduces a cognitive load. To limit the number of paradigms that people have to deal with, we are limiting ourselves to the most impactful kind of tests and that is a functional test run in a browser-like environment.

We have a framework built using Cypress and integrated with Percy for visual regression testing. It may be found on GitHub along with instructions on how to integrate it in a project. These tests must run via CI but should also be run locally before pushing the changes. A CI run example may be found in the contrib-tracker example pipeline.

Reverse Proxy #

A reverse proxy is highly recommended for most of the sites that we build. Varnish is a popular option for use with Drupal and it is supported out of the box with most of the PaaS providers we use.

Varnish is a separate application from Drupal’s server stack. You do not need any module or customization on Drupal as long as the cache headers are correct. In practice, it is useful to install a module such as Purge to invalidate the cache when required.

block-beta
  columns 7
  Internet(("Internet")) space varnish{{"Varnish"}} space web("Web Server") space:2
  space:7
  space:4 fpm("PHP-FPM") space:2
  space:4 app(["Drupal"]) space db[("Database")]
  space:6 other("Other Services...")
  Internet-->varnish
  varnish-->web
  web-->fpm
  fpm-->app
  app-->db
  app-->other
  style fpm stroke-dasharray: 5 5
  style other stroke-dasharray: 5 5

ddev get ddev/ddev-varnish
ddev restart

Backup and Recovery Plan #

Use the backup capabilities of the PaaS provider where the websites are hosted. PaaS providers provide mechanisms to set the backup frequency and retention. Here is a list of some of the common ones that we use.

• Acquia Cloud
• Acquia Site Factory
• Pantheon
• Platform.sh

Incident Reporting #

We use OpsGenie for incident reporting and it is integrated with our other tools such as Jira and Slack. We use UptimeRobot to monitor the websites that we support and create alerts in OpsGenie through its integration.

block-beta
  columns 5
  space:4 jira("Jira")
  uptimerobot("UptimeRobot") space opsgenie{"OpsGenie"} space slack("Slack")
  jsd("Jira Service Desk") space:3 app("OpsGenie App")
  uptimerobot--"Site down"-->opsgenie
  jsd-->opsgenie
  opsgenie-->jira
  opsgenie-->slack
  opsgenie-->app

Each website that we maintain must be configured on UptimeRobot for monitoring. This sends alerts to OpsGenie which, in turn, sends notifications as per our team structures. For teams to be able to respond to alerts, there must be a runbook with all the information required to understand the alert and resolve the situation.

Infrastructure (non-prod) #

To streamline our development and deployment processes, we use Platform.sh as the platform for our non-production environments. We try to match the stack as closely as possible with the production environment. This means that we replicate all parts of the stack–components like varnish, redis, and SOLR. We also try to match the version as closely as possible.

To configure your project to run on platform.sh, you need at least these three files.

• .platform.app.yaml
• .platform/routes.yaml
• .platform/services.yaml

The links below point to sample configuration for a Drupal 10 website. Here are a few more templates you can evaluate

• Drupal 10
• Drupal 9
• WordPress
• Laravel
• Decoupled Drupal

Secret Management #

We use various platforms to securely store and access our secrets. Of them, 1Password is our canonical source for all types of secrets that we need on a project. This includes account logins, SSH keys, API keys and tokens, credentials, and other sensitive information that would be used by the team or by the system during build or runtime.

Additional places where secrets may be stored are:

• Repository secrets on GitHub.
• Any mechanism to store sensitive information on various PaaS we use.

Access control for team members #

All the platforms that we use to collaborate on code and projects provide some sort of access control. Apply the principle of least privilege required when assigning permissions. Specifically on GitHub, teams must be used to grant access to a repository or multiple repositories. Using teams helps us manage access control in a single place and gives a suitable level of access to any number of repositories. Typically, multiple teams should be created for team leads and other engineers so that the team consisting of team leads can get elevated rights on the repository.

Deployment (prod) #

When continuous deployment is not possible, tagged deployment is preferred when deploying a release to the production. Since there is usually only a single version of a website in production, a simple branched workflow with a single trunk branch (often called main or master) is sufficient. This branch can be continuously deployed to a staging environment for internal or user testing. Once a test run is complete, that commit is tagged and deployed to the production instance.

gitGraph:
  commit
  commit tag: "v20240503"
  branch feature-123
  commit
  commit
  checkout main
  commit
  merge feature-123
  commit tag: "v20240517"
  commit

In the above, the main branch represents the staging environment and the tags of v20240503 and v20240517 are the tags deployed to production. As you might have deduced, the version numbers are dates in vYYYYMMDD format. Any version format may be used as long as this is well-defined and understood within the team. For websites, versions of this format make it easy to create tags and figure out what a tag means without worrying about semantic versioning.

Search Engine #

Common search engines used with Drupal are SOLR and Elastic Search. We typically always prefer SOLR because of the available ecosystem within Drupal, hosting support, and tools. But Elastic Search and other proprietary options such as Algolia or open-source ones like Meilisearch. When the search index is externally hosted, we don’t have to worry about setting it up for local development or in other infrastructure. For self-hosted options, there are ways to configure local environments to run the search indexer.

For SOLR, keep in mind that there are multiple DDEV recipes and you should use the one that matches your production instance. For example, Acquia Search is currently Solr 7 (planned upgrade to 8). In this case, you should use

• ddev-solr (recommended). It installs SolrCloud with Solr 9.
• ddev-drupal-solr. It installs Solr 8 and does not work with earlier versions.
• ddev-contrib. This includes support for many versions of Solr and you can match with your production instance. As mentioned before, use Solr 7 for Acquia Search.