The `docker-compose.yml` file¶

Function of `docker-compose.yml`¶

docker-compose.yml is used exclusively for local project set-up

In the Divio project architecture, the docker-compose.yml file is not used for cloud deployments, but only for configuration of the local environment. On the cloud, the deployment is taken care of by dedicated systems on our servers.

This means that entries in or changes to docker-compose.yml will not affect cloud deployments in any way.

In order to do something useful with containers, they have to be arranged - orchestrated - as part of a project, usually referred to as an ‘application’.

There are multiple ways of orchestrating a Docker application, but Docker Compose is probably the most human-friendly. It’s what we use for our local development environments.

To configure the orchestration, Docker Compose uses a docker-compose.yml file. It specifies what images are required, what ports they need to expose, whether thy e have access to the host filesystem, what commands should be run when they start up, and so on.

Services defined in `docker-compose.yml`¶

In a docker-compose.yml file, services represent the containers that will be created in the application.

When you create a new Divio project using one of our defined project types (e.g. Aldryn Django or PHP/Laravel using Flavours) it will include a docker-compose.yml file ready for local use, with the services already defined.

If you start with a blank project type, you will need to assemble the docker-compose.yml file yourself. This is a fairly straightforward process once you know what you are doing. Our Django tutorial includes steps for creating a complete docker-compose.yml file from scratch. It’s a very good way to become familiar with using Docker Compose, even if you aren’t going to be using Django.

For a working local project, various things need to be defined in the file. In a Divio project, there will be a web service, that’s built in a container using the Dockerfile. There will typically also be a db service, from a standard postgres or other database image.

Most Divio projects will use a docker-compose.yml that contains entries along these lines.

web:
 build: .
 links:
  - "database_default"
 ports:
  - "8000:80"
 volumes:
  - ".:/app:rw"
  - "./data:/data:rw"
 command: python manage.py runserver 0.0.0.0:80
 env_file: .env-local

database_default:
 image: postgres:9.6
 volumes:
  - ".:/app:rw"

Some projects will have additional services (such as Celery for example) defined.

Let’s look at the components of the file more closely.

The application container service, `web`¶

The first definition in the file is for the web service. In order, the directives mean:

build: build it from the Dockerfile in the current directory
links: a link to the database container (database_default)
ports: map the external port 8000 to the internal port 80
volumes:
- .:/app:rw maps the parent directory on the host to /app in the container, with read and write access
- /data:/data:rw maps the data directory on the host to /data in the container, with read and write access
command: by default, when the command docker-compose run is issued, execute python manage.py runserver 0.0.0.0:80 (this will override the CMD instruction in the Dockerfile)
env_file: use the .env-local to supply environment variables to the container

The `volumes` directive¶

When you execute a docker-compose command, the volumes directive in docker-compose.yml file mounts source directories or volumes from your computer at target paths inside the container. If a matching target path exists already as part of the container image, it will be overwritten by the mounted path.

For example:

volumes:
  - ".:/app:rw"
  - "./data:/data:rw"

will mount the entire project code (at the relative path .) as the /app directory inside the container, even if there was already an /app directory there*, in read-write mode (i.e. the container can write as well as read files on the host).

This allows you to make changes to the project from your computer during the local development process, that will be picked up by project inside Docker. These changes will be available to the project only as long as the host directory is mounted inside the container. In order to be made permanent, they need to be committed into the repository so that they will be picked up when the image and container are rebuilt.

Implications for local testing

Nearly everything in /app in the container is also present in the project repository and thus on the host machine. This means that it is safe to replace the container’s /app files with those from the host.

However, any files in /app that are placed there during the build process, i.e. the execution of the Dockerfile, will not be available in the local environment. For a standard Django project, these will include:

the compiled pip requirements, in requirements.txt
collected static files, in static_collected

In most cases, this will not matter, but sometimes these files are required in local development. For example, the requirements.txt may contain useful information about dependency relationships, or the Dockerfile may have performed custom processing of static files.

In that case, the - ".:/app:rw" line can be commented out in docker-compose.yml. In this case, the container will use the files baked into the image, and will not use the local host’s files.

This will allow local configuration to replicate the cloud environment even more closely.

Environment variables¶

Environment variables are loaded from a file, specified by:

env_file: .env-local

The database container service, `database_default`¶

The second definition is for the database_default service.

On the cloud, the project’s database runs on one of our database clusters; locally, it runs on a Postgres instance in database_default.

The directives mean:

image: build the container from the postgres:9.6 image
volumes: map the parent directory on the host to /app in the container, with read and write access

See Expose the database’s port for an example of adding configuration to docker-compose.yml.

Required database service configuration

The Divio CLI expects that the database service will be called database_default (or, in some older projects, db). If the name is changed, operations such as divio project pull db will fail.

The volumes directive needs to map the container’s /app directory as described above, for the same reason.

The docker-compose.yml file¶

Function of docker-compose.yml¶

Services defined in docker-compose.yml¶

The application container service, web¶

The volumes directive¶