| About Me

A Rails development environment with Docker

Docker and Docker Compose reduces bringing up a development environment on any system which supports Docker to a single command. For me this means the time to a working development environment, even on a fresh machine for a project I haven't worked on for months or years, is just a few seconds.

This tutorial is not a comprehensive introduction to either Docker or Docker Compose, for this this I highly recomend the official docker compose tutorial as a starting point. The steps here will stand alone but it's definitely beneficial to get familiar with the docker-compose fundamentals at some point.

Installing Docker

We'll need to have Docker installed locally https://docs.docker.com/install/. We'll also need docker-compose installed, for OSX then this is included when installing Docker, for Linux, it must be installed separately and instructions are here; https://docs.docker.com/compose/install/#install-compose.

Dockerising a Rails Application

For an existing Rails application the files below should be added to the root of the project. If generating a new project we should create an empty directory with the name matching the desired name of the Rails project and create the files here. We'll cover generating the Rails application itself using Docker below.

First we add a file called Dockerfile to the project root which defines how to build the runtime environment for our application:

FROM ruby:2.4.2

# Install apt based dependencies required to run Rails as
# well as RubyGems. As the Ruby image itself is based on a
# Debian image, we use apt-get to install those.
RUN apt-get update && apt-get install -y \
  build-essential \

# Configure the main working directory. This is the base
# directory used in any further RUN, COPY, and ENTRYPOINT
# commands.
RUN mkdir -p /app

# Copy the Gemfile as well as the Gemfile.lock and install
# the RubyGems. This is a separate step so the dependencies
# will be cached unless changes to one of those two files
# are made.
COPY Gemfile Gemfile.lock ./
RUN gem install bundler && bundle install --jobs 20 --retry 5

COPY . /app
RUN rm -rf tmp/*

ADD . /app

Replacing YOUR_EMAIL with your email address.

Then we add a second file called docker-compose.yml to the project root containing the following:

version: "3"
    build: .
    command: rails server -p 3000 -b ''
      - .:/app:delegated
      - "3000:3000"
      - DB_USERNAME=postgres
      - DB_PASSWORD=postgres
      - DB_HOST=postgres
      - postgres
      - redis

    image: postgres:9.4
      - postgresdata:/var/lib/postgresql/data
      - "5432:5432"

    image: redis
    command: redis-server --appendonly yes
      - redisdata:/data
      - "6379:6379"


The compose file

A few elements of this compose file are worth noting as they differ from the many other Rails compose files out there:

  • We don't use links, these are no longer the recommended approach for establishing communication between containers. Instead services are available by hostname where the hostname will be the name of the service. E.g. in this case our app conainer can access postgres via the hostname postgres and redis via the hostname redis
  • The :delegated option on the volume mount for our app directory. This is specific to Docker for Mac, although won't cause problems on other platforms. Details of what this does are available here but it provides a substantial improvement in filesystem performance on OSX, without which day to day development of Rails applications can be painful. If file performance is still not sufficient for your application, it's worth considering Docker Sync although the 60x slowdown references on that site is substantially reduced by the sue of delegated on OSX

Generating a new application

When working with an existing Rails application, we can skip this section. Using Docker to generate the entire application is especially useful when trying to maintain no local development environment at all, so we can develop Rails applications without ever needing a local Ruby install.

One of the quickest ways to generate a new Rails application is simply to use the pre-built Bitnami image and execute:

docker run --rm -v $(pwd):/app bitnami/rails:latest rails new my_app --database postgresql

Which will pull the Bitnami image and generate a new Rails app called my_app in the current directory.

Alternatively, if you wish to avoid use of a third party image, alongside our Dockerfile, we'll need to add a Gemfile containing the following:

source 'https://rubygems.org' do
  gem "rails",

and a blank Gemfile.lock

Then execute:

docker build -t my_name/my_rails_image .
docker run --rm -v $(pwd):/app my_name/my_rails_image rails new my_app --database postgresql

Which builds an image based on our Dockerfile, tags it with the name my_rails_image and then runs rails new myapp --database postgresql within a container based on that image and creates a new Rails application as a subfolder.

Once we have this image built locally, we can re-use it in future with:

docker run --rm -v $(pwd):/app my_name/my_rails_image rails new my_app --database postgresql

Dockerising the application

Our docker-compose.yml file will bring up not only our rails application, but supporting Postgres and Redis instances.

First we'll make some changes to our Rails application so that all configuration is taken from environment variables. We'll then update our docker-compose.yml file to set some environment variables. This lays the groundwork for using something like Hashicorp's envconsul for managing configuration in production at a later date. It's also a core tenant of building 12 factor applications.

We'll begin by modifying config/database.yml to pull connection details from the environment by adding these three lines:

username: <%= ENV.fetch("DB_USERNAME") %>
password: <%= ENV.fetch("DB_PASSWORD") %>
host: <%= ENV.fetch("DB_HOST") %>

These can either be added to the default section or individually to the development and test sections.

It may look like we could skip this entirely and just use the standard DATABASE_URL environment variable with something like postgres://username:password@postgres/DB_NAME but this will cause problems when running commands such as rake db:migrate. These do not reload environment variables between operating on the test and development databases and therefore raise exceptions when trying to apply migrations to test and development as these will both try and use the same database name.

We can see these map to the values set in the environment section of our app definition in docker-compose.yml:

    build: .
    command: rails server -p 3000 -b ''
      - .:/app:delegated
      - "3000:3000"
      - DB_USERNAME=postgres
      - DB_PASSWORD=postgres
      - DB_HOST=postgres

The environment section allows us to set up the environment variables which will be set within the container. As discussed above, the postgres service will be available to our container automatically on the hostname postgres which matches the service name in docker-compose.yml. The default credentials for the official postgres image are postgres and postgres and there's more about customising here.

We can apply the same approach to the config/secrets.yml file if needed as well as anywhere else in the application we want to pass in configuration dynamically. If something like dotenv is in use then we can simply update our local .env file with the above environment variables, rather than defining them in the compose file.

Starting the rails application

We can now build the docker image based on our Dockerfile by executing:

docker-compose build

Once this is complete we can start our application with:

docker-compose up

This will start postgres, redis and our rails application, exposing the rails application on port 3000 so that we can access it by visiting http://localhost:3000 as usual.

Using with pry

If we want to use something like pry-rails to debug our rails application, instead of executing docker-compose up, we should use:

docker-compose run --service-ports app

Running one off commands

Usually when setting up a rails application we would run rake db:create and rake db:migrate. If dockerising an existing application, we're likely to find that we're greeted with an error page that the database does not exist.

To run one off commands using docker-compose we use the run command in the format:

docker-compose run SERVICE_NAME CMD

So to run rake db:create db:migrate within the context of our rails application we would use:

docker-compose run app rake db:migrate db:create

Similarly to start a console we would run:

docker-compose run app rails console

We could run rspec tests with:

docker-compose run app rspec spec

Adding Gems

When adding new gems, we first update the Gemfile, then execute:

docker-compose run app bundle

To update the Gemfile.lock and then:

docker-compose build

To cache the gems in the image.


We can launch a bash shell in our app container using:

docker-compose run app bash

It's important to bear in mind that each invocation of the above runs in a separate, completely isolated, container and so outside the /app directory which is bind mounted to our local directory, the file-systems are transitory and independent of one another.

Being able to run a shell within our app container can be the key to avoiding a lot of frustrating workflow issues when working in a Docker based development environment.

A great example is when upgrading a Rails version. The workflow is typically:

  • Update the version of GEM1 within the Gemfile
  • Execute bundle update GEM1
  • Look to see which dependency issues are raised
  • Update the Gemfile again
  • etc etc

This can be slow and painful if a separate docker-compose run app ... is required every time. Instead we can use docker-compose run app bash once and then iterate within that container as many times as we want. Once it works we simply jump out of the container and execute docker-compose build to persist the new gems to the image.

Deploying to production

Part two covers how to deploy this application to a Kubernetes Cluster on any VPS or Bare Metal provider without having to learn several thousand previously unheard of pieces of terminology.

Feel free to ping me on twitter @TalkingQuickly with any questions or feedback, in particular I'd love to know if a screencast version of this tutorial would be useful?