Docker 101 - Dockerfile

This is the second post of the Docker 101 tutorial; you can check the first one if you missed it. So far we have installed Docker, we've made a couple of different “Hello, world” with Docker and we have learnt basic modes of running (interactive, daemon).

This new post is meant to introduce a new scenario of use (more real, but still simple), and the Dockerfile; so we'll learn how to build custom images.

Let's go!

Our new scenario

In this part of the tutorial we're going to work with a real project, in its simplest form, to go through the main interaction. I've prepared a very simple project so you can use it; now you'll need both git and python knowledge. If you don't, you can ask for some help or you can be imaginative creating a similar environment.

The code is available in my Github. It's a very simple API made with Anillo and Python. To get ready with your environment:

  • You need to copy myapp.ini and pytest.ini files outside the repository
  • You need to export two env vars: MY_APP and MY_APP_TEST, pointing the previous files
  • You need sqlite (usually it's in all Linux)
  • I encourage you to use a virtualenv (with Python3!!!)
  • You need to install requirements.txt and requirements-server.txt

Once you've all of this, you can run the API locally directly with Anillo:

(myapp) src/ $ python serve --create-db --with-fixtures --no-hot-reload

Or, if you installed requirements-server.txt, you can run the application with Gunicorn:

(myapp) src/ $ gunicorn -b --access-logfile - --error-logfile - --log-level debug 'wsgi:load_application("create-db", "with-fixtures")'

You can check it visiting http://localhost:5005/api/v1 or http://localhost:5005/api/v1/users. Yay! You have a running API in your computer! ;-) Congratulations!

Well, this was not about docker, was it? Be patient, this was the project we're going to dockerize, so it's interesting spending some time to have the same environment. Now, why would we use docker for this? For instance, we may want to run automatically the tests, and if they pass, deploy our API in PRE environment connected to a postgres container. To do this, we need an image with our API (we already cloned it in our machine manually), with all the dependences (we installed them manually) and with the proper command (which we launched manually). How can we do an automatic process with so many manual operations?

Dockerfile will help us!


Dockerfile is a file which sets a “FROM” image, and adds all “the commands to assemble an image”. The simplest Dockerfile could be:

FROM ubuntu:14:04

Just a file called Dockerfile with this single line is neccessary to build our own image. Let's see how to build an image:

$ docker build -t my-image:1.0 .

You have to run it in the directory where the Dockerfile is. The -t option sets the name and tag for the new image. Now, you can check you have a new image available:

$ docker images
REPOSITORY          TAG                 IMAGE ID            CREATED             VIRTUAL SIZE
ubuntu              latest              6cc0fc2a5ee2        3 days ago          187.9 MB
my-image            1.0                 6cc0fc2a5ee3        11 days ago         187.9 MB

Take some minutes, go over the previous post and try to run the same “Hello, world” exercises with the new images. Great!!!

Our Dockerfiles

Let's create a couple of Dockerfiles to build our images. Those images will be:

  • An image with postgres and a database called “myapp”
  • An image with the API, conected to the database, serving the API with gunicorn

To do this, we need this tree directory:


You can find all these files in the repository, but I invite you to write them on your own. All commands from now will be executed from docker-101 directory.

Next: create a Dockerfile in postgres directory, with the following content:

FROM postgres:9.4.5
MAINTAINER Yamila Moreno

What's the meaning of this Dockerfile?

  • The FROM statement says that we are starting from a postgres (official) image, in its version 9.4.5
  • The MAINTAINER line just says who am I, and who should you send the huge amounts of money ;-)
  • As the documentation says, if we build the image with “POSTGRES_DB” env var, the value will be used as the name of the database. We use the statement ENV.

Easy peasy. Let's build our image:

$ docker build -t myapp-postgres:1.0 -f api-example/docker/postgres/Dockerfile .

And check the new images exists. Let's run it to check if everything went ok:

$ docker run -d --name myapp-postgres myapp-postgres:1.0

Now, enter in the container and execute psql:

$ docker exec -ti myapp-postgres /bin/bash
root@ec861b83cdee:/# psql -U postgres myapp
psql (9.4.5)
Type "help" for help.

See that we are entering in “myapp” database, instead of “postgres” default database. So yes, everything went well!

Next, let's do the same operations for our API, which is a bit more complex. Create a Dockerfile in api directory with the following content:

FROM ubuntu:14.04
MAINTAINER Yamila Moreno

# Install dependencies
RUN apt-get update
RUN apt-get install -y -qq curl python3-pip git libpq-dev
RUN pip3 install virtualenv

# Configure locales
RUN locale-gen "en_US.UTF-8"
RUN dpkg-reconfigure locales


# Setup the application
RUN virtualenv -p python3 venv
COPY api-example /api-example
RUN /venv/bin/pip install -r /api-example/requirements.txt
RUN /venv/bin/pip install -r /api-example/requirements-server.txt

# Set specific application env vars. We're using the default configuration
ENV MY_APP=/api-example/src/settings/myapp.ini

# Run application
WORKDIR /api-example/src
CMD /venv/bin/gunicorn -b --access-logfile - --error-logfile - --log-level debug 'wsgi:load_application("create-db", "with-fixtures")'

Lots of things!! Let's go through each new line:

  • EXPOSE: informs the environment is exposing this port. It's a good practice
  • WORKDIR: just indicates that next command will be executed from that directory
  • RUN: runs a command. In our case, we're updating the system and installing some dependencies. We also create a virtualenv.
  • CMD: remember the “one command” we have talked before? This is THE one command! In fact it's an array of commands, but in our example, we just need one.

Now, we can bulid the new image:

$ docker build -t myapp-api:1.0 -f api-example/docker/api/Dockerfile .

And again, run it:

$ docker run -d --name myapp-api myapp-api:1.0

Now, we should find our API in http://localhost:5005; but if you have followed my instructions, you won't find anything. Why? In fact, lets run a different image and make the same test:

$ docker run -ti --name myubuntu ubuntu:14:04 /bin/bash
root@ec861b83cdee:/# apt-get install curl # yes, you need to install it
root@ec861b83cdee:/# curl http://localhost:5005

Still not working. This is because, by default, docker doesn't share anything, nor with the host (your machine), neither with the other containers.

Visibility between containers

First of all, we want to make our containers visible between them. Time to introduce new concepts:

–link myapp-api:myapp-api; it's an option when running the image that creates a link between the container you are running and the container in the parameter. Here we're saying that the container running our image is linked to “myapp-api” with the name “myapp-api”.

So, let's stop the ubuntu machine:

$ docker stop myubuntu

And run it again linking the containers:

$ docker run -ti --name myubuntu --link myapp-api:myapp-api ubuntu:14.04 /bin/bash
root@ec861b83cdee:/# apt-get install curl # yes, you need to install it. Yes, again. Each time.
root@ec861b83cdee:/# curl http://myapp-api:5005/api/v1 # we are using the mapped name
{"api-version": "1.0", "message": "Welcome to this API."}

At last!! It worked!! :__) It's being a bit hard, isn't it? but we're learning step by step lots of concepts. Take a moment to read again any part and make different tests. And take a moment to watch over your screen, through the window, where the sun shines and the wind… winds… ;-) And let's move on!

Note about “link”: this argument is to be deprecated and soon; the “network” feature will cover the old functionality and will add more. When I'm writing this post, “network” is still not released in stable version, so I'm using links.

Visibility with the host

We have seen how to make the containers visible between them; but usually, we'll need that some containers are visible from the host. And time to introduce new concepts:

-p 5433:5432; it's an option when running the images that exposes the container's port (5432) in the host's port (5433). I'm mapping in my 5433 because my own 5432 is “busy” with my own postgresql.

Let's test it with our postgres image:

$ docker run -d -p 5433:5432 --name myapp-postgres postgres:9.4.5

And now, access the database from our machine (you'll need to have postgres installed):

$ psql -h localhost -p 5433 -U postgres postgres
# And you'll see some output like:
Line style is unicode.
Border style is 2.
Output format is wrapped.
psql (9.4.5)
Type "help" for help.


There you are! This database is running in a container :D But previously we were trying to make the api accesible from the host. Take a moment to think how would you do this and then continue reading.

Stop and delete running containers and run the api from scratch with the new param:

$ docker run -d -p 5005:5005 --name myapp-api myapp:1.0

And then, from our machine:

$ curl http://localhost:5005/api/v1
{"message": "Welcome to this API.", "api-version": "1.0"}%

Total success!!

This post has been a bit dense, with many new concepts, not easy at all (dockerfile, link, port mapping). Congratulations to get here. The reward will be the next post, where we will make it easier to deal with all these running options. We'll talk about docker-compose.

Happy hacking!

comments powered by Disqus