Deploying web apps with Docker

By Martin Rusev read

In 2014 Docker has grown immenesely and has a become a prefered tool for many people in the devops community. I've been using it for over a year and the moment Docker hit 1.0 a couple of months ago I decided to bite the bullet and migrate my project Amon to Docker. These are some of the mistakes I made and the lessons I learned during the last couple of months.

Don't put everything in the container

Some articles compare Docker to Git and that is not entirely true. With Git you are encouraged to commit often, with Docker however "commiting" often(running docker build) will add history layers, eventually hitting the limit of 127.

Docker containers are designed to be portable and immutable - you can install libraries, databases, servers. The dynamic parts like data directories and log files however remain outside - on the host machine.

Putting your web application in a container should follow the exact same principles - install the requirements beforehand with docker build, but leave the source code in a directory on the host machine and mount it on boot.

// You have to rebuild your container on every change
// adding history layers, eventually hitting the limit which is 127. 
ADD myapp /home/myapp

// You can do this as many times as needed
VOLUMES ['myapp']
docker run -v /myapp:/home/myapp

Use the Docker API

The recomended way to boot a container is:

docker run
    --name=myapp myapp:latest

Written like this it already looks error prone and in a dire need of simplification. This is not all though - we have to add a couple more commands for restarting, stopping the container. It doesn't get better when you put these commands in a deployment capistrano/fabric script.

There are a lot of tools out there for managing docker containers - Panamax, Fig, Flynn and Deis to name a few, but at least for me a lot can be acomplished with a simple wrapper around the Docker API The example below shows my own Python script, but I think it is easy to adapt and write something similar for Ruby, Nodejs, etc.

from datetime import datetime
from docker import Client
from djangoapp.instance.models import Instance

class AmonInstance(object):
    def __init__(self, account=None):
        self.client = Client(base_url='unix://var/run/docker.sock', version='1.15')
        self.account_id = 1
        self.instance = Instance.objects.get(account_id=self.account_id)

    def reload_container(self):


    def start_container(self):
        container = self.client.create_container('martinrusev/djangoapp',

            port_bindings={8000: 8000},
                'bind': '/djangoapp',
                'ro': True
            '/var/log/djangoapp': {
                'bind': '/var/log',

        container_id = container['Id']


    def stop_container(self):


Debugging running containers

The container is running, but your web app is not working. There are two ways to the debug a docker container:

With a more complicated web app with multiple services running on different machines and ports, ENV variables, etc these methods are not very helpful. In my case with a Django app I was given generic errors that the server failed to start or I could not get cron to work.

Loggin in the container is the next logical step, but this is something you can't easily do with the default docker commands. I found a tool called docker-bash part of the popular Phusion Baseimage. It is not locked to Baseimage though - you can log into any image/container. To install it follow the instructions here and then:

//To see the list of running containers
$ docker ps -a

// To log in the container
$ docker-bash container_id

Running more than 1 process in a container

This is a highly debatable issue in the Docker community. Some people prefer to stick to the Unix philosophy ("Make each program do one thing well."), others like me prefer to group containers into logical units. I think there is no right or wrong way to do this and it really depends on the problem you are trying to solve.

In my specific case I am grouping cron and django app in the same container:

Config files vs ENV

This is not a common issue, but I will put it here anyway. You will encounter this problem if your app has some soer of background processing with cron jobs.

The most important thing you should remember when running cron in a container is the fact that cron does not have access to the ENV variables set on boot. Things like DB_URL=postgres://myhost

One way to solve the problem (and simplify that docker run command) is config files instead of ENV variables.

// The config file, mounted on boot

// In our webapp 
CONFIG_FILE = '/home/webapp/config.cfg'
config = ConfigParser.ConfigParser()


If you are a web developer and still on the fences about giving Docker a try, I can asure you that it has become a very stable and reliable tool.

Even if you don't go all the way and deploy it in production - you can use it for building local enviroments, sharing those enviroments and keeping your dev machine clean.

I definitely prefer docker pull redis vs all these steps required to install just one libary and then replicating those steps on the production server(I use Ansible, Docker is still easier)