Django + Postgres + Docker + Travis CI + Heroku CD

Basic workflow of testing a dockerized Django & Postgres web app with Travis (continuous integration) & deployment to Heroku (continuous deployment).


Prerequisite:

1) This post assumes that the reader has accounts with Github, Travis & Heroku & already has the accounts configured. For example, linking Travis with Github, adding a Postgres server in Heroku & setting OS environment variables in Travis & Heroku websites.

2) Basic working knowledge of Django & Docker.


Code on my Github


Which copy of Postgres to use during the different stages in the workflow?

During development, we won’t be using a local copy of Postgres database server. The Docker’s copy of Postgres is used.

When testing in Travis, we don’t have to ask Travis for a copy of Postgres, we won’t be using Travis’s copy, we’ll be using the Docker’s copy.

During deployment, we have to use Heroku’s copy.


makemigrations

Always run docker-compose web run python manage.py makemigrations before deployment to Heroku or in our case, before pushing to Github.

The actual python manage.py migrate for the Postgres server addon from Heroku will be run in the Procfile file.


Listing files & directories in a tree:

cd web_app_DPDTH

$ tree -a -I "CS50_web_dev|staticfiles|static|templates|LICENSE|README.md|__init__.py|settings_DPTH_.py|urls.py|wsgi.py|db.sqlite3|airline4_tests_.py|apps.py|migrations|views.py|models.py|flights.csv|manage.py|wait-for-it.sh|admin.py|.git|.travis_DPTH_.yml|__pycache__"

.
├── .travis.yml
├── .travis_DPDTH_.yml
├── .travis_old.yml
├── Dockerfile
├── Procfile
├── airline
│   └── settings.py
├── docker-compose.yml
├── docker_push.sh
├── flights
│   └── tests.py
├── heroku-container-release.sh
└── requirements.txt

As shown in the tree above, the 9 files that matter in the workflow:

1) tests.py

2) settings.py

3) requirements.txt

4) Dockerfile

5) docker-compose.yml

6) .travis.yml

7) docker_push.sh

8) heroku-container-release.sh

9) Procfile

We will look at the contents of each of the 9 files in the sections below.


tests.py

This is the test file that Travis will use for testing the app. You write whatever test you want for Travis to run with.

from django.db.models import Max
from django.test import Client, TestCase

from .models import Airport, Flight, Passenger

# Create your tests here.
class FlightsTestCase(TestCase):

    def setUp(self):

        # Create airports.
        a1 = Airport.objects.create(code="AAA", city="City A")
        a2 = Airport.objects.create(code="BBB", city="City B")

        # Create flights.
        Flight.objects.create(origin=a1, destination=a2, duration=100)
        Flight.objects.create(origin=a1, destination=a1, duration=200)
        Flight.objects.create(origin=a2, destination=a1, duration=300)

    # 1
    def test_departures_count(self):
        a = Airport.objects.get(code="AAA")
        self.assertEqual(a.departures.count(), 2)

    # 2
    def test_arrivals_count(self):
        a = Airport.objects.get(code="AAA")
        self.assertEqual(a.arrivals.count(), 2)

    # 3
    def test_valid_flight(self):
        a1 = Airport.objects.get(code="AAA")
        a2 = Airport.objects.get(code="BBB")
        f = Flight.objects.get(origin=a1, destination=a2)
        self.assertTrue(f.is_valid_flight())

settings.py

Under the database section, the DATABASES['default'] sets the default database so the default database is the one connected by the OS environment variable DATABASE_URL, however if this is unavailable, we’ll use the one defined with 'HOST': 'db'.

This setup allows us to use 'HOST': 'db' which is the Docker’s copy of postgres during development phase & also during tesing phase with Travis while using the Heroku’s copy during deployment which is provided by connecting to the DATABASE_URL.

The DATABASE_URL, as an OS environment variable which is generated by Heroku after a Database is added to the web app in the Heroku website.

Add and/or edit the following to the settings.py file:

import django_heroku
import dj_database_url
MIDDLEWARE = [
    'django.middleware.security.SecurityMiddleware',

    'whitenoise.middleware.WhiteNoiseMiddleware',  # new

    'django.contrib.sessions.middleware.SessionMiddleware',
    'django.middleware.common.CommonMiddleware',
    'django.middleware.csrf.CsrfViewMiddleware',
    'django.contrib.auth.middleware.AuthenticationMiddleware',
    'django.contrib.messages.middleware.MessageMiddleware',
    'django.middleware.clickjacking.XFrameOptionsMiddleware',
]
DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.postgresql',
        'NAME': 'postgres',
        'USER': 'postgres',
        'PASSWORD': 'postgres',
        'HOST': 'db', # Docker's copy of postgres
        'PORT': 5432,
        #'PORT': 5433,
    }
}

DATABASE_URL = os.environ.get('DATABASE_URL')
db_from_env = dj_database_url.config(default=DATABASE_URL, conn_max_age=500, ssl_require=True)
DATABASES['default'].update(db_from_env)
STATIC_ROOT = os.path.join(BASE_DIR, 'staticfiles')
django_heroku.settings(locals())


requirements.txt

This file lets Docker & Travis know what packages are needed for the app. This is needed in Dockerfile & .travis.yml files.

django>=2.0.11
psycopg2
psycopg2-binary
dj-database-url==0.5.0
gunicorn
whitenoise
django-heroku
pytz
sqlparse

Dockerfile

This contains the instructions for building a Docker image.

The CMD gunicorn airline.wsgi:application --bind 0.0.0.0:$PORT tells Docker to use gunicorn as the web server. See here for details.

FROM python:3

WORKDIR /usr/src/app

ADD requirements.txt /usr/src/app

RUN pip install -r requirements.txt

ADD . /usr/src/app

# collect static files
RUN python manage.py collectstatic --noinput

CMD gunicorn airline.wsgi:application --bind 0.0.0.0:$PORT

docker-compose.yml

This contains the instructions on how to run a Docker containers which is an instance of a Docker image.

Notice the sleep delay introduced in the 2 command: sections. See here for details.

EDIT: The environment variables are added under db as newly required when testing on Travis.

version: '3'

services:
    db:
        image: postgres
        environment:
            - POSTGRES_DB=${POSTGRES_DB}
            - POSTGRES_USER=${POSTGRES_USER}
            - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}           
    migration:
        build: .
        command: bash -c 'while !</dev/tcp/db/5432; do sleep 1; done; python3 manage.py migrate'
        volumes:
            - .:/usr/src/app
        depends_on:
            - db
    web:
        build: .
#        container_name: webapp-dpdth
        image: webapp-dpdth
        command: bash -c 'while !</dev/tcp/db/5432; do sleep 1; done; python3 manage.py runserver 0.0.0.0:8000'
        volumes:
            - .:/usr/src/app
        ports:
            - "8000:8000"
        depends_on:
            - db
            - migration


.travis.yml

This file contains instructions for Travis. Notice that we’re not using the Postgres from Travis because we’re using Postgres from Docker directly. The postgresql is therefore commented out under the services: section.

Under script:, we ask Travis to run the test using Docker with the docker-compose command.

Under deploy:, we execute a docker_push.sh script, more details in the sections below.

The skip_cleanup: true tells Travis not to remove any files that it deems unnecessary after deployment. Travis does not have permission to do that on Heroku anyway.

language: python
python:
    - 3.6
services:
    - docker
#    - postgresql
install:
    - pip install -r requirements.txt
script:
    - docker-compose run web python manage.py test
deploy:
    provider: script
    script: bash docker_push.sh
    skip_cleanup: true
    on:
        branch: master

Workflow for after testing with Travis to deployment to Heroku

The main workflow after testing to deployment is as such:

tag image -> push image to registry -> release image

We’ll see how to do that in the following script files:

1) docker_push.sh

2) heroku-container-release.sh


docker_push.sh

This file does several things listed as follows:

1) Login to the Heroku’s image registry.

2) tag the source image webapp-dpdth:latest to the target image registry.heroku.com/webapp-dpdth/web. Replace webapp-dpdth with your app name on Heroku.

3) Push the target image to Heroku’s registry if the branch tested on Travis is a master branch & that it’s not a PR.

4) Change ownership & permission of files to allow Travis to execute the heroku-container-release.sh script.

#!/bin/bash

sudo docker login --username $HEROKU_DOCKER_USERNAME --password $HEROKU_AUTH_TOKEN registry.heroku.com
sudo docker tag webapp-dpdth:latest registry.heroku.com/webapp-dpdth/web
if [ $TRAVIS_BRANCH == "master" ] && [ $TRAVIS_PULL_REQUEST == "false" ]; then sudo docker push registry.heroku.com/webapp-dpdth/web; fi

chmod +x heroku-container-release.sh
sudo chown $USER:docker ~/.docker
sudo chown $USER:docker ~/.docker/config.json
sudo chmod g+rw ~/.docker/config.json

./heroku-container-release.sh


heroku-container-release.sh

This file is for releasing a Docker image via Heroku’s API. Replace webapp-dpdth with your app name on Heroku.

#!/bin/bash
imageId=$(docker inspect registry.heroku.com/webapp-dpdth/web --format={{.Id}})
payload='{"updates":[{"type":"web","docker_image":"'"$imageId"'"}]}'
curl -n -X PATCH https://api.heroku.com/apps/webapp-dpdth/formation \
-d "$payload" \
-H "Content-Type: application/json" \
-H "Accept: application/vnd.heroku+json; version=3.docker-releases" \
-H "Authorization: Bearer $HEROKU_AUTH_TOKEN"

See here for details.


Procfile

This file is for Heroku. The command in the release: section will run after a Docker image is released. It will run the migrate command with --noinput option. Without running migrate, the database on Heroku may not function correctly.

It also tells Heroku to deploy the web app using Gunicorn as the production server.

Note that airline is the Django project name. It’s not the web app name in the Django project & is also not the web app name in Heroku.

release: python manage.py migrate --noinput
web: gunicorn airline.wsgi

The deployed web app

With the above files in place, push to Github & Travis will start testing. After all tests passed, deployment starts. If there isn’t any failures, the web app will be running on:

https://webapp-dpdth.herokuapp.com

This link brings you to the admin page. It is using the Heroku’s copy of Postgres.

This link brings you to my built log in Travis.com which shows how a successful test/deploy built looks like.


Web security:

Please note that web security has not been throughly consider in this basic workflow describe above. Do NOT simply use the above workflow for production.

For example the SECRET_KEY in the settings.py isn’t dealt with at all and web security is really beyond the scope of this post.



2020

PBT for MARL

46 minute read

My attempt to implement a water down version of PBT (Population based training) for MARL (Multi-agent reinforcement learning).

Back to top ↑

2019

.bash_profile for Mac

13 minute read

This post demonstrates how to create customized functions to bundle commands in a .bash_profile file on Mac.

DPPO distributed tensorflow

68 minute read

This post documents my implementation of the Distributed Proximal Policy Optimization (Distributed PPO or DPPO) algorithm. (Distributed continuous version)

A3C distributed tensorflow

26 minute read

This post documents my implementation of the A3C (Asynchronous Advantage Actor Critic) algorithm (Distributed discrete version).

Distributed Tensorflow

76 minute read

This post demonstrates a simple usage example of distributed Tensorflow with Python multiprocessing package.

N-step targets

76 minute read

This post documents my implementation of the N-step Q-values estimation algorithm.

Dueling DDQN with PER

49 minute read

This post documents my implementation of the Dueling Double Deep Q Network with Priority Experience Replay (Duel DDQN with PER) algorithm.

Dueling DDQN

24 minute read

This post documents my implementation of the Dueling Double Deep Q Network (Dueling DDQN) algorithm.

DDQN

29 minute read

This post documents my implementation of the Double Deep Q Network (DDQN) algorithm.

DQN

24 minute read

This post documents my implementation of the Deep Q Network (DQN) algorithm.

Back to top ↑