Where to run collectstatic when deploying django app to heroku using docker?

Question

I am deploying a Django app to Heroku using Docker. When I put RUN manage.py collectstatic --noinput in the Dockerfile, it fails because there is no value set for the environment variable DJANGO_SECRET_KEY. My understanding is that this is because config vars aren't available during build time.

When I run collectstatic as a release command, it works without error, and successfully copies the static files. However, when I hit the app url, it returns a 500 error because the static files can't be found. I believe this is because the release command is run as a dyno on an ephemeral filesystem, and the copied files are therefore not found.

It seems to be a catch-22. Putting collectstatic in the Dockerfile fails because there are no config variables available, but putting it as a release command fails because only file changes from the build phase are saved?

What to do?

Here are my collectstatic settings in settings.py

MIDDLEWARE = [
    'django.middleware.security.SecurityMiddleware',
    'whitenoise.middleware.WhiteNoiseMiddleware',
    ...
]
...
STATIC_ROOT = os.path.join(BASE_DIR, 'staticfiles')
STATIC_URL = '/static/'
STATICFILES_DIRS = (
    os.path.join(BASE_DIR, 'static'),
)
STATICFILES_STORAGE = 'backend.storage.WhiteNoiseStaticFilesStorage'

Dockerfile

# Pull base image
FROM python:3.7-slim

# Set environment varibles
ENV PYTHONDONTWRITEBYTECODE 1
ENV PYTHONUNBUFFERED 1

# Set work directory
RUN mkdir /code
WORKDIR /code

# Install dependencies
RUN pip install pipenv
COPY Pipfile Pipfile.lock /code/
RUN pipenv install --system

# Copy project
COPY . /code/

## collect static files
RUN mkdir backend/staticfiles

# This fails because DJANGO_SECRET_KEY can't be empty
RUN python manage.py --noinput

heroku.yml

build:
  docker:
    web: Dockerfile
run:
  web: gunicorn backend.config.wsgi:application --bind 0.0.0.0:$PORT

Answer 1

After confirming with Heroku support, this does indeed appear to be a bit of a catch-22.

The solution was to put collectstatic in the Dockerfile so that it runs during build time and the files persist.

We got around not having a secret key config var by setting a default secret key using the get_random_secret_key function from Django.

The run phase uses the secret key from the Heroku config vars, so we aren't actually changing the secret key every time -- the default only applies to the build process. collectstatic doesn't index on the secret key, so this is fine.

In settings.py

from django.core.management.utils import get_random_secret_key ... SECRET_KEY = os.getenv('DJANGO_SECRET_KEY', default=get_random_secret_key())  

Answer 2

I don't use heroku so can't test, but you should be able to run collect static before you run the app;

Dockerfile

# Pull base image FROM python:3.7-slim  # Set environment varibles ENV PYTHONDONTWRITEBYTECODE 1 ENV PYTHONUNBUFFERED 1  # Set work directory WORKDIR /code/  # Install dependencies RUN pip install pipenv COPY Pipfile Pipfile.lock . RUN pipenv install --system  # Copy project COPY . .  # Collect static files RUN python manage.py collectstatic --noinput  # run gunicorn CMD gunicorn hello_django.wsgi:application --bind 0.0.0.0:$PORT 

You could also not run collectstatic in your dockerfile, or event run the application because these can be ran by heroku.yml, for example;

build:   docker:     web: Dockerfile   config:     DJANGO_SETTINGS_MODULE: project.settings run:   web: gunicorn backend.config.wsgi:application --bind 0.0.0.0:$PORT release:   image: web   command:     - python manage.py collectstatic --noinput 

You also shouldn't need to mkdir for your working directory. Just set WORKDIR /code/ early in your dockerfile and after that things will run based on that directory.

There's a decent article on this here; https://testdriven.io/blog/deploying-django-to-heroku-with-docker/