So it’s been a while, but lately I’ve been writing Python (and Erlang!) for my full-time job at Nextiva, where we have a microservice architecture mainly comprised of single-page apps, backed by Python applications accessed through an API gateway. And as you’d expect, we make heavy use of Docker and Docker Compose for local development¹. One thing that has come up a handful of times, however, is that an application’s Docker image builds just fine, and even runs by itself, but when mounting the application’s source code as a Docker Compose volume, the image’s Python fails to find the package.

For example, take this basic Dockerfile for a Python application my-app that uses a requirements.txt and setup.py:

FROM python:3-alpine
WORKDIR /app

COPY requirements.txt ./
RUN pip install --no-cache-dir -r requirements.txt

COPY . ./
RUN pip install -e .

CMD ["my-app"]

and an accompanying docker-compose.yml:

version: 3
services:
  app:
    build: .
    volumes:
      - .:/app

Try to start it, and:

docker-compose up
Traceback (most recent call last):
  File "/usr/local/bin/my-app", line 5, in <module>
    from pkg_resources import load_entry_point
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 2927, in <module>
    @_call_aside
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 2913, in _call_aside
    f(*args, **kwargs)
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 2940, in _initialize_master_working_set
    working_set = WorkingSet._build_master()
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 635, in _build_master
    ws.require(__requires__)
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 943, in require
    needed = self.resolve(parse_requirements(requirements))
  File "/usr/local/lib/python3.7/site-packages/pkg_resources/__init__.py", line 829, in resolve
    raise DistributionNotFound(req, requirers)
pkg_resources.DistributionNotFound: The 'my-app' distribution was not found and is required by the application

everything fails horribly.

What happens is that the RUN pip install -e . command in the Dockerfile builds the Python package for the application in /app, creating a /app/my-app.egg-info directory in the Docker image. But the copy of the source code on your host machine probably doesn’t have a .egg-info directory, so when you start the application with docker-compose up, your .egg-info-less copy of the source code directory replaces the image’s copy, leaving Python within the image unable to find it.

The solution² is twofold: first, add your applicaiton directory to the $PYTHONPATH environment variable in your Dockerfile:

ENV PYTHONPATH=/app

Then, replace your Dockerfile’s RUN pip install -e . command with:

RUN cd /usr/local/lib/python3.7/site-packages && \
    python /app/setup.py develop

This creates your app’s .egg-info in the Python site-packages directory, and tells the Docker image’s Python interpreter that it can find packages in the /app directory. As a result, the application will run both as a standalone Docker image in production, and in local development with a Docker Compose volume mount.