I say

This is a highly opinionated example of how to manage dependencies of a python application and how to package it in a small alpine-based Docker image.

How do I use this?

Copy the Dockerfile into the root of your python project folder (next to your pyproject.toml) and run:

docker build -t myimagetag .

Features

Small Docker image

The application can be packaged into a slim (< 80 MB overhead) alpine-based Docker image by simply running:

docker build -t isay .

in the root folder.

Python dependency management using poetry

If your app is using poetry as the dependency management system, you are good to go. Check out the docs at https://python-poetry.org/docs/.

System dependencies

Some python packages (including, maybe, your own) may have system dependencies like openssl or gcc either in order to be compiled or in order to run. You can define these dependencies in the [tools.sys.alpine3-11.dev-dependencies] and [tools.sys.alpine3-11.dependencies] sub-sections of your pyproject.toml file. They each include the list of system dependencies along with their exact versions. The dev-dependencies are the ones needed for building wheels for the dependencies (and for the project) while the dependencies are needed to run the application.

The alpine3-11 part in the section name has to match the version of alpine linux that the python image is based on. The second line in the Dockerfile reads:

FROM python:3.8-alpine3.11 as base

The alpine3.11 part in the Dockerfile must match the alpine3-11 part in pyproject.toml (with the . replaced by a - due to restrictions on section names in toml).

If you'd like to use another version of alpine for your project, change that in the second line of the Dockerfile. If you rely on any system dependencies, refer to the section below.

Cache-friendly Docker builds

Whether you are running the build on your development machine or on a continuous integration (CI) build agent, the build process will try to make use of Docker's caching mechanism by building layers that depend on data likely to change less frequently ahead of the layers that depend on data that is likely to change more frequently. Here is the sequence:

1. Python version: Dockerfile
2. Dev & runtime dependencies (system and python): pyproject.toml, poetry.lock
3. Application installed in both the dev and runtime virtual environments: app source code