ubuntu-python

A base Docker container for running Python apps with an Ubuntu userland, based on alpine-python and build for three different architectures:

• amd64
• arm32v7 (Pi 2+ or other ARM 32-bit boards like the ODROID U2+)
• arm64v8 (Pi 4 or other ARM 64-bit boards, as well as the WSL images for Windows on ARM)

Why

alpine-python makes for awesome small containers, but it's a pain to deal with all the binary wheels related to machine/deep learning stuff like Tensorflow, numpy, etc., so I decided to bite the bullet and take on the extra bloat that comes with an Ubuntu distro.

Status & Roadmap

• [ ] Multi-stage, "stripped" builds (requires some tuning of the onbuild images)
• [x] Multi-arch builds
• [ ] CI builds (WIP due to long build times - Travis errors out)
• [x] Python 3.8.5
• [x] Python 3.7.5
• [x] Python 3.7.3
• [x] Normalize architecture tags (arm32v7, amd64, etc. to match Docker naming conventions)
• [x] Python 3.7.0
• [x] Move to Ubuntu 18.04 base image
• [x] LTO (experimental) optimizations, inspired by revsys
• [x] Initial armhf containers
• [x] Initial amd64 containers, plain + onbuild
• [x] Python 3.6.3 (amd64)
• [x] Scaffolding for multiarch builds
• [x] Base userland with required libraries for building Python 3.6

Usage

This image runs the python command on docker run. You can either specify your own command, e.g:

docker run --rm -ti rcarmo/ubuntu-python python hello.py

Or extend this image using your custom Dockerfile, e.g:

FROM rcarmo/ubuntu-python:onbuild
# or FROM rcarmo/ubuntu-python:3.8-onbuild-amd64 if you prefer to specify your architecture

# for a flask server
EXPOSE 5000
CMD python manage.py runserver

Dont' forget to build your image:

docker build --rm=true -t rcarmo/app .

You can also access bash inside the container:

docker run --rm -ti rcarmo/ubuntu-python /bin/bash

Another option is to build an extended Dockerfile version (like shown above), and mount your application inside the container:

docker run --rm -v "$(pwd)":/home/app -w /home/app -p 5000:5000 -ti rcarmo/app

Details

• Builds the Python interpreter with optimizations enabled, for a significant performance boost
• You can use the latest or onbuild tags, but every variant is tagged with its architecture and build step purpose
• Uses make altinstall to have Python 3.8 coexist with the built-in Ubuntu Python (which is nearly impossible to eradicate anyway)
• Just like the main python docker image, it creates useful symlinks that are expected to exist, e.g. python3.8 > python, pip3.8 > pip, etc.)