l3dlp-sandbox / minimal-docker-python-setup

A demo of a minimal Nginx-uWSGI-Flask-Redis stack using Docker in less than 42 mb.

This project is only for demontration purposes and will not be updated.

Minimal Docker Python setup

A demo of a minimal Nginx-uWSGI-Flask-Redis stack using Docker. The app counts amount of visits per IP and stores it in Redis before returning it to the client.

Setup consists of 3 containers:

• Nginx
• uWSGI, Flask app
• Redis

The whole setup runs with using less then 45 mb of disk space!

Quickstart

Create images with Nginx and Redis from the tarred filesystem:

user@host $ docker import - orangetux/nginx < nginx/rootfs.tar
user@host $ docker import - orangetux/redis < redis/rootfs.tar

Now build the other images and run them by using docker-compose:

user@host $ docker-compose build
user@host $ docker-compose up

And head over http://localhost:1337 and you should see an IP address.

uWSGI and Flask

uWSGI and the Flask app are running from a custom image based on advancedclimatesystems/python:2.7.10. The Dockerfile for this image can be found here. See 'Installing C extensions' for more information about how uWSGI is installed.

Nginx

Nginx runs from an image which is based on a small filesystem. This filesystem is built with Buildroot. You can build the filesystem using the supplied defconfig.

Redis

Like Nginx, Redis runs from an very small, custom crafted Docker image. To build the filesystem which this image is based on, use docker_redis_defconfig.

Installing C extensions

Wheel

Installation of Python dependencies is done in 2 steps. First the dependency is built. Compiling a C extension is one of the steps during this stage. After building the depedency is installed. This step does't do much more than copying files produced by the build phase to their correct location.

The Python image doesn't contain a C compiler. For pure Python dependencies this is not a problem, but C extensions can't be built inside the image. These dependencies must be built on the host before they can be installed inside the image.

Dependencies can be built using Wheel: a built-package format for Python. All required wheels are supplied and can be found at app/wheelhouse, but if you want to build the wheels yourself run the following command. It requires pip, wheel and setuptools >= 0.8.0.

user@host $ pip wheel --wheel-dir=wheelhouse flask flask-redis uwsgi

Installing depedencies from wheels is now simple:

root@container # pip install --find-links wheelhouse flask flask-redis uwsgi

Shared libraries

uWSGI relies on a few shared libraries which are not all available in the image. They need to be added. If you run uWSGI without supplying all missing shared libraries you'll see see something like this:

root@container # uwsgi
uwsgi: error while loading shared libraries: libpcre.so.3: cannot open shared object file: No such file or directory

If you add this library to the Docker image and run uwsgi you'll find out that another depedency is missing. The following shared libraries must be added before uWSGI is able to run:

• liblzma.so.5
• libpcre.so.3
• libxml2.so.2

These libraries are included in the repository and located at app/shared_libs. But they can be copied from any x86_64 machine to /lib/ in the image. To locate them, use ldd:

user@host $ ldd uwsgi | grep -e lzma -e xml -e pcre
libpcre.so.3 => /lib/x86_64-linux-gnu/libpcre.so.3 (0x00007f2ad8b91000)
libxml2.so.2 => /usr/lib/x86_64-linux-gnu/libxml2.so.2 (0x00007f2ad81d7000)
liblzma.so.5 => /lib/x86_64-linux-gnu/liblzma.so.5 (0x00007f2ad72ed000)

Demo

License

This software is licensed under the MIT license.

© 2015 Auke Willem Oosterhoff