pypakr

Python containers for virtualenv

This Python project implements simple containers in the sense similar to Docker: you can generate an image file that contains complete information needed to create a virtual environment on any system that has pypakr installed. NOTE: don't confuse this with the other meaning of the term container, which is "a holder object that stores a collection of other objects (its elements)." [http://www.cplusplus.com/reference/stl/]

Introduction

Just like Docker can run isolated environments, called containers, Python has virtualenv. This tool creates isolated Python environments. But it misses what Docker has, which is a way to have a file that somehow defines the environment and that can be used to create a virtual environment on any machine that has Docker installed.

In Docker, containers are created from "images" that specify their precise contents (source: Wikipedia). We need something like that for Python.

pypakr fulfills that need. It can take a description of which Python packages need to be installed and use it to generate an image file that describes what is installed in a virtual environment. Then it can take that image and generate a directory that contains a complete virtual environment.

This can be done with Docker too. The advantage of pypakr is that the images are much smaller - they contain only Python files that can be overlaid on top of a generic virtual environment to produce a "container" that can run a Python application or microservice. Containers are also smaller than a typical Docker container, because they only contain a virtualenv-style environment.

Dependencies

pypakr currently runs on Linux, and other than Python, it depends on the following software:

• virtualenv
• vex, for running code in a virtual environment from a script
• unionfs-fuse, for mounting UnionFS volumes

Install these tools like this:

pip install virtualenv
pip install vex

On Ubuntu, install unionfs-fuse like this:

sudo apt-get install unionfs-fuse

Testing

pypakr can be tested before installing by running

tox

or after installing by running

pytest

Installation

From source: run

pip install pypakr

From PYPI: run

pip install pypakr-pkg

Initialization

After you installed pypakr, run command

pypakr init

It will create configuration file .pypakr in your home directory and also create a base virtual environment in directory

$HOME/pypakr/BASE

By default, if .pypakr does not exist before you run pypakr init, it will be created to look like this:

[Global]
base = /home/george/pypakr/BASE

where /home/george stands for your home directory.

However, if you don't like this default, create .pypakr and set value of key base to point to some other path. Then run pypakr init, and it will try to create the base directory if it doesn't exist and then create a virtual environment in it.

Help

Run pypakr by itself on the command line to get help, or

pypakr help

It will give you text like this:

pypakr
Python containers
Usage:
pypakr <command> <parameters>
  - Commands:
    - init              - initialize
    - create-image      - create image
       - Arguments:
         -s, --src <source-file>
         -i, --image <image-file>
    - create-container  - create container
       - Arguments:
         -i, --image <image-file>
         -c, --container <container-directory>
    - run               - run container (execute script in the
                          container's virtual environment)
       - Arguments:
         -c, --container <container-directory>
         -r, --script <script-to-execute>

Configuration is in file ~/.pypakr
[Global]
base = /home/george/pypakr/BASE

See documentation in directory doc. As a quick reminder, here are the steps to create a container:

1. Create source file CUSTOM.tar.

2. Create image file IMAGE.tar:

   pypakr create-image -s CUSTOM.tar -i IMAGE.tar

3. Create container CONT:

   pypakr create-container -i IMAGE.tar -c CONT

4. Run script run in the container CONT:

   pypakr run -c CONT -r ./run

TODO

This project is still young and there is so much more that can be done. Here are some ideas:

• Port to Windows. unionfs is a Linux thing; a Windows version of pypakr is possible, by generating a custom virtual environment and then extracting the difference from the base virtual environment and tarring that to create an image.
• A hub site like Docker Hub.
• Serverless web application that uses pypakr containers as units of functionality.

Contributions

All comments, questions, issue reports, and pull requests are welcome!