Overview

The idea behind this module is to support self-updating Python applications, namely command-line tools. Transmute probes remote repositories for updated components, fetches updates and adds them to sys.path, making them available for use in the application.

Components are assumed to be standard Python Eggs. Repositories are then simple containers for these eggs. Currently PyPI and S3 “folders” are supported as repositories. (Mostly for testing purposes, local directories are also supported as repositories).

Under the hood, pkg_resources (from setuptools) is used to parse and fulfill requirements, based on listings of available eggs obtained from each repository. Once updated packages are made available, their modules can be imported or the application can be re-launched with an adjusted environment to pick up updated modules.

The application writer controls the packages to update, repositories each package will be grabbed from, and when an update is actioned.

The philosophy has been that an absent or flaky network should not prevent (or significantly delay) an application from running on top of outdated packages, if they’ve been cached locally. That said, there are currently no provisions for testing and verifying a successful update or rolling back a failed update.

Example

This is a script that requests the ‘hello’ package to be updated from the dist folder in the current working directory:

import transmute

transmute.require([ 'hello' ], sources=[ 'dist' ])
transmute.update()

import hello
hello.greet('world')

Bootstrapping an application with bootstrap.py

The submodule in `transmute/bootstrap.py <https://github.com/comoyo/python-transmute/blob/master/transmute/bootstrap.py>`__ can be used on its own to bootstrap other Python modules and applications. It is capable of downloading packages from PyPI. In this way, transmute itself can be loaded and further used to download additional packages.

The script provides a bunch of hooks where users can place their code. In particular, main() can be filled in to fetch application specific packages and actually launch the application. At the point it is called transmute has been added to sys.path (after downloading from PyPI, if needed).

def main():
    import transmute
    import transmute.s3

    transmute.require([ 'foobar' ],
            sources=[ 's3://foobar-repository/eggs/foobar' ])
    transmute.update()

    import foobar.cli
    return foobar.cli.run()

It can also be used as a placeholder for a Python module. If the module itself is available from PyPI, the corresponding package name would be added to the requirements variable. For other use cases, I sense a pull request coming :-)

At the time of this writing, the only non-standard dependency of the script is the pkg_resources module from the setuptools package. The assumption here is that the module is more or less available everywhere. If this turns out to be a problem in practice, I suppose the script could be simplified to not require it.

Supported package formats

Currently, only standard Python eggs are supported. I don’t mind adding support for other formats, formats supported natively by Python are preferred.

In that regard, source tarballs look particularly interesting for pure Python packages, and seem to be more generally available from PyPI. Unpacking and importing the packages locally could be a way forward.

Python wheels also look interesting and gaining some traction.

Supported repositories

transmute.require([ 'foobar' ], sources=[ '/opt/basket' ])

transmute.require([ 'foobar' ], sources=[ transmute.PYPI_SOURCE ])

- AWS_CREDENTIAL_FILE
- AWS_ACCESS_KEY, AWS_SECRET_KEY, and (optionally) AWS_SECURITY_TOKEN
- AWS_DEFAULT_REGION and EC2_REGION

import transmute.s3
transmute.require([ 'foobar' ], sources=[ 's3://bucket/key-prefix' ])

Open issues

• Logging is sorely missing. This can be helpful in debugging, but also to keep track of updates and possibly tie in to enabling rollbacks.

• Rolling back a b0rked update.

• Provide hooks for verifying an update before activating it.

• We shouldn’t use the network on every run of a given command. Keeping track of metadata about repository queries would allow us to limit updates to daily or weekly schedules.

• Currently MD5 hashes are used to verify integrity of downloaded packages, as advertised by repositories. It would be nice to be able to verify package signatures.

• Your pet peeve?