Because the Apache Python avro package is written in pure Python, it is relatively slow. In one test case, it takes about 14 seconds to iterate through a file of 10,000 records. By comparison, the JAVA avro SDK reads the same file in 1.9 seconds.

The fastavro library was written to offer performance comparable to the Java library. With regular CPython, fastavro uses C extensions which allow it to iterate the same 10,000 record file in 1.7 seconds. With PyPy, this drops to 1.5 seconds (to be fair, the JAVA benchmark is doing some extra JSON encoding/decoding).

fastavro supports the following Python versions:

• Python 2.7
• Python 3.4
• Python 3.5
• Python 3.6
• PyPy
• PyPy3

Documentation is available at http://fastavro.readthedocs.io/en/latest/

fastavro is available both on PyPi

pip install fastavro

and on conda-forge conda channel.

conda install -c conda-forge fastavro

As recommended by Cython, the C files output is distributed. This has the advantage that the end user does not need to have Cython installed. However it means that every time you change fastavro you need to run make.

We release both to pypi and to conda-forge.

We assume you have twine installed and that you've created your own fork of fastavro-feedstock.

• Make sure the tests pass
• Copy the windows build artifacts for the new version from https://ci.appveyor.com/project/scottbelden/fastavro to the dist folder
• Run make publish
• Note the sha signature emitted at the above
• Switch to feedstock directory and edit recipe/meta.yaml
  • Update version and sha256 variables at the top of the file
  • Run python recipe/test_recipe.py
  • Submit a PR

See the ChangeLog

Project Home