athena-udf 0.2

Athena User Defined Functions(UDFs) made easy!

athena-udf

Athena User Defined Functions(UDFs) in Python made easy!

This library implements the Athena UDF protocol in Python so you don't have to use Java and you can use any Python library you wish including numpy/pandas!

Installation

Install this library using pip:

pip install athena-udf

Usage

Simply install the package, create a lambda handler Python file, subclass BaseAthenaUDF and implement the handle_athena_record static method with your required functionality like this:

import athena_udf


class SimpleVarcharUDF(athena_udf.BaseAthenaUDF):

    @staticmethod
    def handle_athena_record(input_schema, output_schema, arguments):
        varchar = arguments[0]
        return varchar.lower()


lambda_handler = SimpleVarcharUDF().lambda_handler

This very basic example takes a varchar input, and returns the lowercase version.

varchar is converted to a python string on the way in and way out.

input_schema contains a PyArrow schema representing the schema of the data being passed

output_schema contains a PyArrow schema representing the schema of what athena expects to be returned.

arguments contains a list of arguments given to the function. Can be more than 1 with different types.

If you package the above into a zip, with dependencies and name your lambda function my-kambda you can then run it from the athena console like so:

USING EXTERNAL FUNCTION my_udf(col1 varchar) RETURNS varchar LAMBDA 'athena-test'

SELECT my_udf('FooBar');

Which will yield the result foobar

See other examples in the examples folder of this repo.

Important information before using.

Each lambda instance will take multiple requests for the same query. Each request can contain multiple rows, athena-udf handles this for you and your implementation will receive a single row.

Athena will group your data into around 1MB chunks in a single request. The maximum your function can return in 6MB per chunk.

This library uses PyArrow. This is a large library so the Lambdas will be around 50MB zipped.

Timestamps seem to be truncated into Python date objects missing the time.

Functions can return one value only. To return more complex data structures consider returning a JSON payload and parsing on athena.

Development

To contribute to this library, first checkout the code. Then create a new virtual environment:

cd athena-udf
python -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

pip install -e '.[test]'

To run the tests:

pytest