Add Python bindings for JTS + GeoMesa Spark UDFs

Add Python bindings for JTS + GeoMesa Spark UDFs

RasterFrames has examples of how to register UDFs with the PySpark context.

https://github.com/locationtech/rasterframes/blob/43bd3b37f0b2931470b2d5b5757474d3c885e659/pyrasterframes/src/main/python/pyrasterframes/rasterfunctions.py#L718-L1062

There may be other Python packaging tricks that the RasterFrames team has sorted out.

Original JIRA Issue: https://geomesa.atlassian.net/browse/GEOMESA-2766

Key: GEOMESA-2766
Type: Improvement
Priority: Major
Status: To Do
Resolution: Unresolved
Reporter: James Hughes
Created: Fri, 10 Jan 2020 10:29:20 -0500
Updated: Fri, 10 Jan 2020 10:35:11 -0500

Comments:

557058:f542d9ac-6efc-4b69-93fb-160ff09c0414 (Fri, 10 Jan 2020 10:31:58 -0500):

This is the easy part. Creating a self-contained Python .whl file that navigates the nine levels of pip hell is the hard part.

557058:878133e6-01b0-4022-8f67-5e77d62963f5 (Fri, 10 Jan 2020 10:35:11 -0500):

Of interest from the RasterFrames side of the house is

• Making sure any SparkSession / SparkContext modifications made by either package are copacetic. Bonus points if we can also include geopyspark in this mix. Extra bonus points for documenting useful patterns
• Packaging / distribution as Simeon Fitch points out. Currently RasterFrames is oriented toward producing a publishable whl. But GeoMesa is not
• Separation of concerns. Defining GeoMesa python functions and python UDTs within RasterFrames is madness.

Start by reviewing RasterFrames' pyrasterframes/rasterfunctions.py, especially lines 718-1062, and its PySpark UDF registration approach. Compare that with GeoMesa's Spark UDF and Python packaging structure; done means Python bindings are distributable, coexist cleanly with SparkSession or SparkContext changes, and include the useful usage patterns described in the issue.