yuribogomolov / skaryna

Spark SQL extension that improves performance

Welcome to Project Skaryna

Here I'm going to experiment with new features of Apache Spark, and document all results/outcomes. I picked GitHub format for two simple reasons:

1. Keep history of code changes, and provide reproducible examples
2. Provide an easy way to reuse created utility functions.

Spark SQL extensions

This section is focused on Spark SQL extensions. A more powerful version of this feature was released in Spark 2.2, but I couldn't find any documentation that cover the details, and provide examples.

Let's take a look at the Catalyst architecture:

Spark SQL has two interfaces:

1. DataFrame API
2. SQL queries

and both of them are transformed into a logical plan. Let's list the main phases:

1. Logical plan analyzis
2. Logical plan optimization
3. Conversion into a Physical Plan (SparkPlan)
4. Conversion into RDD operations

Spark extensions allow to customize all major phases:

skaryna's Issues

Images for README