congruity

In many ways, the migration from using classic Spark applications using the full power and flexibility to be using only the Spark Connect compatible DataFrame API can be challenging.

The goal of this library is to provide a compatibility layer that makes it easier to adopt Spark Connect. The library is designed to be simply imported in your application and will then monkey-patch the existing API to provide the legacy functionality.

Non-Goals

This library is not intended to be a long-term solution. The goal is to provide a compatibility layer that becomes obsolete over time. In addition, we do not aim to provide compatibility for all methods and features but only a select subset. Lastly, we do not aim to achieve the same performance as using some of the native RDD APIs.

Usage

Spark JVM & Spark Connect compatibility library.

pip install spark-congruity

import congruity.patch

Example

Here is code that works on Spark JVM:

from pyspark.sql import SparkSession

spark = SparkSession.builder.remote("sc://localhost").getOrCreate()
data = [("Java", "20000"), ("Python", "100000"), ("Scala", "3000")]
spark.sparkContext.parallelize(data).toDF()

This code doesn't work with Spark Connect. The congruity library rearranges the code under the hood, so the old syntax works on Spark Connect clusters as well:

import congruity.patch  # noqa: F401
from pyspark.sql import SparkSession

spark = SparkSession.builder.remote("sc://localhost").getOrCreate()
data = [("Java", "20000"), ("Python", "100000"), ("Scala", "3000")]
spark.sparkContext.parallelize(data).toDF()

Contributing

We very much welcome contributions to this project. The easiest way to start is to pick any of the below RDD or SparkContext methods and implement the compatibility layer. Once you have done that open a pull request and we will review it.

What's supported?

RDD

| RDD | API | Comment | |-----------------------------------|--------------------|-------------------------------------------------------------------| | aggregate | :white_check_mark: | | | aggregateByKey | :x: | | | barrier | :x: | | | cache | :x: | | | cartesian | :x: | | | checkpoint | :x: | | | cleanShuffleDependencies | :x: | | | coalesce | :x: | | | cogroup | :x: | | | collect | :white_check_mark: | | | collectAsMap | :white_check_mark: | | | collectWithJobGroup | :x: | | | combineByKey | :x: | | | count | :white_check_mark: | | | countApprox | :x: | | | countByKey | :x: | | | countByValue | :x: | | | distinct | :x: | | | filter | :white_check_mark: | | | first | :white_check_mark: | | | flatMap | :x: | | | fold | :white_check_mark: | First version | | foreach | :x: | | | foreachPartition | :x: | | | fullOuterJoin | :x: | | | getCheckpointFile | :x: | | | getNumPartitions | :x: | | | getResourceProfile | :x: | | | getStorageLevel | :x: | | | glom | :white_check_mark: | | | groupBy | :white_check_mark: | | | groupByKey | :white_check_mark: | | | groupWith | :x: | | | histogram | :white_check_mark: | | | id | :x: | | | intersection | :x: | | | isCheckpointed | :x: | | | isEmpty | :white_check_mark: | | | isLocallyCheckpointed | :x: | | | join | :x: | | | keyBy | :white_check_mark: | | | keys | :white_check_mark: | | | leftOuterJoin | :x: | | | localCheckpoint | :x: | | | lookup | :x: | | | map | :white_check_mark: | | | mapPartitions | :white_check_mark: | First version, based on mapInArrow. | | mapPartitionsWithIndex | :x: | | | mapPartitionsWithSplit | :x: | | | mapValues | :white_check_mark: | | | max | :white_check_mark: | | | mean | :white_check_mark: | | | meanApprox | :x: | | | min | :white_check_mark: | | | name | :x: | | | partitionBy | :x: | | | persist | :x: | | | pipe | :x: | | | rand