PySpark — Python on Apache Spark¶

PySpark is the Python API (frontend) of Apache Spark.

How It Works¶

When a Python script is executed using spark-submit shell script (Spark Core), PythonRunner is started (and --verbose option can show it as Main class).

$ ./bin/spark-submit --version hello_pyspark.py
Using properties file: null
Parsed arguments:
  master                  local[*]
  ...
  primaryResource         file:/Users/jacek/dev/oss/spark/hello_pyspark.py
  name                    hello_pyspark.py
...
Main class:
org.apache.spark.deploy.PythonRunner
Arguments:
file:/Users/jacek/dev/oss/spark/hello_pyspark.py
null
Spark config:
(spark.app.name,hello_pyspark.py)
(spark.app.submitTime,1684188276759)
(spark.master,local[*])
(spark.submit.deployMode,client)
(spark.submit.pyFiles,)
...

spark-submit execution above could be translated to the following:

./bin/spark-class org.apache.spark.deploy.PythonRunner hello_pyspark.py ""

PythonRunner then launches a Py4JServer (on a py4j-gateway-init daemon thread) and waits until it is started.

Finally, PythonRunner launches a Python process (to run the Python script) and waits until the process finishes (successfully or not).

$ ps -o pid,command | grep python3 | grep -v grep
12607 python3 /Users/jacek/dev/oss/spark/hello_pyspark.py

sudo lsof -p [pid of the python process]

Python 3.8 and Later¶

The minimum version of Python is 3.8.

shell.py¶

pyspark shell defines PYTHONSTARTUP environment variable to execute shell.py before the first prompt is displayed in Python interactive mode.

Py4J¶

java_gateway uses Py4J - A Bridge between Python and Java:

Py4J enables Python programs running in a Python interpreter to dynamically access Java objects in a Java Virtual Machine. Methods are called as if the Java objects resided in the Python interpreter and Java collections can be accessed through standard Python collection methods. Py4J also enables Java programs to call back Python objects.

pyspark.sql Package¶

pyspark.sql is a Python package for Spark SQL.

from pyspark.sql import *

__init__.py¶

The __init__.py files are required to make Python treat directories containing the file as packages.

Per 6.4.1. Importing * From a Package:

The import statement uses the following convention: if a package's __init__.py code defines a list named __all__, it is taken to be the list of module names that should be imported when from package import * is encountered.

Per Public and Internal Interfaces in PEP 8 -- Style Guide for Python Code:

To better support introspection, modules should explicitly declare the names in their public API using the __all__ attribute.

From python/pyspark/sql/__init__.py:

__all__ = [
    'SparkSession', 'SQLContext', 'HiveContext', 'UDFRegistration',
    'DataFrame', 'GroupedData', 'Column', 'Catalog', 'Row',
    'DataFrameNaFunctions', 'DataFrameStatFunctions', 'Window', 'WindowSpec',
    'DataFrameReader', 'DataFrameWriter', 'PandasCogroupedOps'
]

pandas¶

The minimum version of Pandas is 0.23.2 (and PandasConversionMixin asserts that).

import pandas as pd

pyarrow¶

The minimum version of PyArrow is 1.0.0 (and PandasConversionMixin asserts that).

import pyarrow

Python Mixins¶

From 8.7. Class definitions:

classdef ::= [decorators] "class" classname [inheritance] ":" suite

The inheritance list usually gives a list of base classes

PySpark uses mixins:

• PandasConversionMixin
• PandasMapOpsMixin
• SparkConversionMixin