PySpark and broadcast join example

Question

I am using Spark 1.3

# Read from text file, parse it and then do some basic filtering to get   data1
data1.registerTempTable('data1')

# Read from text file, parse it and then do some basic filtering to get data1
data2.registerTempTable('data2')

# Perform join
data_joined = data1.join(data2, data1.id == data2.id);

My data is quite skewed and data2 (few KB) << data1 (10s of GB) and the performance is quite bad. I was reading about broadcast join, but not sure how I can do the same using Python API.

Answer 1

Spark 1.3 doesn't support broadcast joins using DataFrame. In Spark >= 1.5.0 you can use broadcast function to apply broadcast joins:

from pyspark.sql.functions import broadcast

data1.join(broadcast(data2), data1.id == data2.id)

For older versions the only option is to convert to RDD and apply the same logic as in other languages. Roughly something like this:

from pyspark.sql import Row
from pyspark.sql.types import StructType

# Create a dictionary where keys are join keys
# and values are lists of rows
data2_bd = sc.broadcast(
    data2.map(lambda r: (r.id, r)).groupByKey().collectAsMap())


# Define a new row with fields from both DFs
output_row = Row(*data1.columns + data2.columns)

# And an output schema
output_schema = StructType(data1.schema.fields + data2.schema.fields)

# Given row x, extract a list of corresponding rows from broadcast
# and output a list of merged rows
def gen_rows(x):
    return [output_row(*x + y) for y in data2_bd.value.get(x.id, [])]

# flatMap and create a new data frame
joined = data1.rdd.flatMap(lambda row: gen_rows(row)).toDF(output_schema)

Answer 2

This code is working in spark-2.0.2-bin-hadoop2.7 version

from pyspark.sql import SparkSession

from pyspark.sql.functions import broadcast

spark = SparkSession.builder.appName("Python Spark SQL basic 
example").config("spark.some.config.option", "some-value").getOrCreate()

df2 = spark.read.csv("D:\\trans_mar.txt",sep="^");

df1=spark.read.csv("D:\\trans_feb.txt",sep="^");

print(df1.join(broadcast(df2),df2._c77==df1._c77).take(10))