Skip to content

Calling Compiled Scala Code from Python using PySpark

Calling compiled Scala code inside the JVM from Python using PySpark

There is no doubt that Java and Scala are the de-facto languages for Data Engineering, whilst Python is certainly the front runner for language of choice with Data Scientists. Spark; a framework for distributed data analytics is written in Scala but allows for usage in Python, R and Java. Interoperability between Java and Scala is a no briner since Scala compiles down to Java byte code, but call Scala from Python is a little more involved, but the process is very simple.

The code used in this post builds upon the code used in a previous post and has the standard maven directory layout. To have a closer look, it can be found under code/posts/2020-05-20-Scala-Python-JVM

We will be calling simple.SimpleApp.hello() function to print "Hello, World!".

The simple Scala we will use is the following:

$ cat -n src/main/scala/simple/SimpleApp.scala

 1  package simple;
 3  object SimpleApp {
 4    def hello(): Unit = {
 5      println("Hello, Wolrd")
 6    }
 7  }

This will then be compiled and packaged using sbt to created a .jar file that can be included in the running JVM instance when launching Spark. Thus, after running:

$ sbt clean compile package

[info] Loading settings for project simpleapp-build from plugins.sbt ...
[info] Loading project definition from /Users/tallamjr/www/blog/code/posts/2020-05-20-Scala-Python-JVM/simpleApp/project
[info] Loading settings for project simpleapp from build.sbt ...
[info] Set current project to SimpleApp (in build file:/Users/tallamjr/www/blog/code/posts/2020-05-20-Scala-Python-JVM/simpleApp/)
[info] Executing in batch mode. For better performance use sbt's shell
[success] Total time: 0 s, completed 21-May-2020 13:18:19
[info] Compiling 1 Scala source to /Users/tallamjr/www/blog/code/posts/2020-05-20-Scala-Python-JVM/simpleApp/target/scala-2.12/classes ...
[success] Total time: 7 s, completed 21-May-2020 13:18:25
[success] Total time: 0 s, completed 21-May-2020 13:18:26

We obtain target/scala-2.12/simpleapp_2.12-1.0.jar which is supplied to Spark like so:

$ spark-submit --driver-class-path target/scala-2.12/simpleapp_2.12-1.0.jar simpleSpark/

simpleSpark/ is the where the pyspark code lives that will be calling the Scala function, let's have a look into that file:

from pyspark.sql import SparkSession

spark = SparkSession.builder \
    .master("local[*]") \
    .appName("simpleSpark") \

sc = spark.sparkContext

def logLevel(sc):
    # REF:
    log4jLogger =
    log = log4jLogger.LogManager.getLogger(__name__)
    log.warn("Custom WARN message")


print(spark.range(5000).where("id > 500").selectExpr("sum(id)").collect())


There is a bit of boilerplate to get the session started and some customisation with logging going on but the key lines of code are:

 8  sc = spark.sparkContext
25  sc._jvm.simple.SimpleApp.hello()

The resulting output after running spark-submit is:

$ spark-submit --driver-class-path target/scala-2.12/simpleapp_2.12-1.0.jar simpleSpark/

20/05/21 13:26:02 WARN Utils: Your hostname, Tareks-MacBook-Pro.local resolves to a loopback address:; using instead (on interface en0)
20/05/21 13:26:02 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address
20/05/21 13:26:02 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Hello, Wolrd!


This post was inspired by Alexis Seigneurin's much more detailed post Spark - Calling Scala code from PySpark which I highly recommend reading.