Write and run Spark Scala jobs on Dataproc

This tutorial illustrates different ways to create and submit a Spark Scala job to a Dataproc cluster, including how to:

• write and compile a Spark Scala "Hello World" app on a local machine from the command line using the Scala REPL (Read-Evaluate-Print-Loop or interactive interpreter) or the SBT build tool
• package compiled Scala classes into a jar file with a manifest
• submit the Scala jar to a Spark job that runs on your Dataproc cluster
• examine Scala job output from the Google Cloud console

This tutorial also shows you how to:

• write and run a Spark Scala "WordCount" mapreduce job directly on a Dataproc cluster using the spark-shell REPL

• run pre-installed Apache Spark and Hadoop examples on a cluster

Note that although the command line examples in this tutorial assume a Linux terminal environment, many or most will also run as written in a macOS or Windows terminal window.

Set up a Google Cloud Platform project

If you haven't already done so:

1. Set up a project
2. Create a Cloud Storage bucket
3. Create a Dataproc cluster

Write and compile Scala code locally

As a simple exercise for this tutorial, write a "Hello World" Scala app using the Scala REPL or the SBT command line interface locally on your development machine.

• Use Scala
• Use SBT

Use Scala

1. Download the Scala binaries from the Scala Install page

2. Unpack the file, set the SCALA_HOME environment variable, and add it to your path, as shown in the Scala Install instructions. For example:

   export SCALA_HOME=/usr/local/share/scala
   export PATH=$PATH:$SCALA_HOME/
   

3. Launch the Scala REPL

   $ scala
   Welcome to Scala version ...
   Type in expressions to have them evaluated.
   Type :help for more information.
   scala>
   

4. Copy and paste HelloWorld code into Scala REPL

   object HelloWorld {
     def main(args: Array[String]): Unit = {
       println("Hello, world!")
     }
   }

5. Save HelloWorld.scala and exit the REPL

   scala> :save HelloWorld.scala
   scala> :q
   

6. Compile with scalac

   $ scalac HelloWorld.scala
   

7. List the compiled .class files

   $ ls HelloWorld*.class
   HelloWorld$.class   HelloWorld.class
   

Use SBT

1. Download SBT

2. Create a "HelloWorld" project, as shown below

   $ mkdir hello
   $ cd hello
   $ echo \
   'object HelloWorld {def main(args: Array[String]) = println("Hello, world!")}' > \
   HelloWorld.scala
   

3. Create an sbt.build config file to set the artifactName (the name of the jar file that you will generate, below) to "HelloWorld.jar" (see Modifying default artifacts)

   echo \
   'artifactName := { (sv: ScalaVersion, module: ModuleID, artifact: Artifact) =>
   "HelloWorld.jar" }' > \
   build.sbt
   

4. Launch SBT and run code

   $ sbt
   [info] Set current project to hello ...
   > run
   ... Compiling 1 Scala source to .../hello/target/scala-.../classes...
   ... Running HelloWorld
   Hello, world!
   [success] Total time: 3 s ...
   

5. Package code into a jar file with a manifest that specifies the main class entry point (HelloWorld), then exit

   > package
   ... Packaging .../hello/target/scala-.../HelloWorld.jar ...
   ... Done packaging.
   [success] Total time: ...
   > exit
   

Create a jar

Create a jar file with SBT or using the jar command.

Download Java? To run the jar command, you must have Java SE (Standard Edition) JRE (Java Runtime Environment) installed on your machine—see Java SE Downloads.

Create a jar with SBT

The SBT package command creates a jar file (see Use SBT).

Create a jar manually

1. Change directory (cd) into the directory that contains your compiled HelloWorld*.class files, then run the following command to package the class files into a jar with a manifest that specifies the main class entry point (HelloWorld).

   $ jar cvfe HelloWorld.jar HelloWorld HelloWorld*.class
   added manifest
   adding: HelloWorld$.class(in = 637) (out= 403)(deflated 36%)
   adding: HelloWorld.class(in = 586) (out= 482)(deflated 17%)
   

   Unpacking and examining the jar's manifest (MANIFEST.MF) shows that it lists the HelloWorld Main-Class entry point:

   Manifest-Version: ...
   Created-By: ...
   Main-Class: HelloWorld
   

Copy jar to Cloud Storage

1. Use the Google Cloud CLI to copy the jar to a Cloud Storage bucket in your project

When passing bucket names to the Google Cloud CLI, make sure to only specify the bucket name after the gs:// prefix. For example, if a project contains a "my-most-unique-bucket-name" bucket, as shown in the following Google Cloud console Storage Browser screenshot,

the following command will list the contents of that bucket:

gcloud storage ls gs://my-most-unique-bucket-name

$ gcloud storage cp HelloWorld.jar gs://<bucket-name>/
Copying file://HelloWorld.jar [Content-Type=application/java-archive]...
Uploading   gs://bucket-name/HelloWorld.jar:         1.46 KiB/1.46 KiB

Submit jar to a Dataproc Spark job

1. Use the Google Cloud console to submit the jar file to your Dataproc Spark job. Fill in the fields on the Submit a job page as follows:

   • Cluster: Select your cluster's name from the cluster list
   • Job type: Spark

   • Main class or jar: Specify the Cloud Storage URI path to your HelloWorld jar (gs://your-bucket-name/HelloWorld.jar).

     If your jar does not include a manifest that specifies the entry point to your code ("Main-Class: HelloWorld"), the "Main class or jar" field should state the name of your Main class ("HelloWorld"), and you should fill in the "Jar files" field with the URI path to your jar file (gs://your-bucket-name/HelloWorld.jar).

2. Click Submit to start the job. Once the job starts, it is added to the Jobs list.

3. Click the Job ID to open the Jobs page, where you can view the job's driver output.

Write and run Spark Scala code using the cluster's spark-shell REPL

You may want to develop Scala apps directly on your Dataproc cluster. Hadoop and Spark are pre-installed on Dataproc clusters, and they are configured with the Cloud Storage connector, which allows your code to read and write data directly from and to Cloud Storage.

This example shows you how to SSH into your project's Dataproc cluster master node, then use the spark-shell REPL to create and run a Scala wordcount mapreduce application.

1. SSH into the Dataproc cluster's master node

   1. Go to your project's Dataproc Clusters page in the Google Cloud console, then click on the name of your cluster.

   2. On the cluster detail page, select the VM Instances tab, then click the SSH selection that appears at the right your cluster's name row.

      A browser window opens at your home directory on the master node

2. Launch the spark-shell

   $ spark-shell
   ...
   Using Scala version ...
   Type in expressions to have them evaluated.
   Type :help for more information.
   ...
   Spark context available as sc.
   ...
   SQL context available as sqlContext.
   scala>
   

3. Create an RDD (Resilient Distributed Dataset) from a Shakespeare text snippet located in public Cloud Storage

   Shakespeare text snippet:
   

   What's in a name? That which we call a rose
   By any other name would smell as sweet.
   

   scala> val text_file = sc.textFile("gs://pub/shakespeare/rose.txt")
   

4. Run a wordcount mapreduce on the text, then display the wordcounts result

   scala> val wordCounts = text_file.flatMap(line => line.split(" ")).map(word =>
   (word, 1)).reduceByKey((a, b) => a + b)
   scala> wordCounts.collect
   ... Array((call,1), (What's,1), (sweet.,1), (we,1), (as,1), (name?,1), (any,1), (other,1),
   (rose,1), (smell,1), (name,1), (a,2), (would,1), (in,1), (which,1), (That,1), (By,1))
   

5. Save the counts in <bucket-name>/wordcounts-out in Cloud Storage, then exit the scala-shell

   scala> wordCounts.saveAsTextFile("gs://<bucket-name>/wordcounts-out/")
   scala> exit
   

6. Use the gcloud CLI to list the output files and display the file contents

   $ gcloud storage ls gs://bucket-name/wordcounts-out/
   gs://spark-scala-demo-bucket/wordcounts-out/
   gs://spark-scala-demo-bucket/wordcounts-out/_SUCCESS
   gs://spark-scala-demo-bucket/wordcounts-out/part-00000
   gs://spark-scala-demo-bucket/wordcounts-out/part-00001
   

7. Check gs://<bucket-name>/wordcounts-out/part-00000 contents

   $ gcloud storage cat gs://bucket-name/wordcounts-out/part-00000
   (call,1)
   (What's,1)
   (sweet.,1)
   (we,1)
   (as,1)
   (name?,1)
   (any,1)
   (other,1)
   

Running Pre-Installed Example code

The Dataproc master node contains runnable jar files with standard Apache Hadoop and Spark examples.

Jar Type	Master node /usr/lib/ location	GitHub Source	Apache Docs
Hadoop	hadoop-mapreduce/hadoop-mapreduce-examples.jar	source link	MapReduce Tutorial
Spark	spark/lib/spark-examples.jar	source link	Spark Examples

Submitting examples to your cluster from the command line

Examples can be submitted from your local development machine using the Google Cloud CLI gcloud command-line tool (see Using the Google Cloud console to submit jobs from the Google Cloud console).

Hadoop WordCount example

gcloud dataproc jobs submit hadoop --cluster=cluster-name \
    --region=region \
    --jars=file:///usr/lib/hadoop-mapreduce/hadoop-mapreduce-examples.jar \
    --class=org.apache.hadoop.examples.WordCount \
    -- URI of input file URI of output file

Spark WordCount example

gcloud dataproc jobs submit spark --cluster=cluster-name \
    --region=region \
    --jars=file:///usr/lib/spark/examples/jars/spark-examples.jar \
    --class=org.apache.spark.examples.JavaWordCount \
    -- URI of input file

Shutdown your cluster

To avoid ongoing charges, shutdown your cluster and delete the Cloud Storage resources (Cloud Storage bucket and files) used for this tutorial.

To shutdown a cluster:

gcloud dataproc clusters delete cluster-name \
    --region=region

To delete the Cloud Storage jar file:

gcloud storage rm gs://bucket-name/HelloWorld.jar

You can delete a bucket and all of its folders and files with the following command:

gcloud storage rm gs://bucket-name/ --recursive

What's next

• Read Managing Java dependencies for Apache Spark applications on Dataproc.

• See Spark job tuning tips