Quick Start

This tutorial provides a quick introduction to using Spark. We will first introduce the API through Spark’s interactive Scala shell (don’t worry if you don’t know Scala – you will not need much for this), then show how to write standalone applications in Scala, Java, and Python. See the programming guide for a more complete reference.

To follow along with this guide, you only need to have successfully built Spark on one machine. Simply go into your Spark directory and run:

$ sbt/sbt assembly

Interactive Analysis with the Spark Shell

Basics

Spark’s interactive shell provides a simple way to learn the API, as well as a powerful tool to analyze datasets interactively. Start the shell by running ./bin/spark-shell in the Spark directory.

Spark’s primary abstraction is a distributed collection of items called a Resilient Distributed Dataset (RDD). RDDs can be created from Hadoop InputFormats (such as HDFS files) or by transforming other RDDs. Let’s make a new RDD from the text of the README file in the Spark source directory:

scala> val textFile = sc.textFile("README.md")
textFile: spark.RDD[String] = spark.MappedRDD@2ee9b6e3

RDDs have actions, which return values, and transformations, which return pointers to new RDDs. Let’s start with a few actions:

scala> textFile.count() // Number of items in this RDD
res0: Long = 74

scala> textFile.first() // First item in this RDD
res1: String = # Apache Spark

Now let’s use a transformation. We will use the filter transformation to return a new RDD with a subset of the items in the file.

scala> val linesWithSpark = textFile.filter(line => line.contains("Spark"))
linesWithSpark: spark.RDD[String] = spark.FilteredRDD@7dd4af09

We can chain together transformations and actions:

scala> textFile.filter(line => line.contains("Spark")).count() // How many lines contain "Spark"?
res3: Long = 15

More on RDD Operations

RDD actions and transformations can be used for more complex computations. Let’s say we want to find the line with the most words:

scala> textFile.map(line => line.split(" ").size).reduce((a, b) => if (a > b) a else b)
res4: Long = 16

This first maps a line to an integer value, creating a new RDD. reduce is called on that RDD to find the largest line count. The arguments to map and reduce are Scala function literals (closures), and can use any language feature or Scala/Java library. For example, we can easily call functions declared elsewhere. We’ll use Math.max() function to make this code easier to understand:

scala> import java.lang.Math
import java.lang.Math

scala> textFile.map(line => line.split(" ").size).reduce((a, b) => Math.max(a, b))
res5: Int = 16

One common data flow pattern is MapReduce, as popularized by Hadoop. Spark can implement MapReduce flows easily:

scala> val wordCounts = textFile.flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey((a, b) => a + b)
wordCounts: spark.RDD[(java.lang.String, Int)] = spark.ShuffledAggregatedRDD@71f027b8

Here, we combined the flatMap, map and reduceByKey transformations to compute the per-word counts in the file as an RDD of (String, Int) pairs. To collect the word counts in our shell, we can use the collect action:

scala> wordCounts.collect()
res6: Array[(java.lang.String, Int)] = Array((need,2), ("",43), (Extra,3), (using,1), (passed,1), (etc.,1), (its,1), (`/usr/local/lib/libmesos.so`,1), (`SCALA_HOME`,1), (option,1), (these,1), (#,1), (`PATH`,,2), (200,1), (To,3),...

Caching

Spark also supports pulling data sets into a cluster-wide in-memory cache. This is very useful when data is accessed repeatedly, such as when querying a small “hot” dataset or when running an iterative algorithm like PageRank. As a simple example, let’s mark our linesWithSpark dataset to be cached:

scala> linesWithSpark.cache()
res7: spark.RDD[String] = spark.FilteredRDD@17e51082

scala> linesWithSpark.count()
res8: Long = 15

scala> linesWithSpark.count()
res9: Long = 15

It may seem silly to use Spark to explore and cache a 30-line text file. The interesting part is that these same functions can be used on very large data sets, even when they are striped across tens or hundreds of nodes. You can also do this interactively by connecting bin/spark-shell to a cluster, as described in the programming guide.

A Standalone App in Scala

Now say we wanted to write a standalone application using the Spark API. We will walk through a simple application in both Scala (with SBT), Java (with Maven), and Python. If you are using other build systems, consider using the Spark assembly JAR described in the developer guide.

We’ll create a very simple Spark application in Scala. So simple, in fact, that it’s named SimpleApp.scala:

/*** SimpleApp.scala ***/
import org.apache.spark.SparkContext
import org.apache.spark.SparkContext._

object SimpleApp {
  def main(args: Array[String]) {
    val logFile = "$YOUR_SPARK_HOME/README.md" // Should be some file on your system
    val sc = new SparkContext("local", "Simple App", "YOUR_SPARK_HOME",
      List("target/scala-2.10/simple-project_2.10-1.0.jar"))
    val logData = sc.textFile(logFile, 2).cache()
    val numAs = logData.filter(line => line.contains("a")).count()
    val numBs = logData.filter(line => line.contains("b")).count()
    println("Lines with a: %s, Lines with b: %s".format(numAs, numBs))
  }
}

This program just counts the number of lines containing ‘a’ and the number containing ‘b’ in the Spark README. Note that you’ll need to replace $YOUR_SPARK_HOME with the location where Spark is installed. Unlike the earlier examples with the Spark shell, which initializes its own SparkContext, we initialize a SparkContext as part of the program. We pass the SparkContext constructor four arguments, the type of scheduler we want to use (in this case, a local scheduler), a name for the application, the directory where Spark is installed, and a name for the jar file containing the application’s code. The final two arguments are needed in a distributed setting, where Spark is running across several nodes, so we include them for completeness. Spark will automatically ship the jar files you list to slave nodes.

This file depends on the Spark API, so we’ll also include an sbt configuration file, simple.sbt which explains that Spark is a dependency. This file also adds a repository that Spark depends on:

name := "Simple Project"

version := "1.0"

scalaVersion := "2.10.3"

libraryDependencies += "org.apache.spark" %% "spark-core" % "0.9.1"

resolvers += "Akka Repository" at "http://repo.akka.io/releases/"

If you also wish to read data from Hadoop’s HDFS, you will also need to add a dependency on hadoop-client for your version of HDFS:

libraryDependencies += "org.apache.hadoop" % "hadoop-client" % "<your-hdfs-version>"

Finally, for sbt to work correctly, we’ll need to layout SimpleApp.scala and simple.sbt according to the typical directory structure. Once that is in place, we can create a JAR package containing the application’s code, then use sbt/sbt run to execute our program.

$ find .
.
./simple.sbt
./src
./src/main
./src/main/scala
./src/main/scala/SimpleApp.scala

$ sbt/sbt package
$ sbt/sbt run
...
Lines with a: 46, Lines with b: 23

A Standalone App in Java

Now say we wanted to write a standalone application using the Java API. We will walk through doing this with Maven. If you are using other build systems, consider using the Spark assembly JAR described in the developer guide.

We’ll create a very simple Spark application, SimpleApp.java:

/*** SimpleApp.java ***/
import org.apache.spark.api.java.*;
import org.apache.spark.api.java.function.Function;

public class SimpleApp {
  public static void main(String[] args) {
    String logFile = "$YOUR_SPARK_HOME/README.md"; // Should be some file on your system
    JavaSparkContext sc = new JavaSparkContext("local", "Simple App",
      "$YOUR_SPARK_HOME", new String[]{"target/simple-project-1.0.jar"});
    JavaRDD<String> logData = sc.textFile(logFile).cache();

    long numAs = logData.filter(new Function<String, Boolean>() {
      public Boolean call(String s) { return s.contains("a"); }
    }).count();

    long numBs = logData.filter(new Function<String, Boolean>() {
      public Boolean call(String s) { return s.contains("b"); }
    }).count();

    System.out.println("Lines with a: " + numAs + ", lines with b: " + numBs);
  }
}

This program just counts the number of lines containing ‘a’ and the number containing ‘b’ in a text file. Note that you’ll need to replace $YOUR_SPARK_HOME with the location where Spark is installed. As with the Scala example, we initialize a SparkContext, though we use the special JavaSparkContext class to get a Java-friendly one. We also create RDDs (represented by JavaRDD) and run transformations on them. Finally, we pass functions to Spark by creating classes that extend spark.api.java.function.Function. The Java programming guide describes these differences in more detail.

To build the program, we also write a Maven pom.xml file that lists Spark as a dependency. Note that Spark artifacts are tagged with a Scala version.

<project>
  <groupId>edu.berkeley</groupId>
  <artifactId>simple-project</artifactId>
  <modelVersion>4.0.0</modelVersion>
  <name>Simple Project</name>
  <packaging>jar</packaging>
  <version>1.0</version>
  <repositories>
    <repository>
      <id>Akka repository</id>
      <url>http://repo.akka.io/releases</url>
    </repository>
  </repositories>
  <dependencies>
    <dependency> <!-- Spark dependency -->
      <groupId>org.apache.spark</groupId>
      <artifactId>spark-core_2.10</artifactId>
      <version>0.9.1</version>
    </dependency>
  </dependencies>
</project>

If you also wish to read data from Hadoop’s HDFS, you will also need to add a dependency on hadoop-client for your version of HDFS:

<dependency>
  <groupId>org.apache.hadoop</groupId>
  <artifactId>hadoop-client</artifactId>
  <version>...</version>
</dependency>

We lay out these files according to the canonical Maven directory structure:

$ find .
./pom.xml
./src
./src/main
./src/main/java
./src/main/java/SimpleApp.java

Now, we can execute the application using Maven:

$ mvn package
$ mvn exec:java -Dexec.mainClass="SimpleApp"
...
Lines with a: 46, Lines with b: 23

A Standalone App in Python

Now we will show how to write a standalone application using the Python API (PySpark).

As an example, we’ll create a simple Spark application, SimpleApp.py:

"""SimpleApp.py"""
from pyspark import SparkContext

logFile = "$YOUR_SPARK_HOME/README.md"  # Should be some file on your system
sc = SparkContext("local", "Simple App")
logData = sc.textFile(logFile).cache()

numAs = logData.filter(lambda s: 'a' in s).count()
numBs = logData.filter(lambda s: 'b' in s).count()

print "Lines with a: %i, lines with b: %i" % (numAs, numBs)

This program just counts the number of lines containing ‘a’ and the number containing ‘b’ in a text file. Note that you’ll need to replace $YOUR_SPARK_HOME with the location where Spark is installed. As with the Scala and Java examples, we use a SparkContext to create RDDs. We can pass Python functions to Spark, which are automatically serialized along with any variables that they reference. For applications that use custom classes or third-party libraries, we can add those code dependencies to SparkContext to ensure that they will be available on remote machines; this is described in more detail in the Python programming guide. SimpleApp is simple enough that we do not need to specify any code dependencies.

We can run this application using the bin/pyspark script:

$ cd $SPARK_HOME
$ ./bin/pyspark SimpleApp.py
...
Lines with a: 46, Lines with b: 23

Running on a Cluster

There are a few additional considerations when running applicaitons on a Spark, YARN, or Mesos cluster.

Including Your Dependencies

If your code depends on other projects, you will need to ensure they are also present on the slave nodes. A popular approach is to create an assembly jar (or “uber” jar) containing your code and its dependencies. Both sbt and Maven have assembly plugins. When creating assembly jars, list Spark itself as a provided dependency; it need not be bundled since it is already present on the slaves. Once you have an assembled jar, add it to the SparkContext as shown here. It is also possible to add your dependent jars one-by-one using the addJar method of SparkContext.

For Python, you can use the pyFiles argument of SparkContext or its addPyFile method to add .py, .zip or .egg files to be distributed.

Setting Configuration Options

Spark includes several configuration options that influence the behavior of your application. These should be set by building a SparkConf object and passing it to the SparkContext constructor. For example, in Java and Scala, you can do:

import org.apache.spark.{SparkConf, SparkContext}
val conf = new SparkConf()
             .setMaster("local")
             .setAppName("My application")
             .set("spark.executor.memory", "1g")
val sc = new SparkContext(conf)

Or in Python:

from pyspark import SparkConf, SparkContext
conf = SparkConf()
conf.setMaster("local")
conf.setAppName("My application")
conf.set("spark.executor.memory", "1g"))
sc = SparkContext(conf = conf)

Accessing Hadoop Filesystems

The examples here access a local file. To read data from a distributed filesystem, such as HDFS, include Hadoop version information in your build file. By default, Spark builds against HDFS 1.0.4.