This tutorial provides a quick introduction to using Spark. We will first introduce the API through Spark’s interactive shell (in Python or Scala), then show how to write applications in Java, Scala, and Python. To follow along with this guide, first, download a packaged release of Spark from the Spark website. Since we won’t be using HDFS, you can download a package for any version of Hadoop. Note that, before Spark 2.0, the main programming interface of Spark was the Resilient Distributed Dataset (RDD). After Spark 2.0, RDDs are replaced by Dataset, which is strongly-typed like an RDD, but with richer optimizations under the hood. The RDD interface is still supported, and you can get a more detailed reference at the RDD programming guide. However, we highly recommend you to switch to use Dataset, which has better performance than RDD. See the SQL programming guide to get more information about Dataset. BasicsSpark’s shell provides a simple way to learn the API, as well as a powerful tool to analyze data interactively. It is available in either Scala (which runs on the Java VM and is thus a good way to use existing Java libraries) or Python. Start it by running the following in the Spark directory: Spark’s primary abstraction is a distributed collection of items called a Dataset. Datasets can be created from Hadoop InputFormats (such as HDFS files) or by transforming other Datasets. Let’s make a new Dataset from the text of the README file in the Spark source directory:
You can get values from Dataset directly, by calling some actions, or transform the Dataset to get a new one. For more details, please read the API doc.
Now let’s transform this Dataset into a new one. We call 9 to return a new Dataset with a subset of the items in the file.
We can chain together transformations and actions:
Or if PySpark is installed with pip in your current environment: Spark’s primary abstraction is a distributed collection of items called a Dataset. Datasets can be created from Hadoop InputFormats (such as HDFS files) or by transforming other Datasets. Due to Python’s dynamic nature, we don’t need the Dataset to be strongly-typed in Python. As a result, all Datasets in Python are Dataset[Row], and we call it 0 to be consistent with the data frame concept in Pandas and R. Let’s make a new DataFrame from the text of the README file in the Spark source directory:
You can get values from DataFrame directly, by calling some actions, or transform the DataFrame to get a new one. For more details, please read the .
Now let’s transform this DataFrame to a new one. We call 9 to return a new DataFrame with a subset of the lines in the file.
We can chain together transformations and actions:
More on Dataset OperationsDataset actions and transformations can be used for more complex computations. Let’s say we want to find the line with the most words:
This first maps a line to an integer value, creating a new Dataset. 2 is called on that Dataset to find the largest word count. The arguments to 3 and 2 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 5 function to make this code easier to understand:
One common data flow pattern is MapReduce, as popularized by Hadoop. Spark can implement MapReduce flows easily: 0Here, we call 6 to transform a Dataset of lines to a Dataset of words, and then combine 7 and 8 to compute the per-word counts in the file as a Dataset of (String, Long) pairs. To collect the word counts in our shell, we can call 9: 1 2This first maps a line to an integer value and aliases it as “numWords”, creating a new DataFrame. 0 is called on that DataFrame to find the largest word count. The arguments to 1 and 0 are both , we can use 3 to get a column from a DataFrame. We can also import pyspark.sql.functions, which provides a lot of convenient functions to build a new Column from an old one.One common data flow pattern is MapReduce, as popularized by Hadoop. Spark can implement MapReduce flows easily: 3Here, we use the 4 function in 1, to transform a Dataset of lines to a Dataset of words, and then combine 6 and 8 to compute the per-word counts in the file as a DataFrame of 2 columns: “word” and “count”. To collect the word counts in our shell, we can call 9: 4CachingSpark 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 9 dataset to be cached: 5It may seem silly to use Spark to explore and cache a 100-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 0 to a cluster, as described in the . 6It may seem silly to use Spark to explore and cache a 100-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 1 to a cluster, as described in the .Suppose we wish to write a self-contained application using the Spark API. We will walk through a simple application in Scala (with sbt), Java (with Maven), and Python (pip). We’ll create a very simple Spark application in Scala–so simple, in fact, that it’s named 2: 7Note that applications should define a 3 method instead of extending 4. Subclasses of 4 may not work correctly.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 SparkSession, we initialize a SparkSession as part of the program. We call 6 to construct a 7, then set the application name, and finally call 8 to get the 7 instance.Our application depends on the Spark API, so we’ll also include an sbt configuration file, 0, which explains that Spark is a dependency. This file also adds a repository that Spark depends on: 8For sbt to work correctly, we’ll need to layout 2 and 0 according to the typical directory structure. Once that is in place, we can create a JAR package containing the application’s code, then use the 3 script to run our program. 9This example will use Maven to compile an application JAR, but any similar build system will work. We’ll create a very simple Spark application, 4: 0This 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 SparkSession, we initialize a SparkSession as part of the program. To build the program, we also write a Maven 5 file that lists Spark as a dependency. Note that Spark artifacts are tagged with a Scala version. 1We lay out these files according to the canonical Maven directory structure: 2Now, we can package the application using Maven and execute it with 6. 3Now we will show how to write an application using the Python API (PySpark). If you are building a packaged PySpark application or library you can add it to your setup.py file as: 4As an example, we’ll create a simple Spark application, 7: 5This 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 SparkSession to create Datasets. For applications that use custom classes or third-party libraries, we can also add code dependencies to 3 through its 9 argument by packaging them into a .zip file (see 0 for details). 1 is simple enough that we do not need to specify any code dependencies.We can run this application using the 2 script: 6If you have PySpark pip installed into your environment (e.g., 3), you can run your application with the regular Python interpreter or use the provided ‘spark-submit’ as you prefer. 7Other dependency management tools such as Conda and pip can be also used for custom classes or third-party libraries. See also Python Package Management. |