Use Azure Toolkit for Eclipse to create Spark applications for an HDInsight cluster

Use HDInsight Tools in Azure Toolkit for Eclipse to develop Spark applications written in Scala and submit them to an Azure HDInsight Spark cluster, directly from the Eclipse IDE. You can use the HDInsight Tools plug-in in a few different ways:

  • To develop and submit a Scala Spark application on an HDInsight Spark cluster
  • To access your Azure HDInsight Spark cluster resources
  • To develop and run a Scala Spark application locally
Important

You can use this tool to create and submit applications only for an HDInsight Spark cluster on Linux.

Prerequisites

Install HDInsight Tools in Azure Toolkit for Eclipse and the Scala plug-in

Install HDInsight Tools

HDInsight Tools for Eclipse is available as part of Azure Toolkit for Eclipse. For installation instructions, see Installing Azure Toolkit for Eclipse.

Install the Scala plug-in

When you open Eclipse, HDInsight Tools automatically detects whether you installed the Scala plug-in. Select OK to continue, and then follow the instructions to install the plug-in from the Eclipse Marketplace.

Automatic installation of the Scala plug-in

Sign in to your Azure subscription

  1. Start the Eclipse IDE and open Azure Explorer. On the Window menu, select Show View, and then select Other. In the dialog box that opens, expand Azure, select Azure Explorer, and then select OK.

    Show View dialog box

  2. Right-click the Azure node, and then select Sign in.
  3. In the Azure Sign In dialog box, choose the authentication method, select Sign in, and enter your Azure credentials.

    Azure Sign In dialog box

  4. After you're signed in, the Select Subscriptions dialog box lists all the Azure subscriptions associated with the credentials. Click Select to close the dialog box.

    Select Subscriptions dialog box

  5. On the Azure Explorer tab, expand HDInsight to see the HDInsight Spark clusters under your subscription.

    HDInsight Spark clusters in Azure Explorer

  6. You can further expand a cluster name node to see the resources (for example, storage accounts) associated with the cluster.

    Expanding a cluster name to see resources

Set up a Spark Scala project for an HDInsight Spark cluster

  1. In the Eclipse IDE workspace, select File, select New, and then select Project.
  2. In the New Project wizard, expand HDInsight, select Spark on HDInsight (Scala), and then select Next.

    Selecting the Spark on HDInsight (Scala) project

  3. The Scala project creation wizard automatically detects whether you installed the Scala plug-in. Select OK to continue downloading the Scala plug-in, and then follow the instructions to restart Eclipse.

    Scala check

  4. In the New HDInsight Scala Project dialog box, provide the following values, and then select Next:

    • Enter a name for the project.
    • In the JRE area, make sure that Use an execution environment JRE is set to JavaSE-1.7 or later.
    • Make sure that the Spark SDK is set to the location where you downloaded the SDK. The link to the download location is included in the prerequisites earlier in this article. You can also download the SDK from the link in the dialog box.

    New HDInsight Scala Project dialog box

  5. In the next dialog box, select the Libraries tab and keep the defaults, and then select Finish.

    Libraries tab

Create a Scala application for an HDInsight Spark cluster

  1. In the Eclipse IDE, from Package Explorer, expand the project that you created earlier, right-click src, point to New, and then select Other.
  2. In the Select a wizard dialog box, expand Scala Wizards, select Scala Object, and then select Next.

    Select a wizard dialog box

  3. In the Create New File dialog box, enter a name for the object, and then select Finish.

    Create New File dialog box

  4. Paste the following code in the text editor:

     import org.apache.spark.SparkConf
     import org.apache.spark.SparkContext
    
     object MyClusterApp{
       def main (arg: Array[String]): Unit = {
         val conf = new SparkConf().setAppName("MyClusterApp")
         val sc = new SparkContext(conf)
    
         val rdd = sc.textFile("wasb:///HdiSamples/HdiSamples/SensorSampleData/hvac/HVAC.csv")
    
         //find the rows that have only one digit in the seventh column in the CSV
         val rdd1 =  rdd.filter(s => s.split(",")(6).length() == 1)
    
         rdd1.saveAsTextFile("wasb:///HVACOut")
       }        
     }
    
  5. Run the application on an HDInsight Spark cluster:

    a. From Package Explorer, right-click the project name, and then select Submit Spark Application to HDInsight.
    b. In the Spark Submission dialog box, provide the following values, and then select Submit:

    • For Cluster Name, select the HDInsight Spark cluster on which you want to run your application.
    • Select an artifact from the Eclipse project, or select one from a hard drive. The default value depends on the item that you right-click from Package Explorer.
    • In the Main class name drop-down list, the submission wizard displays all object names from your project. Select or enter one that you want to run. If you selected an artifact from a hard drive, you must enter the main class name manually.
    • Because the application code in this example does not require any command-line arguments or reference JARs or files, you can leave the remaining text boxes empty.

      Spark Submission dialog box

  6. The Spark Submission tab should start displaying the progress. You can stop the application by selecting the red button in the Spark Submission window. You can also view the logs for this specific application run by selecting the globe icon (denoted by the blue box in the image).

    Spark Submission window

Access and manage HDInsight Spark clusters by using HDInsight Tools in Azure Toolkit for Eclipse

You can perform various operations by using HDInsight Tools, including accessing the job output.

Access the job view

  1. In Azure Explorer, expand HDInsight, expand the Spark cluster name, and then select Jobs.

    Job view node

  2. Select the Jobs node. HDInsight Tools automatically detects whether you installed the E(fx)clipse plug-in. Select OK to continue, and then follow the instructions to install the Eclipse Marketplace and restart Eclipse.

    Install E(fx)clipse

  3. Open the Job View from the Jobs node. In the right pane, the Spark Job View tab displays all the applications that were run on the cluster. Select the name of the application for which you want to see more details.

    Application details

    You can then take any of these actions:

    • Hover on the job graph. It displays basic info about the running job. Select the job graph, and you can see the stages and info that every job generates.

      Job stage details

    • Select the Log tab to view frequently used logs, including Driver Stderr, Driver Stdout, and Directory Info.

      Log details

    • Open the Spark history UI and the YARN UI (at the application level) by selecting the hyperlinks at the top of the window.

Access the storage container for the cluster

  1. In Azure Explorer, expand the HDInsight root node to see a list of HDInsight Spark clusters that are available.
  2. Expand the cluster name to see the storage account and the default storage container for the cluster.

    Storage account and default storage container

  3. Select the storage container name associated with the cluster. In the right pane, double-click the HVACOut folder. Open one of the part- files to see the output of the application.

Access the Spark history server

  1. In Azure Explorer, right-click your Spark cluster name, and then select Open Spark History UI. When you're prompted, enter the admin credentials for the cluster. You specified these while provisioning the cluster.
  2. In the Spark history server dashboard, you use the application name to look for the application that you just finished running. In the preceding code, you set the application name by using val conf = new SparkConf().setAppName("MyClusterApp"). So, your Spark application name was MyClusterApp.

Start the Ambari portal

  1. In Azure Explorer, right-click your Spark cluster name, and then select Open Cluster Management Portal (Ambari).
  2. When you're prompted, enter the admin credentials for the cluster. You specified these while provisioning the cluster.

Manage Azure subscriptions

By default, HDInsight Tools in Azure Toolkit for Eclipse lists the Spark clusters from all your Azure subscriptions. If necessary, you can specify the subscriptions for which you want to access the cluster.

  1. In Azure Explorer, right-click the Azure root node, and then select Manage Subscriptions.
  2. In the dialog box, clear the check boxes for the subscription that you don't want to access, and then select Close. You can also select Sign Out if you want to sign out of your Azure subscription.

Run a Spark Scala application locally

You can use HDInsight Tools in Azure Toolkit for Eclipse to run Spark Scala applications locally on your workstation. Typically, these applications don't need access to cluster resources such as a storage container, and you can run and test them locally.

Prerequisite

While you're running the local Spark Scala application on a Windows computer, you might get an exception as explained in SPARK-2356. This exception occurs because WinUtils.exe is missing in Windows.

To resolve this error, you must download the executable to a location like C:\WinUtils\bin. You must then add the environment variable HADOOP_HOME and set the value of the variable to C\WinUtils.

Run a local Spark Scala application

  1. Start Eclipse and create a project. In the New Project dialog box, make the following choices, and then select Next.

    • In the left pane, select HDInsight.
    • In the right pane, select Spark on HDInsight Local Run Sample (Scala).

    New Project dialog box

  2. To provide the project details, follow steps 3 through 6 from the earlier section Set up a Spark Scala project for an HDInsight Spark cluster.

  3. The template adds a sample code (LogQuery) under the src folder that you can run locally on your computer.

    Location of LogQuery

  4. Right-click the LogQuery application, point to Run As, and then select 1 Scala Application. Output like this appears on the Console tab:

    Spark application local run result

Known problems

To submit an application to Azure Data Lake Store, select Interactive mode during the Azure sign-in process. If you select Automated mode, you might get an error.

Interactive sign-in

You can choose an Azure Data Lake cluster to submit your application with any sign-in method.

Currently, viewing Spark outputs directly is not supported.

Feedback

If you have any feedback, or if you encounter any other problems when using this tool, send us an email at hdivstool@microsoft.com.

See also

Scenarios

Creating and running applications

Tools and extensions

Managing resources