This topic describes how to set up a Spark on MaxCompute development environment.


Before you start to set up the development environment, make sure that you have installed the following softwares:
  • JDK 1.8
  • Python2.7
  • Maven
  • Git

Download the Spark on MaxCompute client

The Spark on MaxCompute package is released with the MaxCompute authorization function. This allows Spark on MaxCompute to serve as a client that submits jobs with the spark-submit script. Currently, we have released two packages for Spark1.x and Spark2.x.

Set environment variables

  • Set JAVA_HOME as follows.
    # We recommend that you use JDK 1.8 or later.
    export JAVA_HOME=/path/to/jdk
    export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar
    export PATH=$JAVA_HOME/bin:$PATH
  • Set SPARK_HOME as follows.
    Download the Spark on MaxCompute client and unzip it to a local path. Replace SPARK_HOME with your unzip path.
    export SPARK_HOME=/path/to/spark_extracted_package
    export PATH=$SPARK_HOME/bin:$PATH
  • To use PySpark, you need to install Python 2.7 and set PATH as follows.
    export PATH=/path/to/python/bin/:$PATH

Configure the spark-defaults.conf file

First time you use the Spark on MaxCompute client, you need to configure the spark-defaults.conf file.

The spark-defaults.conf.template file is in the $SPARK_HOME/conf path. You can use it as a template for configurations of the spark-defaults.conf file.
# spark-defaults.conf
# Enter the MaxCompute project and account information = XXX = XXX     
spark.hadoop.odps.access.key = XXX

# Retain the following default settings.
Spark.hadoop.odps.end.point = # Configure the endpoint through which the Spark on MaxCompute client accesses MaxCompute projects. Specify the actual endpoint based on your requirements. For more information, see Configure endpoints.
spark.hadoop.odps.runtime.end.point = # This is the endpoint of the Spark, which is the endpoint of the MaxCompute VPC where the region is located. You can modify this parameter based on your requirements.
spark.hadoop.odps.task.major.version = cupid_v2
spark.hadoop.odps.cupid.container.image.enable = true
spark.hadoop.odps.cupid.container.vm.engine.type = hyper

spark.hadoop.odps.cupid.webproxy.endpoint = =

For special scenarios and functions, you need to enable some other configurations. For more information, see Spark on MaxCompute configuration details.

Configure dependencies

  • Configure the dependency for Spark on MaxCompute jobs to access MaxCompute tables.
    Spark on MaxCompute jobs use the odps-spark-datasource module to access MaxCompute tables. The following example shows Maven configuration:
    <! -- Spark-2.x uses the following module -->
    <! -- Spark-1.x uses the following module -->
  • Configure the dependency for Spark on MaxCompute jobs to access OSS.
    If Spark on MaxCompute jobs need to access OSS, add the following dependency.

Project preparation

Spark on MaxCompute provides a demo project template. We recommend that you download and copy the template to delevelop your application.
Note In the demo project, the scope parameter for the Spark dependency is set to provided. Do not modify this parameter, otherwise the submitted job will not run normally.
  • Download and compile Spark-1.x template
    git clone
    cd spark-1.x
    mvn clean package
  • Download and compile Spark-2.x
    git clone
    cd spark-2.x
    mvn clean package

SparkPi smoke test

After completing the preceding tasks, perform a smoke test to check end-to-end connectivity. For example, for Spark-2.x, you can run the following commands to conduct a SparkPi test:
#/Path/to/MaxCompute-Spark Set the correct path of the compiled JAR package.
bin/spark-submit --master yarn-cluster --class com.aliyun.odps.spark.examples.SparkPi \

# The following log means that the smoke test is successful.
19/06/11 11:57:30 INFO Client: 
         client token: N/A
         diagnostics: N/A
         ApplicationMaster host:
         ApplicationMaster RPC port: 38965
         queue: queue
         start time: 1560225401092
         final status: SUCCEEDED

Notes on using IDEA locally

Usually, run the code on the cluster after successful local debugging. However, Spark supports local execution in IDEA. Read the following notes before running the code:
  • Set the spark.master parameter manually.
    val spark = SparkSession
          .config ("spark.master", "local [4]") // The code can run directly after you set spark.master to local[N]. N is the number of concurrency.
  • Manually add the related dependency of the Spark on MaxCompute client in IDEA.
    In the pom.xml file, set the scope parameter to provided to avoid the "NoClassDefFoundError" error.
    Exception in thread "main" java.lang.NoClassDefFoundError: org/apache/spark/sql/SparkSession$
        at com.aliyun.odps.spark.examples.SparkPi$.main(SparkPi.scala:27)
        at com.aliyun.odps.spark.examples.Spa. r. kPi.main(SparkPi.scala)
    Caused by: java.lang.ClassNotFoundException: org.apache.spark.sql.SparkSession$
        at java.lang.ClassLoader.loadClass(
        at sun.misc.Launcher$AppClassLoader.loadClass(
        at java.lang.ClassLoader.loadClass(
        ... 2 more
    Use the following method to manually add the JARs directory in Spark on MaxCompute to the IDEA project template. The parameter scope=provided will remain unchanged. The code will not throw errors when it runs in IDEA directly.
    1. Click File on the top menu bar in IDEA and select Project Structure….44
    2. On the Project Structure page, click Modules in the left navigation bar. Select Resource Packages, and click the Dependencies tab in Resource Packages.
    3. On the Dependencies tab in the resource package, click + in the lower left corner and select JARs or directories…. Add the JARs directory in Spark on MaxCompute.
  • The spark-defaults.conf file cannot be used in local mode. Set your configurations manually.
    When you submit a job in spark-submit mode, the system will read the configurations in the spark-defaults.conf file. When in local mode, you need to manually set the configurations. For example, if you allow Spark SQL to access MaxCompute tables, set your configurations as follows in local mode.
    val spark = SparkSession
          .config ("spark.master", "local [4]") // The code can run directly after you set spark.master to local[N]. N is the number of concurrency.
          .config("", "****")
          .config("", "****")
          .config("spark.hadoop.odps.access.key", "****")
          .config("spark.hadoop.odps.end.point", "")
          .config("spark.sql.catalogImplementation", "odps")