spark-instrumented-optimizer/docs/building-spark.md

layout	title	redirect_from
global	Building Spark	building-with-maven.html

• This will become a table of contents (this text will be scraped). {:toc}

Building Apache Spark

Apache Maven

The Maven-based build is the build of reference for Apache Spark. Building Spark using Maven requires Maven 3.3.9 or newer and Java 7+. Note that support for Java 7 is deprecated as of Spark 2.0.0 and may be removed in Spark 2.2.0.

Setting up Maven's Memory Usage

You'll need to configure Maven to use more memory than usual by setting MAVEN_OPTS:

export MAVEN_OPTS="-Xmx2g -XX:ReservedCodeCacheSize=512m"

When compiling with Java 7, you will need to add the additional option "-XX:MaxPermSize=512M" to MAVEN_OPTS.

If you don't add these parameters to MAVEN_OPTS, you may see errors and warnings like the following:

[INFO] Compiling 203 Scala sources and 9 Java sources to /Users/me/Development/spark/core/target/scala-{{site.SCALA_BINARY_VERSION}}/classes...
[ERROR] PermGen space -> [Help 1]

[INFO] Compiling 203 Scala sources and 9 Java sources to /Users/me/Development/spark/core/target/scala-{{site.SCALA_BINARY_VERSION}}/classes...
[ERROR] Java heap space -> [Help 1]

[INFO] Compiling 233 Scala sources and 41 Java sources to /Users/me/Development/spark/sql/core/target/scala-{site.SCALA_BINARY_VERSION}/classes...
OpenJDK 64-Bit Server VM warning: CodeCache is full. Compiler has been disabled.
OpenJDK 64-Bit Server VM warning: Try increasing the code cache size using -XX:ReservedCodeCacheSize=

You can fix these problems by setting the MAVEN_OPTS variable as discussed before.

Note:

• If using build/mvn with no MAVEN_OPTS set, the script will automatically add the above options to the MAVEN_OPTS environment variable.
• The test phase of the Spark build will automatically add these options to MAVEN_OPTS, even when not using build/mvn.
• You may see warnings like "ignoring option MaxPermSize=1g; support was removed in 8.0" when building or running tests with Java 8 and build/mvn. These warnings are harmless.

build/mvn

Spark now comes packaged with a self-contained Maven installation to ease building and deployment of Spark from source located under the build/ directory. This script will automatically download and setup all necessary build requirements (Maven, Scala, and Zinc) locally within the build/ directory itself. It honors any mvn binary if present already, however, will pull down its own copy of Scala and Zinc regardless to ensure proper version requirements are met. build/mvn execution acts as a pass through to the mvn call allowing easy transition from previous build methods. As an example, one can build a version of Spark as follows:

./build/mvn -DskipTests clean package

Other build examples can be found below.

Building a Runnable Distribution

To create a Spark distribution like those distributed by the Spark Downloads page, and that is laid out so as to be runnable, use ./dev/make-distribution.sh in the project root directory. It can be configured with Maven profile settings and so on like the direct Maven build. Example:

./dev/make-distribution.sh --name custom-spark --pip --r --tgz -Psparkr -Phadoop-2.7 -Phive -Phive-thriftserver -Pmesos -Pyarn

This will build Spark distribution along with Python pip and R packages. For more information on usage, run ./dev/make-distribution.sh --help

Specifying the Hadoop Version and Enabling YARN

You can specify the exact version of Hadoop to compile against through the hadoop.version property. If unset, Spark will build against Hadoop 2.6.X by default.

You can enable the yarn profile and optionally set the yarn.version property if it is different from hadoop.version.

Examples:

# Apache Hadoop 2.6.X
./build/mvn -Pyarn -DskipTests clean package

# Apache Hadoop 2.7.X and later
./build/mvn -Pyarn -Phadoop-2.7 -Dhadoop.version=2.7.3 -DskipTests clean package

Building With Hive and JDBC Support

To enable Hive integration for Spark SQL along with its JDBC server and CLI, add the -Phive and Phive-thriftserver profiles to your existing build options. By default Spark will build with Hive 1.2.1 bindings.

# With Hive 1.2.1 support
./build/mvn -Pyarn -Phive -Phive-thriftserver -DskipTests clean package

Packaging without Hadoop Dependencies for YARN

The assembly directory produced by mvn package will, by default, include all of Spark's dependencies, including Hadoop and some of its ecosystem projects. On YARN deployments, this causes multiple versions of these to appear on executor classpaths: the version packaged in the Spark assembly and the version on each node, included with yarn.application.classpath. The hadoop-provided profile builds the assembly without including Hadoop-ecosystem projects, like ZooKeeper and Hadoop itself.

Building with Mesos support

./build/mvn -Pmesos -DskipTests clean package

Building for Scala 2.10

To produce a Spark package compiled with Scala 2.10, use the -Dscala-2.10 property:

./dev/change-scala-version.sh 2.10
./build/mvn -Pyarn -Dscala-2.10 -DskipTests clean package

Note that support for Scala 2.10 is deprecated as of Spark 2.1.0 and may be removed in Spark 2.2.0.

Building submodules individually

It's possible to build Spark sub-modules using the mvn -pl option.

For instance, you can build the Spark Streaming module using:

./build/mvn -pl :spark-streaming_2.11 clean install

where spark-streaming_2.11 is the artifactId as defined in streaming/pom.xml file.

Continuous Compilation

We use the scala-maven-plugin which supports incremental and continuous compilation. E.g.

./build/mvn scala:cc

should run continuous compilation (i.e. wait for changes). However, this has not been tested extensively. A couple of gotchas to note:

• it only scans the paths src/main and src/test (see docs), so it will only work from within certain submodules that have that structure.

• you'll typically need to run mvn install from the project root for compilation within specific submodules to work; this is because submodules that depend on other submodules do so via the spark-parent module).

Thus, the full flow for running continuous-compilation of the core submodule may look more like:

$ ./build/mvn install
$ cd core
$ ../build/mvn scala:cc

Speeding up Compilation with Zinc

Zinc is a long-running server version of SBT's incremental compiler. When run locally as a background process, it speeds up builds of Scala-based projects like Spark. Developers who regularly recompile Spark with Maven will be the most interested in Zinc. The project site gives instructions for building and running zinc; OS X users can install it using brew install zinc.

If using the build/mvn package zinc will automatically be downloaded and leveraged for all builds. This process will auto-start after the first time build/mvn is called and bind to port 3030 unless the ZINC_PORT environment variable is set. The zinc process can subsequently be shut down at any time by running build/zinc-<version>/bin/zinc -shutdown and will automatically restart whenever build/mvn is called.

Building with SBT

Maven is the official build tool recommended for packaging Spark, and is the build of reference. But SBT is supported for day-to-day development since it can provide much faster iterative compilation. More advanced developers may wish to use SBT.

The SBT build is derived from the Maven POM files, and so the same Maven profiles and variables can be set to control the SBT build. For example:

./build/sbt package

To avoid the overhead of launching sbt each time you need to re-compile, you can launch sbt in interactive mode by running build/sbt, and then run all build commands at the command prompt. For more recommendations on reducing build time, refer to the Useful Developer Tools page.

## Encrypted Filesystems

When building on an encrypted filesystem (if your home directory is encrypted, for example), then the Spark build might fail with a "Filename too long" error. As a workaround, add the following in the configuration args of the scala-maven-plugin in the project pom.xml:

<arg>-Xmax-classfile-name</arg>
<arg>128</arg>

and in project/SparkBuild.scala add:

scalacOptions in Compile ++= Seq("-Xmax-classfile-name", "128"),

to the sharedSettings val. See also this PR if you are unsure of where to add these lines.

IntelliJ IDEA or Eclipse

For help in setting up IntelliJ IDEA or Eclipse for Spark development, and troubleshooting, refer to the Useful Developer Tools page.

Running Tests

Tests are run by default via the ScalaTest Maven plugin. Note that tests should not be run as root or an admin user.

The following is an example of a command to run the tests:

./build/mvn test

The ScalaTest plugin also supports running only a specific Scala test suite as follows:

./build/mvn -P... -Dtest=none -DwildcardSuites=org.apache.spark.repl.ReplSuite test
./build/mvn -P... -Dtest=none -DwildcardSuites=org.apache.spark.repl.* test

or a Java test:

./build/mvn test -P... -DwildcardSuites=none -Dtest=org.apache.spark.streaming.JavaAPISuite

Testing with SBT

The following is an example of a command to run the tests:

./build/sbt test

To run only a specific test suite as follows:

./build/sbt "test-only org.apache.spark.repl.ReplSuite"
./build/sbt "test-only org.apache.spark.repl.*"

To run test suites of a specific sub project as follows:

./build/sbt core/test

Running Java 8 Test Suites

Running only Java 8 tests and nothing else.

./build/mvn install -DskipTests
./build/mvn -pl :java8-tests_2.11 test

or

./build/sbt java8-tests/test

Java 8 tests are automatically enabled when a Java 8 JDK is detected. If you have JDK 8 installed but it is not the system default, you can set JAVA_HOME to point to JDK 8 before running the tests.

PySpark pip installable

If you are building Spark for use in a Python environment and you wish to pip install it, you will first need to build the Spark JARs as described above. Then you can construct an sdist package suitable for setup.py and pip installable package.

cd python; python setup.py sdist

Note: Due to packaging requirements you can not directly pip install from the Python directory, rather you must first build the sdist package as described above.

Alternatively, you can also run make-distribution with the --pip option.

PySpark Tests with Maven

If you are building PySpark and wish to run the PySpark tests you will need to build Spark with Hive support.

./build/mvn -DskipTests clean package -Phive
./python/run-tests

The run-tests script also can be limited to a specific Python version or a specific module

./python/run-tests --python-executables=python --modules=pyspark-sql

Note: You can also run Python tests with an sbt build, provided you build Spark with Hive support.

Running R Tests

To run the SparkR tests you will need to install the R package testthat (run install.packages(testthat) from R shell). You can run just the SparkR tests using the command:

./R/run-tests.sh

Running Docker-based Integration Test Suites

In order to run Docker integration tests, you have to install the docker engine on your box. The instructions for installation can be found at the Docker site. Once installed, the docker service needs to be started, if not already running. On Linux, this can be done by sudo service docker start.

./build/mvn install -DskipTests
./build/mvn -Pdocker-integration-tests -pl :spark-docker-integration-tests_2.11

or

./build/sbt docker-integration-tests/test