Spark Overview
Apache Spark is a fast and general-purpose cluster computing system. It provides high-level APIs in Scala, Java, and Python that make parallel jobs easy to write, and an optimized engine that supports general computation graphs. It also supports a rich set of higher-level tools including Shark (Hive on Spark), MLlib for machine learning, Bagel for graph processing, and Spark Streaming.
Downloading
Get Spark by visiting the downloads page of the Apache Spark site. This documentation is for Spark version 0.8.1-incubating.
Spark runs on both Windows and UNIX-like systems (e.g. Linux, Mac OS). All you need to run it is to have java
to installed on your system PATH
, or the JAVA_HOME
environment variable pointing to a Java installation.
Building
Spark uses Simple Build Tool, which is bundled with it. To compile the code, go into the top-level Spark directory and run
sbt/sbt assembly
For its Scala API, Spark 0.8.1-incubating depends on Scala 2.9.3. If you write applications in Scala, you will need to use this same version of Scala in your own program – newer major versions may not work. You can get the right version of Scala from scala-lang.org.
Running the Examples and Shell
Spark comes with several sample programs in the examples
directory.
To run one of the samples, use ./run-example <class> <params>
in the top-level Spark directory
(the run-example
script sets up the appropriate paths and launches that program).
For example, try ./run-example org.apache.spark.examples.SparkPi local
.
Each example prints usage help when run with no parameters.
Note that all of the sample programs take a <master>
parameter specifying the cluster URL
to connect to. This can be a URL for a distributed cluster,
or local
to run locally with one thread, or local[N]
to run locally with N threads. You should start by using
local
for testing.
Finally, you can run Spark interactively through modified versions of the Scala shell (./spark-shell
) or
Python interpreter (./pyspark
). These are a great way to learn the framework.
Launching on a Cluster
The Spark cluster mode overview explains the key concepts in running on a cluster. Spark can run both by itself, or over several existing cluster managers. It currently provides several options for deployment:
- Amazon EC2: our EC2 scripts let you launch a cluster in about 5 minutes
- Standalone Deploy Mode: simplest way to deploy Spark on a private cluster
- Apache Mesos
- Hadoop YARN
A Note About Hadoop Versions
Spark uses the Hadoop-client library to talk to HDFS and other Hadoop-supported
storage systems. Because the HDFS protocol has changed in different versions of
Hadoop, you must build Spark against the same version that your cluster uses.
By default, Spark links to Hadoop 1.0.4. You can change this by setting the
SPARK_HADOOP_VERSION
variable when compiling:
SPARK_HADOOP_VERSION=2.2.0 sbt/sbt assembly
In addition, if you wish to run Spark on YARN, set
SPARK_YARN
to true
:
SPARK_HADOOP_VERSION=2.0.5-alpha SPARK_YARN=true sbt/sbt assembly
Note that on Windows, you need to set the environment variables on separate lines, e.g., set SPARK_HADOOP_VERSION=1.2.1
.
For this version of Spark (0.8.1) Hadoop 2.2.x (or newer) users will have to build Spark and publish it locally. See Launching Spark on YARN. This is needed because Hadoop 2.2 has non backwards compatible API changes.
Where to Go from Here
Programming guides:
- Quick Start: a quick introduction to the Spark API; start here!
- Spark Programming Guide: an overview of Spark concepts, and details on the Scala API
- Java Programming Guide: using Spark from Java
- Python Programming Guide: using Spark from Python
- Spark Streaming: using the alpha release of Spark Streaming
- MLlib (Machine Learning): Spark’s built-in machine learning library
- Bagel (Pregel on Spark): simple graph processing model
API Docs:
- Spark for Java/Scala (Scaladoc)
- Spark for Python (Epydoc)
- Spark Streaming for Java/Scala (Scaladoc)
- MLlib (Machine Learning) for Java/Scala (Scaladoc)
- Bagel (Pregel on Spark) for Scala (Scaladoc)
Deployment guides:
- Cluster Overview: overview of concepts and components when running on a cluster
- Amazon EC2: scripts that let you launch a cluster on EC2 in about 5 minutes
- Standalone Deploy Mode: launch a standalone cluster quickly without a third-party cluster manager
- Mesos: deploy a private cluster using Apache Mesos
- YARN: deploy Spark on top of Hadoop NextGen (YARN)
Other documents:
- Configuration: customize Spark via its configuration system
- Tuning Guide: best practices to optimize performance and memory use
- Hardware Provisioning: recommendations for cluster hardware
- Job Scheduling: scheduling resources across and within Spark applications
- Building Spark with Maven: build Spark using the Maven system
- Contributing to Spark
External resources:
- Spark Homepage
- Shark: Apache Hive over Spark
- Mailing Lists: ask questions about Spark here
- AMP Camps: a series of training camps at UC Berkeley that featured talks and exercises about Spark, Shark, Mesos, and more. Videos, slides and exercises are available online for free.
- Code Examples: more are also available in the examples subfolder of Spark
- Paper Describing Spark
- Paper Describing Spark Streaming
Community
To get help using Spark or keep up with Spark development, sign up for the user mailing list.
If you’re in the San Francisco Bay Area, there’s a regular Spark meetup every few weeks. Come by to meet the developers and other users.
Finally, if you’d like to contribute code to Spark, read how to contribute.