CAPS extends Apache Spark™ with Cypher, the industry's most widely used property graph query language defined and maintained by the openCypher project. It allows for the integration of many data sources and supports multiple graph querying. It enables you to use your Spark cluster to run analytical graph queries. Queries can also return graphs to create processing pipelines.

CAPS allows you to develop complex processing pipelines orchestrated by a powerful and expressive high-level language. In addition to developers and big data integration specialists, CAPS is also of practical use to data scientists, offering tools allowing for disparate data sources to be integrated into a single graph. From this graph, queries can extract subgraphs of interest into new result graphs, which can be conveniently exported for further processing.

CAPS builds on the Spark SQL DataFrame API, offering integration with standard Spark SQL processing and also allows integration with GraphX. To learn more about this, please see our examples.

The functionality and APIs are stabilizing but surface changes (e.g. to the Cypher syntax and semantics for multiple graph processing and graph projections/construction) are still likely to occur. We invite you to try out the project, and we welcome feedback and contributions

A 1.0 release of the project with a stable feature set and API/language surface is targeted for Q4 of 2018.

We expect continuing development of the project after 1.0. If you are interested in contributing to the project we would love to hear from you; email us at [email protected] or just raise a PR. Please note that this is an openCypher project and contributions can only be accepted if you’ve agreed to the openCypher Contributors Agreement (oCCA).

CAPS is built on top of the Spark DataFrame API and uses features such as the Catalyst optimizer. The Spark representations are accessible and can be converted to representations that integrate with other Spark libraries.

CAPS supports a subset of Cypher and is the first implementation of multiple graphs and graph query compositionality.

CAPS currently supports importing graphs from both Neo4j and from custom CSV format in HDFS and local file system. CAPS has a data source API that allows you to plug in custom data importers for external graphs.

CAPS is under rapid development and we are planning to offer support for:

• a large subset of the Cypher language
• new Cypher Multiple Graph features
• integration with Spark SQL
• injection of custom graph data sources

Currently CAPS is a third-party add-on to the Spark ecosystem. We, however, believe that property graphs and graph processing has the potential to be come a vital part of data analytics. We are thus working, in cooperation with Databricks, on making CAPS a core part of Spark. The first step on this road is the specification of a PropertyGraph API, similar to SQL and Dataframes, along with porting Cypher 9 features of CAPS to the core Spark project in a so called Spark Project Improvement Proposal (SPIP). We are currently in the first phase of this process. The SPIP describing the motivation and goals is published here SPARK-25994. Additionally SPARK-26028 proposes an API design and implementation strategies.

CAPS is currently easiest to use with Scala. Below we explain how you can import a simple graph and run a Cypher query on it.

CAPS is built using Gradle

./gradlew build

In order to use CAPS add the following dependency:

Maven:

<dependency>
  <groupId>org.opencypher</groupId>
  <artifactId>spark-cypher</artifactId>
  <version>0.2.3</version>
</dependency>

sbt:

libraryDependencies += "org.opencypher" % "spark-cypher" % "0.2.3"

Remember to add fork in run := true in your build.sbt for scala projects; this is not CAPS specific, but a quirk of spark execution that will help prevent problems.

mvn scala:doc

Documentation will be generated and placed under [MODULE_DIRECTORY]/target/site/scaladocs/index.html

Cypher is based on the property graph data model, comprising labelled nodes and typed relationships, with a relationship either connecting two nodes, or forming a self-loop on a single node. Both nodes and relationships are uniquely identified by an ID (in CAPS this is of type Long), and contain a set of properties.

The following example shows how to convert a social network represented as Scala case classes to a PropertyGraph representation. The PropertyGraph representation is internally transformed into Spark data frames. If you have existing data frames which you would like to treat as a graph, have a look at our DataFrameInputExample.

Once the property graph is constructed, it supports Cypher queries via its cypher method.

import org.opencypher.spark.api.CAPSSession
import org.opencypher.spark.api.io.{Node, Relationship, RelationshipType}
import org.opencypher.spark.util.ConsoleApp

/**
  * Demonstrates basic usage of the CAPS API by loading an example network via Scala case classes and running a Cypher
  * query on it.
  */
object CaseClassExample extends ConsoleApp {

  // 1) Create CAPS session
  implicit val session: CAPSSession = CAPSSession.local()

  // 2) Load social network data via case class instances
  val socialNetwork = session.readFrom(SocialNetworkData.persons, SocialNetworkData.friendships)

  // 3) Query graph with Cypher
  val results = socialNetwork.cypher(
    """|MATCH (a:Person)-[r:FRIEND_OF]->(b)
       |RETURN a.name, b.name, r.since
       |ORDER BY a.name""".stripMargin
  )

  // 4) Print result to console
  results.show
}

/**
  * Specify schema and data with case classes.
  */
object SocialNetworkData {

  case class Person(id: Long, name: String, age: Int) extends Node

  @RelationshipType("FRIEND_OF")
  case class Friend(id: Long, source: Long, target: Long, since: String) extends Relationship

  val alice = Person(0, "Alice", 10)
  val bob = Person(1, "Bob", 20)
  val carol = Person(2, "Carol", 15)

  val persons = List(alice, bob, carol)
  val friendships = List(Friend(0, alice.id, bob.id, "23/01/1987"), Friend(1, bob.id, carol.id, "12/12/2009"))
}

The above program prints:

╔═════════╤═════════╤══════════════╗
║ a.name  │ b.name  │ r.since      ║
╠═════════╪═════════╪══════════════╣
║ 'Alice' │ 'Bob'   │ '23/01/1987' ║
║ 'Bob'   │ 'Carol' │ '12/12/2009' ║
╚═════════╧═════════╧══════════════╝
(2 rows)

More examples, including multiple graph features, can be found in the examples module.

See the documentation in org.opencypher.spark.impl.io.hdfs.CsvGraphLoader, which specifies how to structure the CSV and the schema mappings that describe the graph structure for the underlying data.

• How to use CAPS in Apache Zeppelin
• Look at and contribute to the Wiki

We would love to find out about any issues you encounter and are happy to accept contributions following a Contributors License Agreement (CLA) signature as per the process outlined in our contribution guidelines.

The project is licensed under the Apache Software License, Version 2.0, with an extended attribution notice as described in the license header.

© Copyright 2016-2019 Neo4j, Inc.

Apache Spark™, Spark, and Apache are registered trademarks of the Apache Software Foundation.