Neo4j Scala wrapper library

The Neo4j Scala wrapper library allows you use Neo4j open source graph database through a domain-specific language.

This wrapper is mostly based on the work done by Martin Kleppmann in his Scala implementation of RESTful JSON HTTP resources on top of the Neo4j graph database and Jersey project.

See this GIST for a usual Neo4j Matrix Example

You may find Neo4j-Spatial-Scala interesting as well.

All discussions (if there are any) see Google Group neo4j-scala

Building

$ git clone git://github.com/FaKod/neo4j-scala.git
$ cd neo4j-scala
$ mvn clean install

Or fetch it with Maven (the Sonatype Maven Repo is only needed if you want to use a SNAPSHOT version):

<repositories>
  <repository>
    <id>sonatype-snapshots</id>
    <url>https://oss.sonatype.org/content/repositories/snapshots/</url>
  </repository>
  ...
</repositories>

<dependencies>
  <dependency>
    <groupId>eu.fakod</groupId>
    <artifactId>neo4j-scala_2.10</artifactId>
    <version>0.3.0</version>
  </dependency>
</dependencies>

Troubleshooting

Please consider using Github issues tracker to submit bug reports or feature requests.

Using this library

Graph Database Service Provider

Neo4j Scala Wrapper needs a Graph Database Service Provider, it has to implement GraphDatabaseServiceProvider trait. One possibility is to use the EmbeddedGraphDatabaseServiceProvider for embedded Neo4j instances where you simply have to define a Neo4j storage directory. The class MyNeo4jClass using the wrapper is f.e.:

class MyNeo4jClass extends SomethingClass with Neo4jWrapper with EmbeddedGraphDatabaseServiceProvider {
  def neo4jStoreDir = "/tmp/temp-neo-test"
  . . .
}

Available are:

• EmbeddedGraphDatabaseServiceProvider
• SingletonEmbeddedGraphDatabaseServiceProvider (singleton version)
• BatchGraphDatabaseServiceProvider (use it with Neo4jBatchIndexProvider)
• REMOVED -> RestGraphDatabaseServiceProvider uses the REST binding

Transaction Wrapping

Transactions are wrapped by withTx. After leaving the "scope" success is called (or rollback if an exception is raised):

withTx {
 implicit neo =>
   val start = createNode
   val end = createNode
   start --> "foo" --> end
}

Using an Index

Neo4j provides indexes for nodes and relationships. The indexes can be configured by mixing in the Neo4jIndexProvider trait. See Indexing

class MyNeo4jClass extends . . . with Neo4jIndexProvider {
  // configuration for the index being created.
  override def NodeIndexConfig = ("MyTest1stIndex", Some(Map("provider" -> "lucene", "type" -> "fulltext"))) ::
                                 ("MyTest2ndIndex", Some(Map("provider" -> "lucene", "type" -> "fulltext"))) :: Nil
}

Use one of the configured indexes with

val nodeIndex = getNodeIndex("MyTest1stIndex").get

Add and remove entries by:

nodeIndex += (Node_A, "title", "The Matrix")
nodeIndex -= (Node_A)

Relations

Using this wrapper, this is how creating two relationships can look in Scala. The String are automatically converted into Dynamic Relationships:

start --> "KNOWS" --> intermediary --> "KNOWS" --> end
left --> "foo" --> middle <-- "bar" <-- right

To return the Property Container for the Relation Object use the '<' method:

val relation = start --> "KNOWS" --> end <

Properties

And this is how getting and setting properties on a node or relationship looks like :

// setting the property foo
start("foo") = "bar"
// cast Object to String and match . . .
start[String]("foo") match {
  case Some(x) => println(x)
  case None => println("aww")
}

Using Case Classes

Neo4j provides storing keys (String) and values (Object) into Nodes. To store Case Classes the properties are stored as key/values to the Property Container, they can be a Node or a Relation. However, Working types are limited to basic types like String, Integer etc.

case class Test(s: String, i: Int, ji: java.lang.Integer, d: Double, l: Long, b: Boolean)

. . .
withTx {
  implicit neo =>
    // create new Node with Case Class Test
    val node1 = createNode(Test("Something", 1, 2, 3.3, 10, true))

    // can Test be created from node
    val b:Boolean = node.toCCPossible[Test]

    // or using Option[T] (returning Some[T] if possible)
    val nodeOption: Option[Test] = node.toCC[Test]

    // yield all Nodes that are of type Case Class Test
    val tests = for(n <- getTraverser; t <- n.toCC[Test]) yield t

    // create new relation with Case Class Test
    node1 --> "foo" --> node2 < Test("other", 0, 1, 1.3, 1, false)
 }

Traversing

Besides, the neo4j scala binding makes it possible to write stop and returnable evaluators in a functional style :

//StopEvaluator.END_OF_GRAPH, written in a Scala idiomatic way :
start.traverse(Traverser.Order.BREADTH_FIRST, (tp : TraversalPosition) => false, 
ReturnableEvaluator.ALL_BUT_START_NODE, "foo", Direction.OUTGOING)

//ReturnableEvaluator.ALL_BUT_START_NODE, written in a Scala idiomatic way :
start.traverse(Traverser.Order.BREADTH_FIRST, StopEvaluator.END_OF_GRAPH, (tp : TraversalPosition) => tp.notStartNode(), 
"foo", Direction.OUTGOING)

Typed Traversing

The traverser mentioned above processes and returns Nodes resp. Property Container. To allow a more type safe traverser the TypedTraverser was introduced. The basic semantic is that you have to define the type (a case class) that should be returned (while inheritance is respected). But first define Relation Types and Directions:

Relation Types and Direction

To define relation types and directions use the follow method. Some examples

follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY" // BOTH for "KNOWS", OUTGOING for "CODED_BY"
follow -<- "BAR" -- "FOO"                       // INCOMING for "BAR", BOTH for "FOO", defaults to DEPTH_FIRST
follow(DEPTH_FIRST) ->- "FOOBAR"                // OUTGOING for "FOOBAR"

So the traverser, that returns an Iterable[MatrixBase], can be written like this:

myNode.doTraverse[MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY") {
    . . .block1. . .
} {
    . . .block2. . .
}

Return and Stop Evaluator

block1 is the Stop Evaluator and block2 the Return Evaluator, both of type PartialFunction[(T, TraversalPosition), Boolean]. Where T is MatrixBase in case of the example, TraversalPosition from the traverser and Boolean as return type.

PartialFunctions can be handled with a case statement, like this:

node.doTraverse[MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY") {
   		END_OF_GRAPH // predefined partial function
 	} {
		// if node is of type Matrix and TraversalPosition.depth == 2 then check lenth
   		case (x: Matrix, tp) if (tp.depth == 2) => x.name.length > 2

		// if node is of type NonMatrix then false
   		case (y: NonMatrix, _) => false
 	}

Assuming that Matrix and NonMatrix are inherited from MatrixBase the traverser will handle that correctly. X and y only matches if the given type can be marshaled to Matrix or NonMatrix and assigned from MatrixBase. Additionally, you can use the TraversalPosition in both case statements or ignore it with '_'. The example defines that depth must be 2 and the Matrix.name parameter must be of length > 2. NonMatrix classes are not returned. The Stop Evaluator always returns false.

Iterable[T]

The example above returns an Iterable[MatrixBase]. This allows to use the powerful Scala collections, f.e.:

...}.toList.sortWith(_.name < _.name) // sorts Nodes by name

...}.toList.flatMap(_.name) // List(N, e, o, M, o, r, p, h, e, u, s)

...}.toList.par. . . // do something with parallel collections

Where '_' is automatically replaced by MatrixBase instances.

Finally we can write, f.e.:

val list:List[MatrixBase] = node.doTraverse[MatrixBase](follow -<- "KNOWS") {
    case _ => false
} {
    case (x: Matrix, _) => true
    case (x: NonMatrix, _) => false
}.toList.sortWith(_.name < _.name)

Using a List of Nodes

Instead of one Node you can use a List of Nodes (List[Node]). The given traverser is started multithreaded for every node in the List. The resulting threads are joined, the result-lists are appended and Node duplicates removed. F.e:

val erg1 = startWithNodes.doTraverse[MatrixBase](follow -<- "KNOWS") {
    case _ => false
  } {
    case (x: Matrix, _) => true
    case (x: NonMatrix, _) => false
  }.toList.sortWith(_.name < _.name)

Where startWithNodes is of type List[Node].

REMOVED -> REST Typed Traversing <-- REMOVED

The main difference between the non-REST Typed Traverser is the ability to provide server side Prune Evaluator and Return Filter. This is important because otherwise all traversed data will be transfered to the client. This is possible but not always the best solution.

Prune Evaluator and Max Depth

The PruneEvaluator defines where to stop traversing relations. It has to be Java Script code that can use the position instance of type org.neo4j.graphdb.Path.

max depth is a short-hand way of specifying a prune evaluator which prunes after a certain depth. If not specified

• a max depth of 1 is used and
• if a "prune evaluator" is specified instead of a max depth, no max depth limit is set.

Examples for Prune Evaluator / MaxDepth

Using the case class PruneEvaluator ("JAVASCRIPT" is dafault, "false" is Java Script code):

startNode.doTraverse[Test_MatrixBase](follow -- "KNOWS" ->- "CODED_BY") {
  PruneEvaluator("false")
} {
  case (x: Test_Matrix, tp) if (tp.depth == 3) => x.name.length > 2
  case (x: Test_NonMatrix, _) => false
}.toList.sortWith(_.name < _.name)

Using a Java Script prune evaluator as a String (implicit conversion involved)

startNode.doTraverse[Test_MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY") {
  "position.length() > 100;"
} {
  case (x: Test_Matrix, tp) if (tp.depth == 2) => x.name.length > 2
  case (x: Test_NonMatrix, _) => false
}.toList.sortWith(_.name < _.name)

Using MaxDepth 100:

startNode.doTraverse[Test_MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY")(100) {
  case (x: Test_Matrix, _) => x.name.length > 2
}.toList.sortWith(_.name < _.name)

Return Filter

The Return Filter has the same semantic as the Return Evaluator. It can be used as with the normal TypedTraverser, can be used with the Java Script and with two builtin functions:

• ReturnAllButStartNode
• ReturnAll

Examples for Return Filter

Traversing with Max Depth 10 and all nodes except start node:

startNode.
       doTraverse[…](follow(BREADTH_FIRST) -- "KNOWS", 10, ReturnAllButStartNode)

Using Java Script Prune Evaluator ("true"):

startNode.
       doTraverse[…](follow(BREADTH_FIRST) ->- "CODED_BY", 1, "true")

Server Side type check and Max Depth 3:

startNode.doTraverse[Test_MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY", 3,
    endNode.isOfType[Test_Matrix]
  ).toList.sortWith(_.name < _.name)

Server Side type check and Server Side Java Script Prune Evaluator

startNode.doTraverse[Test_MatrixBase](follow(BREADTH_FIRST) -- "KNOWS" ->- "CODED_BY",
    "position.length() >= 1",
    endNode.isOfType[Test_Matrix]
  ).toList.sortWith(_.name < _.name)

Simple Cypher Support

The following example shows how to use Cypher together with typed results. In this case "execute" returns the case class Test_Matrix.

class MyClass extends Neo4jWrapper with SingletonEmbeddedGraphDatabaseServiceProvider with Cypher {
	. . .
	val query = "start n=node(" + nodeId + ") return n, n.name"

    val typedResult = query.execute.asCC[Test_Matrix]("n")

    typedResult.next.name must be_==("Neo")
    . . .
}

Batch Processing

Neo4j has a batch insertion mode intended for initial imports, which must run in a single thread and bypasses transactions and other checks in favor of performance. See Batch insertion.

The Java interfaces are slightly different. I wrote some wrapper classes to support nearly transparent usage of batch node and batch relation insertion. Means same code for batch insertion and for normal non batch mode. Instead of using

class Builder extends Neo4jWrapper with SingletonEmbeddedGraphDatabaseServiceProvider with Neo4jIndexProvider {...}

simply exchange the provider traits with

class Builder extends Neo4jWrapper with Neo4jBatchIndexProvider with BatchGraphDatabaseServiceProvider {...}

getting the indexes is still the same code

val nodeIndex = getNodeIndex("NodeIndex").get
val relationIndex = getRelationIndex("RelationIndex").get

setting cache size:

nodeIndex.setCacheCapacity("NodeIndex", 1000000)
relationIndex.setCacheCapacity("RelationIndex", 1000000)

Nevertheless, indexes are not available till flushing. To flush call:

nodeIndex.flush
relationIndex.flush

After insertion, the batch index manager and batch insertion manager have to be shut down

class Builder extends Neo4jWrapper . . .{
	. . .
	shutdownIndex
	shutdown(ds)
	. . .
}

Copyright and License

Copyright (C) 2012 Christopher Schmidt

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.