Neo4j.rb is a graph database for JRuby.
It provides:
-
Mapping of ruby objects to nodes in networks rather than in tables.
-
Dynamic, and schema-free - no need to declare nodes/properties/relationships in advance.
-
Storage of ruby object to a file system.
-
Fast traversal of relationships between nodes in a huge node space.
-
Transaction with rollbacks support.
-
Indexing and querying of ruby objects.
-
Can be used instead of ActiveRecord in Ruby on Rails or Merb
-
Can be accessible as REST resources.
It uses two powerful and mature java libraries:
-
Neo4J (www.neo4j.org/) - for persistance and traversal of the graph
-
Lucene (lucene.apache.org/java/docs/index.html) for quering and indexing.
-
There are over 300 RSpecs.
-
Has been tested with a simple rails application, used Neo4j.rb instead of ActiveRecord
-
Has been load tested (loaded 18000 nodes and done queries/traversal in several threads.)
-
Issue Tracking - neo4j.lighthouseapp.com
-
Twitter - twitter.com/ronge
-
API Documentation - neo4j.rubyforge.org/ (of the released version)
-
Source repo - git://github.com/andreasronge/neo4j.git
-
Mailing list - groups.google.com/group/neo4jrb (neo4jrb@googlegroups.com)
-
Ruby Manor 2008 - Jonathan Conway: jaikoo.com/assets/presentations/neo4j.pdf
-
Neo4j wiki - wiki.neo4j.org/content/Main_Page (check the guidelines and domain modeling gallery pages)
Have you found a bug, need help or have a patch ? Just clone neo4j.rb and send me a pull request or email me. Do you need help - send me an email (andreas.ronge at gmail dot com). Please also check/add issues at lighthouse, neo4j.lighthouseapp.com
-
Neo4j.rb - MIT, see the LICENSE file github.com/andreasronge/neo4j/tree/master/LICENSE.
-
Lucene - Apache, see lucene.apache.org/java/docs/features.html
-
Neo4j - Dual free software/commercial license, see neo4j.org/
This page contains the following information:
-
Installation guide
-
Three Minute Tutorial
-
Ten Minute Tutorial
-
Neo4j API Documentation
-
Extension: REST (see Neo4j::RestMixin)
-
Extension: find_path
-
Ruby on Rails with Neo4j.rb
-
Lucene API Documentation
There are also some complete examples in the example folder
-
admin - an uncomplete admin web gui for the Neo4j.rb/REST interface
-
railway - an example of a railway network application
-
imdb - an example of a neo4j database consisting of movies, role and actors nodes/relationships (over 18000 nodes).
-
rest - an example how to expose neo4j nodes as REST resources
-
Ruby on Rails - see github.com/andreasronge/neo4j-rails-example/tree/master
To install it:
gem install neo4j
To install from the latest source:
git clone git://github.com/andreasronge/neo4j.git cd neo4j rake gem:install
This has been verified to work on JRuby 1.3.0
To check that neo4j.rb is working:
cd neo4j # the folder containing the Rakefile rake # you may have to type jruby -S rake depending how you installed JRuby
Neo node space consists of three basic elements: nodes, relationships that connect nodes and properties attached to both nodes and relationships. All relationships have a type, for example if the node space represents a social network, a relationship type could be KNOWS. If a relationship of the type KNOWS connects two nodes, that probably represents two people that know each other. A lot of the semantics, the meaning, of a node space is encoded in the relationship types of the application.
Example of creating a neo4j node:
require "rubygems" require 'neo4j' Neo4j::Transaction.run do node = Neo4j::Node.new end
Almost all Neo4j operation must be wrapped in a transaction as shown above. In all the following examples we assume that the operations are inside an Neo4j transaction. There are two ways of creating transaction - in a block or the Transaction.new method
Using a block:
Neo4j::Transaction.run do # neo4j operations goes here end
Using the Neo4j::Transaction#new and Neo4j::Transaction#finish methods:
Neo4j::Transaction.new # neo4j operations goes here Neo4j::Transaction.finish
Example of setting properties
node = Neo4j::Node.new node[:name] = 'foo' node[:age] = 123 node[:hungry] = false node[4] = 3.14
Example of getting properties
node[:name] # => 'foo'
Example of creating a relationship
node1 = Neo4j::Node.new node2 = Neo4j::Node.new node1.relationships.outgoing(:friends) << node2 # connects node1 to node2 with relationship type 'friends'
Example of getting relationships
node1.relationships.empty? # => false # The relationships method returns an enumeration of relationship objects. # The nodes method on the relationships returns the nodes instead. node1.relationships.nodes.include?(node2) # => true node1.relationships.first # => the first relationship this node1 has which is between node1 and node2 of any type node1.relationships.nodes.first # => node2 first node of any relationship type node1.relationships.incoming(:friends).nodes.first # => node1 first node of relationship type 'friends' node2.relationships.incoming(:friends).first # => a relationship object between node1 and node2
Example of setting properties on relationships
rel = node1.relationships.outgoing(:friends).first rel[:since] = 1982 node1.relationships.first[:since] # => 1982 (there is only one relationship defined on node1 in this example)
Here is the complete implemenation of the Neo4j::Node class
module Neo4j class Node include Neo4j::NodeMixin end end
In the next tutorial you will learn how to use this Neo4j::NodeMixin in your own domain model classes.
The following example specifies how to map a Neo4j node to a Ruby Person instance.
require "rubygems" require "neo4j" class Person include Neo4j::NodeMixin # define Neo4j properties property :name, :salary, :age # define an one way relationship to any other node has_n :friends # adds a lucene index on the following properties index :name, :salary index 'friends.age' # index each friend age as well end
Neo properties and relationships are declared using the ‘property’ and ‘has_n’/‘has_one’ NodeMixin class method. Adding new types of properties and relationships can also be done without declaring those properties/relationships by using the operator ‘[]’ on Neo4j::NodeMixin and the ‘<<’ on the Neo4j::Relationships::RelationshipTraverser.
By using the NodeMixin and by declaring properties and indices, all instances of the Person class can now be stored in the Neo4j node space and be retrieved/queried by traversing the node space or performing Lucene queries.
A lucene index will be updated when the name or salary property changes. The salary of all friends are also indexed which means we can query for people who have friends with a certain salary.
Creating a Person node instance
person = Person.new
Setting a property:
person.name = 'kalle' person.salary = 10000
Notice that it is not required to specify which attributes should be available on a node. Any attributes can be set using the [] operator. Declared properties set an expectation, not an requirement. It can be used for documenting your model objects and catching typos.
Example:
person['an_undefined_property'] = 'hello'
So, why declare properties in the class at all? By declaring a property in the class, you get the sexy dot notation. But also, if you declare a Lucene index on the declared property and update the value, then the Lucene index will automatically be updated. The property declaration is required before declaring an index on the property.
Like properties, relationships do not have to be defined using has_n or has_one for a class. A relationship can be added at any time on any node.
Example:
person.relationships.outgoing(:best_friends) << other_node person.relationship(:best_friend).end_node # => other_node (if there is only one relationship of type 'best_friend' on person)
There are three ways of finding/quering nodes in neo4j:
1. by traversing the graph 2. by using lucene queries 3. using the unique neo4j id (Neo4j::NodeMixin#neo_node_id).
When doing a traversal one starts from a node and traverses one or more relationships (one or more levels deep). This start node can be either the reference node which is always found (Neo4j.ref_node) or by finding a start node from a lucene query.
There are different ways to write lucene queries. Using a hash:
Person.find (:name => 'kalle', :age => 20..30) # find people with name kalle and age between 20 and 30
or using the lucene query language:
Person.find("name:kalle AND salary:[10000 TO 30000]")
The Lucene query language supports wildcard, grouping, boolean, fuzzy queries, etc… For more information see: lucene.apache.org/java/2_4_0/queryparsersyntax.html
Person.find(:name => 'kalle').sort_by(:salary) Person.find(:name => 'kalle').sort_by(Desc[:salary], Asc[:country]) Person.find(:name => 'kalle').sort_by(Desc[:salary, :country])
The query is not performed until the search result is requested. Example of using the search result.
res = Person.find(:name => 'kalle') res.size # => 10 res.each {|x| puts x.name} res[0].name = 'sune'
Adding a relationship between two nodes:
person2 = Person.new person.friends << person2
The person.friends returns an object that has a number of useful methods (it also includes the Enumerable mixin). Example
person.friends.empty? # => false person.friends.first # => person2 person.friends.include?(person2) # => true
To delete the relationship between person and person2:
person.relationships[person2].delete
If a node is deleted then all its relationship will also be deleted Deleting a node is performed by using the delete method:
person.delete
The has_one and has_many methods create a convenient method for traversals and managing relationships to other nodes. Example:
Person.has_n :friends # generates the friends instance method
# all instances of Person now has a friends method so that we can do the following
person.friends.each {|n| ... }
Traversing using a filter
person.friends{ salary == 10000 }.each {|n| ...}
Traversing with a specific depth (depth 1 is default)
person.friends{ salary == 10000}.depth(3).each { ... }
There is also a more powerful method for traversing several relationships at the same time - Neo4j::NodeMixin#traverse, see below.
In the first example the friends relationship can have relationships to any other node of any class. In the next example we specify that the ‘acted_in’ relationship should use the Ruby classes Actor, Role and Movie. This is done by using the has_n class method:
class Role include Neo4j::RelationshipMixin # notice that neo4j relationships can also have properties property :name end class Actor include Neo4j::NodeMixin # The following line defines the acted_in relationship # using the following classes: # Actor[Node] --(Role[Relationship])--> Movie[Node] # has_n(:acted_in).to(Movie).relationship(Role) end class Movie include Neo4j::NodeMixin property :title property :year # defines a method for traversing incoming acted_in relationships from Actor has_n(:actors).from(Actor, :acted_in) end
Creating a new Actor-Role-Movie relationship can be done like this:
keanu_reeves = Actor.new matrix = Movie.new keanu_reeves.acted_in << matrix
or you can also specify this relationship on the incoming node (since we provided that information in the has_n methods).
keanu_reeves = Actor.new matrix = Movie.new matrix.actors << keanu_reeves
Example of accessing the Role relationship object between an Actor and a Movie
keanu_reeves.relationships.outgoing(:acted_in)[matrix].name = 'neo'
More information about neo4j can be found after the Lucene section below.
The Neo4j module is used to map Ruby objects to nodes and relationships in a network. It supports two different ways of retrieval/quering:
-
Neo4j traversal, Neo4j::NodeMixin#traverse (or Neo4j::NodeMixin#has_n)
-
Lucene indexes, Neo4j::NodeMixin#find
Unlike the Java Neo4j implementation it is not neccessarly to start Neo4j. It will automatically be started when needed. It also uses a hook to automatically shutdown Neo4j. Shutdown of neo4j can also be done using the stop method, example:
Neo4j.stop
Before using Neo4j the location where the database is stored on disk should be configured. The neo4j configuration is kept in the Neo4j::Config class:
Neo4j::Config[:storage_path] = '/home/neo/neodb'
Neo4j.rb uses the Lucene module. That means that the Neo4j::NodeMixin has method for both traversal and lucene queries/indexing.
By default lucene indexes are kept in memory. Keeping index in memory will increase the performance of lucene operations (such as updating the index).
Example to configure Lucene to store indexes on disk instead
Lucene::Config[:store_on_file] = true Lucene::Config[:storage_path] => '/home/neo/lucene-db'
If index is stored in memory then one needs to reindex all nodes when the application starts up again.
MyNode.update_index # will traverse all MyNode instances and (re)create the lucene index in memory.
Neo4j::NodeMixin is a mixin that lets instances to be stored as a node in the neo node space on disk. A node can have properties and relationships to other nodes.
Example of how declare a class that has this behaviour:
class MyNode include Neo4j::NodeMixin end
node = MyNode.new
The Neo4j::NodeMixin mixin defines a delete method that will delete the node and all its relationships.
Example:
node = MyNode.new node.delete
The node in the example above will be removed from the neo database on the filesystem and the lucene index
In order to use properties they have to be declared first
class MyNode include Neo4j::NodeMixin property :foo, :bar end
These properties (foo and bar) will be stored in the Neo database. You can set those properties:
# create a node with two properties in one transaction node = MyNode.new { |n| n.foo = 123 n.bar = 3.14 } # access those properties puts node.foo
You can also set a property like this:
f = SomeNode.new f.foo = 123
Neo4j.rb supports properties to by of type String, Fixnum, Float and true/false
If you want to set a property of a different type then String, Fixnum, Float or true/false you have to specify its type.
Example, to set a property to any type
class MyNode include Neo4j::NodeMixin property :foo, :type => Object end node = MyNode.new node.foo = [1,"3", 3.14] Neo4j.load(node.neo_node_id).foo.class # => Array
Example of using Date queries:
class MyNode include Neo4j::NodeMixin property :since, :type => Date index :since, :type => Date end node.since = Date.new 2008,05,06 MyNode.find("born:[20080427 TO 20100203]")[0].since # => Date 2008,05,06
Example of using DateTime queries:
class MyNode include Neo4j::NodeMixin property :since, :type => DateTime index :since, :type => DateTime end node.since = DateTime.civil 2008,04,27,15,25,59 MyNode.find("since:[200804271504 TO 201002031534]")[0].since # => DateTime ...
Only UTC timezone is allowed.
To find all nodes use the Neo4j#all_nodes method.
Example
Neo4j.all_nodes.each {|node| puts node}
Neo relationships are asymmetrical. That means that if A has a relationship to B then it may not be true that B has a relationship to A.
Relationships can be declared by using the ‘has_n’ or ‘has_one’ Neo4j::NodeMixin class methods.
The has_n Neo4j::NodeMixin class method creates a new instance method that can be used for both traversing and adding new objects to a specific relationship type.
For example, let say that Person can have a relationship to any other node class with the type ‘friends’:
class Person include Neo4j::Node has_n :knows # will generate a knows method for outgoing relationships end
The generated knows method will allow you to add new relationships, example:
me = Person.new neo = Person.new me.knows << neo # me knows neo but neo does not know me
You can add any object to the ‘knows’ relationship as long as it includes the Neo4j::NodeMixin, example:
person = Person.new car = Volvo.new # Volvo is a class that includes the Neo4j::NodeMixin person.knows << car
If you want to express that the relationship should point to a specific class use the ‘to’ method on the has_n method.
class Person include Neo4j::Node has_n(:knows).to(Person) end
It is also possible to generate methods for incoming relationships by using the ‘from’ method on the has_n method.
Example:
class Person include Neo4j::Node has_n :knows # will generate a knows method for outgoing relationships has_n(:known_by).from(:knows) # will generate a known_by method for incomming knows relationship end
By doing this you can add a relationships on either the incoming or outgoing node. The from method can also take an additional class parameter if it has incoming nodes from a different node class (see the Actor-Role-Movie example at the top of this document).
Example of adding a ‘knows’ relationship from the other node:
me = Person.new neo = Person.new neo.known_by << me # me knows neo but neo does not know me
The known_by method creates a ‘knows’ relationship between the me and neo nodes. This is the same as doing:
me.knows << neo # me knows neo but neo does not know me
Example of has_one: A person can have at most one Address
class Person include Neo4j::NodeMixin has_one(:address).to(Address) end class Address include Neo4j::NodeMixin property :city, :road has_n(:people).from(Person, :address) end
In the example above we have Neo4j.rb will generate the following methods
-
in Person, the method ”address=” and ”address”
-
in Address, the traversal method ”people” for traversing incomming relationships from the Person node.
Example of usage:
p = Person.new p.address = Address.new p.address.city = 'malmoe'
Or from the incoming ”address” relationship
a = Address.new {|n| n.city = 'malmoe'} a.people << Person.new
The has_n relationship will not maintain the order of when items are inserted to the relationship. If order should be preserved then use the has_list class method instead.
Example
class Company has_list :employees end company = Company.new company.employees << employee1 << employee2 # prints first employee2 and then employee1 company.employees.each {|employee| puts employee.name}
If the optional parameter :size is given then the list will contain a size counter.
Example
class Company has_list :employees, :counter => true end company = Company.new company.employees << employee1 << employee2 company.employees.size # => 2
The list will be updated if an item is deleted in a list. Example:
company = Company.new company.employees << employee1 << employee2 << employee3 company.employees.size # => 3 employee2.delete company.employees.to_a # => [employee1, employee3] company.employees.size # => 2
Each node in a list knows which lists it belongs to the next and previous item in the list Example: employee1.list(:employees).next # => employee2 employee2.list(:employees).next = Neo4j::Node.new employee1.list(:employees).size # => 2 # the size counter is available if the :counter parameter is given as shown above
(The list method takes an optional extra parameter - the list node. Needed if one node is member of more then one list with the same name).
Each type of relationship has a method that returns an Enumerable object that enables you to traverse that type of relationship.
For example the Person example above declares one relationship of type friends. You can traverse all Person’s friend (depth 1 is default)
f.friends.each { |n| puts n }
It is also possible to traverse a relationship of an arbitrary depth. Example finding all friends and friends friends.
f.friends.depth(2).each { ...}
Traversing to the end of the graph
f.friends.depth(:all).each { ...}
If you want to find one node in a relationship you can use a filter. Example, let say we want to find a friend with name ‘andreas’
n1 = Person.new n2 = Person.new {|n| n.name = 'andreas'} n3 = Person.new n1.friends << n2 << n3 n1.friends{ name == 'andreas' }.to_a # => [n2]
The block { name == ‘andreas’ } will be evaluated on each node in the relationship. If the evaluation returns true the node will be included in the filter search result.
The Neo4j::NodeMixin#traverse method is a more powerful method compared to the generated has_n and has_one methods. Unlike those generated method it can traverse several relationship types at the same time. The types of relationships being traversed must therefore always be specified in the incoming, outgoing or both method. Those three methods can take one or more relationship types parameters if more then one type of relationship should be traversed.
The depth method allows you to specify how deep the traverse should be. If not specified only one level traverse is done.
Example:
me.traverse.incoming(:friends).depth(4).each {} # => people with a friend relationship to me
It is possible to traverse sevaral relationship types at the same type. The incoming, both and outgoing methods takes list of arguments.
Example, given the following holiday trip domain:
# A location contains a hierarchy of other locations # Example region (asia) contains countries which contains cities etc... class Location include Neo4j::NodeMixin has_n :contains has_n :trips property :name index :name # A Trip can be specific for one global area, such as "see all of sweden" or # local such as a 'city tour of malmoe' class Trip include Neo4j::NodeMixin property :name end # create all nodes # ... # setup the relationship between all nodes @europe.contains << @sweden << @denmark @sweden.contains << @malmoe << @stockholm @sweden.trips << @sweden_trip @malmoe.trips << @malmoe_trip @malmoe.trips << @city_tour @stockholm.trips << @city_tour # the same city tour is available both in malmoe and stockholm
Then we can traverse both the contains and the trips relationship types Example:
@sweden.traverse.outgoing(:contains, :trips).to_a # => [@malmoe, @stockholm, @sweden_trip]
It is also possible to traverse both incoming and outgoing relationships, example:
@sweden.traverse.outgoing(:contains, :trips).incoming(:contains).to_a # => [@malmoe, @stockholm, @sweden_trip, @europe]
It’s possible to filter which nodes should be returned from the traverser by using the filter function. This filter function will be evaluated differently depending on if it takes one argument or no arguments, see below.
If the provided filter function does not take any parameter it will be evaluted in the context of the current node being traversed. That means that one can writer filter functions like this:
@sweden.traverse.outgoing(:contains, :trips).filter { name == 'sweden' }
If the filter method takes one parameter then it will be given an object of type TraversalPosition which contains information about current node, how many nodes has been returned, depth etc.
The information contained in the TraversalPostion can be used in order to decide if the node should be included in the traversal search result. If the provided block returns true then the node will be included in the search result.
The filter function will not be evaluated in the context of the current node when this parameter is provided.
The TraversalPosition is a thin wrapper around the java interface TraversalPosition, see api.neo4j.org/current/org/neo4j/api/core/TraversalPosition.html
For example if we only want to return the Trip objects in the example above:
# notice how the tp (TraversalPosition) parameter is used in order to only # return nodes included in a 'trips' relationship. traverser = @sweden.traverse.outgoing(:contains, :trips).filter do |tp| tp.last_relationship_traversed.relationship_type == :trips end traverser.to_a # => [@sweden_trip]
A relationship between two nodes can have properties just like a node.
Example:
p1 = Person.new p2 = Person.new relationship = p1.friends.new(p2) # set a property 'since' on this relationship bewteen p1 and p2 relationship.since = 1992
If a Relationship class has not been specified for a relationship then any properties can be set on the relationship. It has a default relationship class: Neo4j::Relationships::Relationship
If you instead want to use your own class for a relationship use the Neo4j::NodeMixin#has_n.relationship method, example:
class Role # This class can be used as the relationship between two nodes # since it includes the following mixin include Neo4j::RelationMixin property :name end class Actor include Neo4j::NodeMixin # use the Role class above in the relationship between Actor and Movie has_n(:acted_in).to(Movie).relationship(Role) end
The Neo4j::NodeMixin#relationships method can be used to find incoming or outgoing relationship objects. Example of listing all types of outgoing (default) relationship objects (of depth one) from the me node.
me.relationships.each {|rel| ... }
If we instead want to list the nodes that those relationships points to then the nodes method can be used.
me.relationships.nodes.each {|rel| ... }
Listing all incoming relationship obejcts of any relationship type:
me.relationships.incoming.each { ... }
Listing both incoming and outgoing relationship object of a specific type:
me.relationships.both(:friends) { }
Finding one outgoing relationship of a specific type and node (you)
me.relationships.outgoing(:friends)[you] # => [#<Neo4j::RelationshipMixin:0x134ae32]
Example, given we have the two nodes with a relationship between them:
n1 = Person.new n2 = Person.new n1.friends << n2
Then we can find all incoming and outgoing relationships like this:
n1.relationships.to_a # => [#<Neo4j::RelationshipMixin:0x134ae32]
A Neo4j::RelationshipMixin object represents a relationship between two nodes.
n1.relationships[0].start_node # => n1 n1.relationships[0].end_node # => n2
A RelationshipMixin contains the relationship type which connects it connects two nodes with, example:
n1.relationships[0].relationship_type # => :friends
Relationships can also have properties just like a node (NodeMixin).
If we are only interested in all incoming nodes, we can do
n2.relationships.incoming # => [#<Neo4j::RelationshipMixin:0x134ae32]
Or outgoing:
n1.relationships.outgoing # => [#<Neo4j::RelationshipMixin:0x134aea2]
To find a specific relationship use the [] operator:
n1.relationships.outgoing[n2] = #<Neo4j::RelationshipMixin:0x134aea2
Or which is better performance wise (since only friends relationships are being traversed):
n1.relationships.outgoing(:friends)[n2] = #<Neo4j::RelationshipMixin:0x134aea2
Use the Neo4j::RelationshipMixin#delete method. For example, to delete the relationship between n1 and n2 from the example above:
n1.relationships.outgoing(:friends)[n2].delete
If you do not want those relationship object but instead want the nodes you can use the ‘nodes’ method in the Neo4j::RelationshipMixin object.
For example:
n2.relationships.incoming.nodes # => [n1]
Let say we want to find who has my phone number and who consider me as a friend
# who has my phone numbers me.relationships.incoming(:phone_numbers).nodes # => people with my phone numbers # who consider me as a friend me.relationships.incoming(:friends).nodes # => people with a friend relationship to me
Remember that relationships are not symmetrical. Notice there is also a otherway of finding nodes, see the Neo4j::NodeMixin#traverse method below.
All operations that work with the node space (even read operations) must be wrapped in a transaction. For example all get, set and find operations will start a new transaction if none is already not runnig (for that thread).
If you want to perform a set of operation in a single transaction, use the Neo4j::Transaction.run method:
Example
Neo4j::Transaction.run { node1.foo = "value" node2.bar = "hi" }
There is also a auto commit feature available which is enabled by requiring ‘neo4j/auto_tx’ instead of ‘neo4j’, see the three minutes tutorial above.
Neo4j support rollbacks on transaction. Example: Example:
include 'neo4j' node = MyNode.new Neo4j::Transaction.run { |t| node.foo = "hej" # something failed so we signal for a failure t.failure # will cause a rollback, node.foo will not be updated }
You can also run it without a block, like this:
transaction = Neo4j::Transaction.new transaction.start # do something transaction.finish
Properties and relationships which should be indexed by lucene can be specified by the index class method. For example to index the proeprties foo and bar
class SomeNode include Neo4j::NodeMixin property :foo, :bar index :foo, :bar end
Everytime a node of type SomeNode (or a subclass) is create, deleted or updated the lucene index of will be updated.
Sometimes it’s neccessarly to change the index of a class after alot of node instances already have been created. To delete an index use the class method ‘remove_index’ To update an index use the class method ‘update_index’ which will update all already created nodes in the neo database.
Example
require 'neo4j'
require 'neo4j/extensions/reindexer' # needed for the update_index method
class Person
include Neo4j
property :name, :age, :phone
index :name, :age
end
p1 = Person.new {|n| n.name = 'andreas'; n.phone = 123}
Person.find (:name => 'andreas') # => [p1]
Person.find (:phone => 123) # => []
# change index and reindex all person nodes already created in the neo database.
Person.remove_index :name
Person.index :phone # add an index on phone
Person.update_index
Person.find (:name => 'andreas') # => []
Person.find (:phone => 123) # => [p1]
In order to use the update_index method you must include the reindexer neo4j.rb extension. This extensions will keep a relationship to each created node so that it later can recreate the index by traversing those relationships.
The lucene index will be updated after the transaction commits. It is not possible to query for something that has been created inside the same transaction as where the query is performed.
You can declare properties to be indexed by lucene by the index method:
Example
class Person include Neo4j::NodeMixin property :name, :age index :name, :age end node = Person.new node.name = 'foo' node.age = 42 Person.find(:name => 'foo', :age => 42) # => [node]
The query parameter (like property on a Neo4j::NodeMixin) can be of type String, Fixnum, Float, boolean or Range. The query above can also be written in a lucene query DSL:
Person.find{(name =='foo') & (age => 42)} # => [node]
Or lucene query language:
Person.find("name:foo AND age:42")
For more information see: lucene.apache.org/java/2_4_0/queryparsersyntax.html or the lucene module above.
In order to use range querie on numbers the property types must be converted. This is done by using the :type optional parameter:
class Person include Neo4j::NodeMixin property :name, :age index :age, :type => Fixnum end
By using :type => Fixnum the age will be padded with ‘0’s (lucene only support string comparsion).
Example, if the :type => Fixnum was not specified then
p = Person.new {|n| n.age = 100 } Person.find(:age => 0..8) # => [p]
The Neo4j::NodeMixin#index method can be used to index relationships to other classes.
Example, let say we have to classes, Customer and Orders:
class Customer include Neo4j::NodeMixin property :name # specifies outgoing relationships to Order has_n(:orders).to(Order) # create an index on customer-->order#total_cost index "orders.total_cost" end class Order include Neo4j::NodeMixin property :total_cost # specifies one incoming relationship from Customer has_one(:customer).from(Customer, :orders) # create an index on the order<--customer#name relationship index "customer.name" end
Notice that we can index both incoming and outgoing relationships.
Let’s create a customer and one order for that customer
Neo4j::Transaction.run do cust = Customer.new order = Order.new cust.name = "kalle" order.total_cost = "1000" cust.orders << order end
Now we can find both Orders with a total cost between 500 and 2000 and Customers with name ‘kalle’ using lucene
Example:
customers = Customer.find('orders.total_cost' => 500..2000, 'name' => 'kalle')
Or also possible from the other way:
orders = Order.find('total_cost' => 500..2000, 'customer.name' => 'kalle')
Neo4j supports full text search by setting the tokenized property to true on an index. (see JavaDoc for org.apache.lucene.document.Field.Index.ANALYZED).
class Comment include Neo4j::NodeMixin property :comment index comment, :tokenized => true end
The neo module will automatically unmarshalling nodes to the correct ruby class. It does this by reading the classname property and loading that ruby class with that node. If this classname property does not exist it will use the default Neo4j::Node for nodes and Neo4j::Relationships::Relationship for relationship.
class Person include Neo4j::Node def hello end end f1 = Person.new {} # load the class again f2 = Neo4j.load(foo.neo_node_id) # f2 will now be new instance of Person, but will be == f1 f1 == f2 # => true
There is one node that can always be find - the reference node, Neo4j::ReferenceNode. Example:
Neo4j.ref_node
This node can have a relationship to the index node (Neo4j::IndexNode), which has relationships to all created nodes. You can add relationships from this node to your nodes.
It is recommended to wrap several Neo4j operations including read operations in a singel transaction if possible for better performance. Updating a lucene index can be slow. A solution to this is to keep the index in memory instead of on disk.
I’m currently looking at how to scale neo4j.rb by a simple master-slave cluster by using REST, see the REST extension below.
There is an experimental extension that makes it possible to replicate an neo4j database to another machine. For example how to use it see the test/replication/test_master.rb and test_slave.rb It has only been tested to work with a very simple node space.
There is an REST extension to Neo4j.rb. It requires the following gems
* Sinatra >= 0.9.4 * Rack >= 1.0 * json-jruby >= 1.1.6
For RSpec testing it also needs:
* rack-test
For more information see the examples/rest/example.rb or the examples/admin or Neo4j::RestMixin.
Extension which finds the shortest path (in terms of number of links) between two nodes. Use something like this:
require 'neo4j/extensions/find_path' node1.traverse.both(:knows).depth(:all).path_to(node2) # => [node1, node42, node1234, node256, node2]
This extension is still rather experimental. The algorithm is based on the one used in the Neo4j Java IMDB example. For more information see Neo4j::Relationships::NodeTraverser#path_to or the RSpec find_path_spec.rb.
Neo4j.rb does work nicely with R&R. There are two ways to use neo4j.rb with rails - embedded or accessing it via REST.
A complete example of embedding neo4j with rails can be found github.com/andreasronge/neo4j-rails-example/tree/master (please fork and improve it).
Config rails to use Neo4j.rb instead of ActiveRecord, edit movies/config/environment.rb environment.rb:
config.frameworks -= [ :active_record ] #, :active_resource, :action_mailer ] config.gem "neo4j", :version => "0.3.1" # or the latest one
If you need to reindex all nodes or use the Neo4j::NodeMixin#all method you must require the reindexer neo4j.rb extension. Add a require in the environment.rb file:
require 'neo4j/extensions/reindexer'
Create a new file for each neo4j node or relationship class Example for an Actor class create the file: app/models/actor.rb
# filename app/models/actor.rb class Actor include Neo4j::NodeMixin property :name, :phone, :salary has_n(:acted_in).to(Movie).relationship(Role) index :name end
Edit the config/routes.rb file Example:
ActionController::Routing::Routes.draw do |map|
map.resources :actors do |actor|
actor.resources :acted_in
actor.resource :movies, :controller => 'acted_in'
end
Since all neo4j operations must be wrapped in a transaction add a around filter for all operations Example:
acted_in_controller.rb:
class ActedInController < ApplicationController around_filter :neo_tx def index @actor = Neo4j.load(params[:actor_id]) @movies = @actor.acted_in.nodes end def create @actor = Neo4j.load(params[:actor_id]) @movie = Movie.new @movie.update(params[:movie]) @actor.acted_in << @movie flash[:notice] = 'Movie was successfully created.' redirect_to(@actor) end def update @actor = Neo4j.load(params[:actor_id]) @movie = Movie.new @movie.update(params[:movie]) @actor.acted_in.new @movie @movie.update(params[:movie]) flash[:notice] = 'Movie was successfully updated.' redirect_to(@movie) end def show @movie = Neo4j.load(params[:id]) end def new @actor = Neo4j.load(params[:actor_id]) @movie = Movie.value_object.new end def edit @movie = Neo4j.load(params[:id]) end private def neo_tx Neo4j::Transaction.new yield Neo4j::Transaction.finish end end
Add the following views in app/views/actors index.html.erb:
<h1>Listing actors</h1>
<table>
<tr>
<th>Name</th>
</tr>
<% for actor in @actors %>
<tr>
<td><%=h actor.name %></td>
<td><%= link_to 'Edit', edit_actor_path(actor) %></td>
<td><%= link_to 'Show', actor %></td>
<td><%= link_to 'Destroy', actor, :confirm => 'Are you sure?', :method => :delete %></td>
</tr>
<% end %>
</table>
<br />
<%= link_to 'New actor', new_actor_path %>
new.html.erb:
<h1>New Actor</h1>
<% form_for(@actor) do |f| %>
<p>
<%= f.label :name %><br />
<%= f.text_field :name %>
</p>
<p>
<%= f.label :phone %><br />
<%= f.text_field :phone %>
</p>
<p>
<%= f.label :salary%><br />
<%= f.text_field :salary %>
</p>
<p>
<%= f.submit "Update" %>
</p>
<% end %>
<%= link_to 'Back', actors_path %>
You can use this module without using the Neo4j module.
Lucene provides:
-
Flexible Queries - Phrases, Wildcards, Compound boolean expressions etc…
-
Field-specific Queries eg. title, artist, album
-
Sorting
-
Ranked Searching
In lucene everything is a Document. A document can represent anything textual: Word Document, DVD (the textual metadata only), or a Neo4j.rb node. A document is like a record or row in a relationship database.
The following example shows how a document can be created by using the ”<<” operator on the Lucene::Index class and found using the Lucene::Index#find method.
Example of how to write a document and find it:
require 'lucene' include Lucene # the var/myindex parameter is either a path where to store the index or # just a key if index is kept in memory (see below) index = Index.new('var/myindex') # add one document (a document is like a record or row in a relationship database) index << {:id=>'1', :name=>'foo'} # write to the index file index.commit # find a document with name foo # hits is a ruby Enumeration of documents hits = index.find{name == 'foo'} # show the id of the first document (document 0) found # (the document contains all stored fields - see below) hits[0][:id] # => '1'
Notice that you have to call the commit method in order to update the index on the disk/RAM. By performing several update and delete operations before a commit will be much faster then performing commit after each operation.
By default Neo4j::Lucene keeps indexes in memory. That means that when the application restarts the index will be gone and you have to reindex everything again.
To keep indexes in memory:
Lucene::Config[:store_on_file] = true Lucene::Config[:storage_path] => '/home/neo/lucene-db'
When creating a new index the location of the index will be the Lucene::Config + index path Example:
Lucene::Config[:store_on_file] = true Lucene::Config[:storage_path] => '/home/neo/lucene-db' index = Index.new('/foo/lucene')
The example above will store the index at /home/neo/lucene-db/foo/lucene
Let say a person can have several phone numbers. How do we index that ?
index << {:id=>'1', :name=>'adam', :phone => ['987-654', '1234-5678']}
All Documents must have one id field. If one is not specified it is :id of type String. A different id can be specified using the field_infos id_field property on the index:
index = Index.new('some/path/to/the/index') index.field_infos.id_field = :my_id
To change the type of the my_id from String to a different type see below.
Lucene.rb can handle type conversion for you. (The java lucene library stores all the fields as Strings) For example if you want the id field to be a fixnum
require 'lucene' include Lucene index = Index.new('var/myindex') # store the index at dir: var/myindex index.field_infos[:id][:type] = Fixnum index << {:id=>1, :name=>'foo'} # notice 1 is not a string now index.commit # find that document, hits is a ruby Enumeration of documents hits = index.find(:name => 'foo') # show the id of the first document (document 0) found # (the document contains all stored fields - see below) doc[0][:id] # => 1
If the field_info type parameter is not set then it has a default value of String.
By default only the id field will be stored. That means that in the example above the :name field will not be included in the document.
Example
doc = index.find('name' => 'foo') doc[:id] # => 1 doc[:name] # => nil
Use the field info :store=true if you want a field to be stored in the index (otherwise it will only be searchable).
Example
require 'lucene' include Lucene index = Index.new('var/myindex') # store the index at dir: var/myindex index.field_infos[:id][:type] = Fixnum index.field_infos[:name][:store] = true # store this field index << {:id=>1, :name=>'foo'} # notice 1 is not a string now index.commit # find that document, hits is a ruby Enumeration of documents hits = index.find('name' => 'foo') # let say hits only contains one document so we can use doc[0] for that one # that document contains all stored fields (see below) doc[0][:id] # => 1 doc[0][:name] # => 'foo'
As shown above you can set field infos like this
index.field_infos[:id][:type] = Fixnum
Or you can set several properties like this:
index.field_infos[:id] = {:type => Fixnum, :store => true}
Lucene.rb support search in several fields: Example
# finds all document having both name 'foo' and age 42 hits = index.find('name' => 'foo', :age=>42)
Range queries
# finds all document having both name 'foo' and age between 3 and 30 hits = index.find('name' => 'foo', :age=>3..30)
If the query is string then the string is a lucene query.
hits = index.find('name:foo')
For more information see: lucene.apache.org/java/2_4_0/queryparsersyntax.html
The queries above can also be written in a lucene.rb DSL:
hits = index.find { (name == 'andreas') & (foo == 'bar')}
Expression with OR (|) is supported, example
# find all documents with name 'andreas' or age between 30 and 40 hits = index.find { (name == 'andreas') | (age == 30..40)}
Sorting is specified by the ‘sort_by’ parameter Example
hits = index.find(:name => 'foo', :sort_by=>:category)
To sort by several fields:
hits = index.find(:name => 'foo', :sort_by=>[:category, :country])
Example sort order
hits = index.find(:name => 'foo', :sort_by=>[Desc[:category, :country], Asc[:city]])
The Lucene::Index is thread safe. It guarantees that an index is not updated from two thread at the same time.
Use the Lucene::Transaction in order to do atomic commits. By using a transaction you do not need to call the Index.commit method.
Example:
Transaction.run do |t| index = Index.new('var/index/foo') index << { id=>42, :name=>'andreas'} t.failure # rollback end result = index.find('name' => 'andreas') result.size.should == 0
You can find which documents are uncommited by using the uncommited index property.
Example
index = Index.new('var/index/foo') index.uncommited #=> [document1, document2]
Notice that even if it looks like a new Index instance object was created the index.uncommited may return an not empty array. This is because Index.new is a singleton - a new instance object is not created.