Make your ActiveRecords stored and searchable on Neo4j graph database, in order to make fast graph queries that MySQL would crawl while doing them.
Neoid to Neo4j is like Sunspot to Solr. You get the benefits of Neo4j speed while keeping your schema on your plain old RDBMS.
Neoid doesn't require JRuby. It's based on the great Neography gem which uses Neo4j's REST API.
Neoid offers querying Neo4j for IDs of objects and then fetch them from your RDBMS, or storing all desired data on Neo4j.
Important: Heroku Support is not available because Herokud doesn't support Gremlin. So until further notice, easiest way is to self host a Neo4j on EC2 in the same zone, and connect from your dyno to it
See Changelog. Including some breaking changes (and solutions) from previos versions.
Add to your Gemfile and run the bundle
command to install it.
gem 'neoid', '~> 0.1.1'
Requires Ruby 1.9.2 or later.
In an initializer, such as config/initializers/01_neo4j.rb
:
ENV["NEO4J_URL"] ||= "http://localhost:7474"
uri = URI.parse(ENV["NEO4J_URL"])
$neo = Neography::Rest.new(uri.to_s)
Neography.configure do |c|
c.server = uri.host
c.port = uri.port
if uri.user && uri.password
c.authentication = 'basic'
c.username = uri.user
c.password = uri.password
end
end
Neoid.db = $neo
Neoid.configure do |c|
# should Neoid create sub-reference from the ref node (id#0) to every node-model? default: true
c.enable_subrefs = true
end
01_
in the file name is in order to get this file loaded first, before the models (initializers are loaded alphabetically).
If you have a better idea (I bet you do!) please let me know.
For nodes, first include the Neoid::Node
module in your model:
class User < ActiveRecord::Base
include Neoid::Node
end
This will help to create/update/destroy a corresponding node on Neo4j when changed are made a User model.
Then, you can customize what fields will be saved on the node in Neo4j, inside neoidable
configuration, using field
. You can also pass blocks to save content that's not a real column:
class User < ActiveRecord::Base
include Neoid::Node
neoidable do |c|
c.field :slug
c.field :display_name
c.field :display_name_length do
self.display_name.length
end
end
end
Let's assume that a User
can Like
Movie
s:
# user.rb
class User < ActiveRecord::Base
include Neoid::Node
has_many :likes
has_many :movies, through: :likes
neoidable do |c|
c.field :slug
c.field :display_name
end
end
# movie.rb
class Movie < ActiveRecord::Base
include Neoid::Node
has_many :likes
has_many :users, through: :likes
neoidable do |c|
c.field :slug
c.field :name
end
end
# like.rb
class Like < ActiveRecord::Base
belongs_to :user
belongs_to :movie
end
Now let's make the Like
model a Neoid, by including the Neoid::Relationship
module, and define the relationship (start & end nodes and relationship type) options with neoidable
config and relationship
method:
class Like < ActiveRecord::Base
belongs_to :user
belongs_to :movie
include Neoid::Relationship
neoidable do |c|
c.relationship start_node: :user, end_node: :movie, type: :likes
end
end
Neoid adds the metohds neo_node
and neo_relationships
to instances of nodes and relationships, respectively.
So you could do:
user = User.create!(display_name: "elado")
user.movies << Movie.create("Memento")
user.movies << Movie.create("Inception")
user.neo_node # => #<Neography::Node…>
user.neo_node.display_name # => "elado"
rel = user.likes.first.neo_relationship
rel.start_node # user.neo_node
rel.end_node # user.movies.first.neo_node
rel.rel_type # 'likes'
If you'd like to save nodes manually rather than after_save, use auto_index: false
:
class User < ActiveRecord::Base
include Neoid::Node
neoidable auto_index: false do |c|
end
end
user = User.create!(name: "Elad") # no node is created in Neo4j!
user.neo_save # now there is!
You can query with all Neography's API: traverse
, execute_query
for Cypher, and execute_script
for Gremlin.
Nodes and relationships are auto indexed in the node_auto_index
and relationship_auto_index
indexes, where the key is Neoid::UNIQUE_ID_KEY
(which is 'neoid_unique_id') and the value is a combination of the class name and model id, Movie:43
, this value is accessible with model.neo_unique_id
. So use the constant and this method, never rely on assebling those values on your own because they might change in the future.
That means, you can query like this:
Neoid.db.get_node_auto_index(Neoid::UNIQUE_ID_KEY, user.neo_unique_id)
# => returns a Neography hash
Neoid::Node.from_hash(Neoid.db.get_node_auto_index(Neoid::UNIQUE_ID_KEY, user.neo_unique_id))
# => returns a Neography::Node
If Subreferences are enabled, you can get the subref node and then get all attached nodes:
Neoid.ref_node.outgoing('users_subref').first.outgoing('users').to_a
# => this, according to Neography, returns an array of Neography::Node so no conversion is needed
These examples query Neo4j using Gremlin for IDs of objects, and then fetches them from ActiveRecord with an in
query.
Of course, you can store using the neoidable do |c| c.field ... end
all the data you need in Neo4j and avoid querying ActiveRecord.
Most liked movies
gremlin_query = <<-GREMLIN
m = [:]
g.v(0)
.out('movies_subref').out
.inE('likes')
.inV
.groupCount(m).iterate()
m.sort{-it.value}.collect{it.key.ar_id}
GREMLIN
movie_ids = Neoid.db.execute_script(gremlin_query)
Movie.where(id: movie_ids)
Side note: the resulted movies won't be sorted by like count because the RDBMS won't necessarily do it as we passed a list of IDs. You can sort it yourself with array manipulation, since you have the ids.
Movies of user friends that the user doesn't have
Let's assume we have another Friendship
model which is a relationship with start/end nodes of user
and type of friends
,
user = User.find(1)
gremlin_query = <<-GREMLIN
u = g.idx('node_auto_index').get(unique_id_key, user_unique_id).next()
movies = []
u
.out('likes').aggregate(movies).back(2)
.out('friends').out('likes')
.dedup
.except(movies).collect{it.ar_id}
GREMLIN
movie_ids = Neoid.db.execute_script(gremlin_query, unique_id_key: Neoid::UNIQUE_ID_KEY, user_unique_id: user.neo_unique_id)
Movie.where(id: movie_ids)
Using search
block inside a neoidable
block, you can store certain fields.
# movie.rb
class Movie < ActiveRecord::Base
include Neoid::Node
neoidable do |c|
c.field :slug
c.field :name
c.search do |s|
# full-text index fields
s.fulltext :name
s.fulltext :description
# just index for exact matches
s.index :year
end
end
end
Records will be automatically indexed when inserted or updated.
# will match all movies with full-text match for name/description. returns ActiveRecord instanced
Movie.neo_search("*hello*").results
# same as above but returns hashes with the values that were indexed on Neo4j
Movie.search("*hello*").hits
# search in multiple types
Neoid.neo_search([Movie, User], "hello")
# search with exact matches (pass a hash of field/value)
Movie.neo_search(year: 2013).results
Full text search with Neoid is very limited and is likely not to develop more than this basic functionality. I strongly recommend using gems like Sunspot over Solr.
Neoid has a batch ability, that is good for mass updateing/inserting of nodes/relationships. It sends batched requests to Neography, and takes care of type conversion (neography batch returns hashes and other primitive types) and "after" actions (via promises).
A few examples, easy to complex:
Neoid.batch(batch_size: 100) do
User.all.each(&:neo_save)
end
With then
:
User.first.name # => "Elad"
Neoid.batch(batch_size: 100) do
User.all.each(&:neo_save)
end.then do |results|
# results is an array of the script results from neo4j REST.
results[0].name # => "Elad"
end
Nodes and relationships in the results are automatically converted to Neography::Node and Neography::Relationship, respectively.
With individual then
as well as then
for the entire batch:
Neoid.batch(batch_size: 30) do |batch|
(1..90).each do |i|
(batch << [:create_node, { name: "Hello #{i}" }]).then { |result| puts result.name }
end
end.then do |results|
puts results.collect(&:name)
end
When in a batch, neo_save
adds gremlin scripts to a batch, instead of running them immediately. The batch flushes whenever the batch_size
option is met.
So even if you have 20000 users, Neoid will insert/update in smaller batches. Default batch_size
is 200.
If you have an existing database and just want to integrate Neoid, configure the neoidable
s and run in a rake task or console.
Use batches! It's free, and much faster. Also, you should use includes
to incude the relationship edges on relationship entities, so it doesn't query the DB on each relationship.
Neoid.batch do
[ Like.includes(:user).includes(:movie), OtherRelationshipModel.includes(:from_model).includes(:to_model) ].each { |model| model.all.each(&:neo_save) }
NodeModel.all.each(&:neo_save)
end
This will loop through all of your relationship records and generate the two edge nodes along with a relationship (eager loading for better performance). The second line is for nodes without relationships.
For large data sets use pagination. Better interface for that in the future.
Whenever the neo_node
on nodes or neo_relationship
on relationships is called, Neoid checks if there's a corresponding node/relationship in Neo4j (with the auto indexes). If not, it does the following:
- Ensures there's a sub reference node (read here about sub references), if that option is on.
- Creates a node based on the ActiveRecord, with the
id
attribute and all other attributes fromneoidable
's field list - Creates a relationship between the sub reference node and the newly created node
- Auto indexes a node in the auto index, for fast lookup in the future
Then, when it needs to find it again, it just seeks the auto index with that ActiveRecord id.
Like Nodes, it uses an auto index, to look up a relationship by ActiveRecord id
- With the options passed in the
neoidable
, it fetches thestart_node
andend_node
- Then, it calls
neo_node
on both, in order to create the Neo4j nodes if they're not created yet, and creates the relationship with the type from the options. - Adds the relationship to the relationship index.
In order to test your app or this gem, you need a running Neo4j database, dedicated to tests.
I use port 7574 for testing.
To run another database locally (read here too):
Copy the entire Neo4j database folder to a different location,
or
symlink bin
, lib
, plugins
, system
, copy conf
to a single folder, and create an empty data
folder.
Then, edit conf/neo4j-server.properties
and set the port (org.neo4j.server.webserver.port
) from 7474 to 7574 and run the server with bin/neo4j start
In environments/test.rb
, add:
ENV["NEO4J_URL"] = 'http://localhost:7574'
In your spec_helper.rb
, add the following configurations:
config.before :all do
Neoid.clean_db(:yes_i_am_sure)
end
config.before :each do
Neoid.reset_cached_variables
end
Run the Neo4j DB on port 7574, and run rake
from the gem folder.
Please create a new issue if you run into any bugs. Contribute patches via pull requests. Write tests and make sure all tests pass.
Unfortunately, as for now, Neo4j add-on on Heroku doesn't support Gremlin. Therefore, this gem won't work on Heroku's add on. You should self-host a Neo4j instance on an EC2 or any other server.
Developed by @elado