Pacer is a JRuby library that enables very expressive graph traversals.
It currently supports 2 major graph database: Neo4j and Dex using the Tinkerpop graphdb stack. Plus there's a very convenient in-memory graph called TinkerGraph which is part of Blueprints.
Pacer allows you to create, modify and traverse graphs using very fast and memory efficient stream processing thanks to the very cool Pipes library. That also means that almost all processing is done in pure Java, so when it comes the usual Ruby expressiveness vs. speed problem, you can have your cake and eat it too, it's very fast!
Pacer is documented with a comprehensive RSpec test suite and with a thorough YARD documentation. Dig in!
If you like, you can also use the documentation locally via
gem install yard yard server
The easiest way to get Pacer is gem install pacer.
If you want to hack on Pacer, you'll need to have
maven installed (I recommend brew install maven), then use rake jar to set up maven's pom.xml file and run the
maven build script.
Note Pacer currently relies on some features that are not yet in the
main Pipes repo. You will need to build the develop branch of my pipes
repo by cloning it and running mvn clean install.
The Tinkerpop suite supports a number of graph data stores. They are all compatible with Pacer, but I have not yet implemented the simple adapters Pacer needs to use them yet. Here is the list of what's supported so far:
- TinkerGraph - In-memory graph db, built in and ready to use without additional dependencies.
- Neo4J - The industry-leading graph db.
gem install pacer-neo4jpangloss/pacer-neo4j - Dex - A very fast, relatively new graph db.
gem install pacer-dexpangloss/pacer-dex
You can run any or all of the above graph databases. Pacer supports running them simultaneuosly and even supports having many of any given type open at once.
Some databases need to be shutdown cleanly when the program exits. You
can shut a database down anytime by calling graph.shutdown, but you
don't need to worry about it much. Pacer uses Ruby's at_exit hook to
automatically shut down all open databases!
Friend recommendation algorithm expressed in basic traversal functions:
friends = person.out_e(:friend).in_v(:type => 'person')
friends.out_e(:friend).in_v(:type => 'person').except(friends).except(person).most_frequent(0...10)
or using Pacer's route extensions to create your own query methods:
person.friends.friends.except(person.friends).except(person).most_frequent(0...10)
or to take it one step further:
person.recommended_friends
To get started, you need to know just a few methods. First, open up a graph (if one doesn't exist it will be automatically created) and add some vertices to it:
dex = Pacer.dex '/tmp/dex_demo'
pangloss = dex.create_vertex :name => 'pangloss', :type => 'user'
okram = dex.create_vertex :name => 'okram', :type => 'user'
group = dex.create_vertex :name => 'Tinkerpop', :type => 'group'
Now, let's see what we've got:
dex.v
produces:
#<V[1024]> #<V[1025]> #<V[1026]>
Total: 3
=> #<GraphV>
There are our vertices. Let's look their properties:
dex.v.properties
{"name"=>"pangloss", "type"=>"user"} {"name"=>"okram", "type"=>"user"}
{"name"=>"Tinkerpop", "type"=>"group"}
Total: 3
=> #<GraphV -> Obj-Map>
Now let's put an edge between them:
dex.create_edge okram, pangloss, :inspired
=> #<E[2048]:1025-inspired-1024>
That's great for creating an edge but what if I've got lots to create? Try this method instead which can add edges to the cross product of all vertices in one route with all vertices in the other:
group.add_edges_to :member, dex.v(:type => 'user')
#<E[4097]:1026-member-1024> #<E[4098]:1026-member-1025>
Total: 2
=> #<Obj 2 ids -> lookup>
There is plenty more to see as well! Please dig into the code and the spec suite to find loads of examples and edge cases. And if you think of a case that I've missed, I'll greatly appreciate your contributions!
I want Pacer and its ecosystem to become a repository for real implementations of ideas, best practices and techniques for streaming data manipulation. I've got lots of ideas that I'd like to add, and although Pacer seems to be quite rock solid right now -- and I am using it in limited production environments -- it is still in flux. If we find a better way to do something, we're going to do it that way even if that means breaking changes from one release to another.
Once Pacer matures further, a decision will be made to 'lock it down' at least a little more, hopefully there will be a community in place by then to help determine the right time for that to happen!
Pacer is meant to be extensible and is built on a very modular architecture. Nearly every chainable route method is actually implemented in an independent module that is plugged into the route only when it's in use. That allows great flexibility in adding methods to routes without clogging up the method namespace. It also makes it natural to make pacer plugin gems.
There are lots of examples of route extensions right inside Pacer. Have a look at the lib/pacer/filter, side_effect and transform folders to see the modules that are built into Pacer. They vary widely in complexity, so take a look around.
If you want to add a traversal technique to Pacer, you can fork Pacer and send me a pull request or just create your own pacer-<feature name> plugin! To see how to build your own Pacer plugin, see my example pacer-bloomfilter plugin which also has a readme file that goes into considerable detail on the process of creating plugins and provides some additional usage examples as well.
As a side note, don't worry about any magic happening behind the scenes to discover or automatically load pacer plugins, there is none of that! If you want to use a pacer plugin, treat it like any other gem, add it to your Gemfile (if that's what you use) and require the gem as normal if you need it.
If you're already familiar with Gremlin, please look at my Introducing Pacer post for a simple introduction and explanation of how Pacer is at once similar to and quite different from Gremlin, the project that inspired it. That post is a little out of date at this point since it refers to the original version of Gremlin. Groovy Gremlin is the latest version, inspired in turn by Pacer!
I'm aiming for 100% test coverage in Pacer and am currently nearly there in the core classes, but there is a way to go with the filter, transform and side effect route modules. Open coverage/index.html to see the current state of test coverage. And of course contributions would be much apreciated.