Graph Databases in Java

What, Why, How

Greater Milwaukee Java Meetup

7/28/2016 @ Digital Measures

Andrew Glassman

Back End Architect - Digital Measures

@a_glassman

http://aglassman.github.io

Nick Halase

Software Engineer - Digital Measures

@nickhalase

​

What is a Graph Database?

A graph database is a type of data storage that uses vertices and edges to directly relate data.

​

Graph Entities

Vertices

• Singular: Vertex
• Represents a single entity, like a row in a database.
• A Vertex can have many properties of different data types. These property definitions can be defined by a schema, or schema-less depending on the implementation.
• Vertices are referred to as Nodes in some libraries.

​

​

Edges

• Edges represent the relationships between vertices.
• Edges typically have a main “type”.
• A Edge can have many properties of different data types. These property definitions can be defined by a schema, or schema-less depending on the implementation.
• Edges are directional. A bidirectional relationship between two nodes would need to be represented using two edges.

​

​

Example Domains

Example Domain - Taxonomy

• Vertices
• Classification
• Name, description
• Characteristic
• Name, description
• Species
• Name, description
• Edges
• belongs_to
• has

​

​

Example Domain - Social Graph

• Vertices
• Person
• Name
• Age
• Height
• Edges
• Knows
• Since
• reports_to

​

​

Example Domain - Logistics

• Vertices
• Hub
• Name
• Geo-location
• Processing Time
• Truck
• Name
• Hauling Capacity
• Fuel Capacity
• Efficiency

​

​

​

• Edges
• Route
• Name
• Travel Time
• Assigned_To

​

​

Example Domain - Logistics

Hub

Truck

Hub

assigned_to

route

route

Hub

route

route

Truck

Truck

Why use a Graph?

Why use a graph model?

• Highly interconnected data
• Complex queries
• Rapidly changing domain
• The relationships between data elements are as important as the data elements themselves

​

Source

Highly Interconnected Data

• Modeling data that is highly interconnected gets very difficult to manage and query in a RDBMS.
• Typical Example
• Movies - relational:
• Actor, Director, Producer, Movie, Show, Episode, Voice Actor, Writer, Studio
• Maintaining all these tables would get tedious. Now you need tons of PK/FK relations, and join tables. Just imagine the SQL queries this would generate! (so many joins!)
• ​

​

​

Highly Interconnected Data

• Movies - Graph
• Vertices
• Person, Studio
• Content
• type
• Relations
• acted_in (Tarantino directed Pulp Fiction(movie))
• wrote (Tarantino wrote Pulp Fiction(movie))
• financed (Miramiax financed Pulp Fiction(movie))
• contains (Pulp Fiction(movie) contains Misirlou(song))

​

​

Complex Queries

• Graph query languages can express the same query much more efficiently than SQL in some domains.
• RDBMS is based on Relational Algebra. Graph query languages are usually based off of conjunctive Regular Path Queries (CRPQs). This makes them suited to solve different problems.
• Graph is great at traversing relationships. Imagine doing a “degrees of separation” query in SQL.
• MYSQL
• Neo4j
• Great for Join intensive queries. Just find the first node, or nodes, and traverse the graph.

​

​

Complex Queries

• Some standards have been specified, such as RDF. Many graph implementations are built on top of RDF, OWL2, and SPARQL specs. (Blazegraph, Stardog)
• OWL2 is a web ontology language. Ontologies are important when working with large graphs, especially when merging data sets.
• SPARQL is the query language used to query RDF stores.
• RDF stores are implemented using “triple stores”. SPARQL was designed to be able to query distributed resources that implement ontologies.

​

Rapidly Changing Domain / Model

• Graph DBs are almost all schemaless. You can add arbitrary properties, even for the same vertex, or edge types.
• You can add new node types, or edge types at will, and interconnect them without any side effects.

​

​

When NOT to use a Graph?

​

Inappropriate Use-Cases

• Collections of key-value pairings
• Storing aggregate data
• Finding the amount spent across a set of categories in a particular budget

​

​

How to use a Graph DB

Graph is on the rise.

But it’s still dwarfed by RDBMS

Neo4j

• Pros
• Free Community Edition + Enterprise Edition
• Low barrier to entry. Easy Install. Great tutorials.
• Provides Java APIs
• Can run Embedded, or Stand Alone
• Provides REST API for all operations.
• Proprietary Cypher query language.
• Provides full, and incremental backups.
• Cons
• Tinkerpop 2 only, Tinkerpop3 has few contributors.

​

​

Stardog

• Pros
• Supports OWL2, Based on RDF, and supports SPARQL
• Provides Tinkerpop3 / support
• Full text, semantically typed search
• Geospatial queries
• High availability mode using Zookeeper.
• Free community license, flexible developer license
• Cons
• No incremental backups
• Expensive per “node”, but flexible on price

​

​

​

Blazegraph

• Pros
• Based on RDF, supports SPARQL (REST client)
• Supports Tinkerpop3
• Supports running on GPUs for fast queries.
• High Availability configurations supported
• Embedded option as well.
• Cons
• Need commercial license to use. Gray area.
• Embedded license is GPLv2
• Tinkerpop implementation only supports embedded mode.

​

​

DataStax Graph

• Pros
• Supports Tinkerpop3
• Built on Cassandra, so you get a robust, scalable platform out of the box.
• Supports incremental backups, and snapshots.
• Cons
• Not built on RDF (not really a con)
• No community edition.
• Other
• DataStax Blog - Benefits of Gremlin

​

​

OrientDB

• Pros
• Community Edition
• Multi-Model (graph + document)
• Supports Tinkerpop3
• High Availability configurations.
• Cons
• Seems to have remote transaction issues.
• Unstable community that seems driven by “most features” over quality features.

​

​

Apache TinkerPop

• An open graph computing framework
• An abstraction layer over graph databases
• Helps to prevents vendor lock-in
• Maturity
• TinkerPop3 has gained the sponsorship of the Apache Foundation

Apache TinkerPop

• In development for 7 years
• TinkerPop2 final release was September 2014
• Since TinkerPop2, TinkerPop3 has been maintained under the Apache Foundation
• As of 3.2.1 the project is no longer “incubating”
• Released July 18th, 2016 (last week!)

Why TinkerPop?

​

​

​

​

source: tinkerpop.apache.org/docs/current/tutorials/getting-started/#_why_tinkerpop

Why TinkerPop?

• Industry adoption and rockstar developers
• Blazegraph
• DSEGraph
• Hadoop (Giraph)
• Hadoop (Spark)
• IBM Graph
• Neo4j
• Sqlg
• Stardog
• TinkerGraph
• Titan
• Titan (Amazon)
• Titan (Tupl)
• Unipop

source: tinkerpop.apache.org

​