Introduction

A separate database plugin to support Neo4j-specific features and configurations.

Use-case

• Users can choose and install Neo4j source and sink plugins.
• Users should see Neo4j logo on plugin configuration page for better experience.
• Users should get relevant information from the tool tip:
  
  • The tool tip should describe accurately what each field is used for.
• Users should not have to specify any redundant configuration.
• Users should get field level lineage for the source and sink that is being used.
• Reference documentation should be updated to account for the changes.
• The source code for Neo4j database plugin should be placed in repo under data-integrations.org.
• The data pipeline using source and sink plugins should run on both mapreduce and spark engines.

User Storie

• User should be able to install Neo4j specific database source and sink plugins from the Hub.
• Users should have each tool tip accurately describe what each field does.
• Users should get field level lineage information for the Neo4j source and sink.
• Users should be able to setup a pipeline avoiding specifying redundant information.
• Users should get updated reference document for Neo4j source and sink.
• Users should be able to read all the DB types.

Plugin Type

• Batch Source
• Batch Sink 
• Real-time Source
• Real-time Sink
• Action
• Post-Run Action
• Aggregate
• Join
• Spark Model
• Spark Compute

Design Tips

• Reference to the Neo4j driver manual: https://neo4j.com/docs/driver-manual/1.7/
• Reference to the Neo4j jdbc driver (it is Community-Contributed Drivers): https://neo4j.com/developer/java-third-party/
• Reference to Cypher Query Language manual: https://neo4j.com/docs/cypher-manual/current/

Design

Neo4j Overview

Neo4j is a graph database management system with native graph storage and processing. In Neo4j, everything is stored in the form of an edge, node, or attribute. Each node and edge can have any number of attributes. Both nodes and edges can be labelled. Labels can be used to narrow searches.

Cypher Query Language

Cypher is a declarative graph query language that allows for expressive and efficient querying and updating of the graph.
Cypher is inspired by a number of different approaches and builds on established practices for expressive querying. Many of the keywords, such as WHERE and ORDER BY, are inspired by SQL. Pattern matching borrows expression approaches from SPARQL. Some of the list semantics are borrowed from languages such as Haskell and Python. 

Here are a few clauses used to read from the graph:

• MATCH: The graph pattern to match. This is the most common way to get data from the graph.
  
• WHERE: Not a clause in its own right, but rather part of MATCH, OPTIONAL MATCH and WITH. Adds constraints to a pattern, or filters the intermediate result passing through WITH.
  
• RETURN: What to return.
  

Here’s an example of simple Cypher Query:

MATCH (n) RETURN n

Example of Neo4j data

Neo4j database contains information about Persons and Movies and relation between them.
For getting information what movies related with person 'Meg Ryan' can be used next CQL query:

MATCH (person:Person {name: "Meg Ryan"})-[rel]-(movie) RETURN person, rel, movie

Result of this query will be next:

"person"

"rel"

"movie"

{"name":"Meg Ryan","born":1961}

{"roles":["DeDe","Angelica Graynamore","Patricia Graynamore"]}

{"title":"Joe Versus the Volcano",
"tagline":"A story of love, lava andburning desire.",
"released":1990}

{"name":"Meg Ryan","born":1961}

{"roles":["Sally Albright"]}

{"title":"When Harry Met Sally",
"tagline":"Can two friends sleep toget her and still love each other in the morning?",
"released":1998}

{"name":"Meg Ryan","born":1961}

{"roles":["Kathleen Kelly"]}

{"title":"You've Got Mail",
"tagline":"At odds in life... in love on-line.",
"released":1998}

{"name":"Meg Ryan","born":1961}

{"roles":["Carole"]}

{"title":"Top Gun",
"tagline":"I feel the need, the need for speed.",
"released":1986}

{"name":"Meg Ryan","born":1961}

{"roles":["Annie Reed"]}

{"title":"Sleepless in Seattle",
"tagline":"What if someone you never met, someone you never saw, someone you never 
knew was the only someone for you?",
"released":1993}

Other case using CQL for getting data from Neo4j:

MATCH (person:Person {name: "Meg Ryan"})-[rel]-(movie) RETURN person.name AS name, rel.roles AS roles, movie.title AS title

Result of this query will be next:

title

Source Splitter

The proposal is to add "Splits Number" Source configuration property, which allows specifying the desired number of splits to divide the query into when reading from Neo4j. 
Fewer splits may be created if the query cannot be divided into the desired number of splits.
Also, we can use '0' as the default value for this configuration property and determine the number of splits according to the number of map tasks (controlled by the "mapreduce.job.maps" property):

public List<InputSplit> getSplits(JobContext job) throws IOException {
 
    ...
 
    int targetNumTasks = job.getConfiguration().getInt(MRJobConfig.NUM_MAPS, 1);
     
    ...

'MATCH ... RETURN COUNT(*)' CQL query can be used in order to get a total number of documents, that will be divided between splits using 'SKIP' and 'LIMIT'
Example:

• Input query
  

  MATCH (person:Person) RETURN person

• Order By

  person.born

In this case each split will be run next query

MATCH (person:Person) RETURN person ORDER BY person.born SKIP x LIMIT y

where 'x' and 'y' determined for each split based on 'Splits Number' and total counts of records.

Source Properties

Source Data Types Mapping

{"name": "n", "type": {
	"type": "record", "name": "n", "fields": [
		{"name": "born", "type": "long"}, 
		{"name": "name", "type": "string"}, 
		{"name": "_id", "type": "long"}, 
		{"name": "_labels", "type": {"type": "array", "items": "string"}}
	]
}}

{"name": "r", "type": {
    "type": "record", "name": "r", "fields": [
        {"name": "_startId", "type": "long"},
        {"name": "roles", "type": {"type": "array", "items": "string"}},
        {"name": "_type", "type": "string"},
        {"name": "_endId", "type": "long"},
        {"name": "_id", "type": "long"}
    ]
}}

{"name": "dr", "type": {
    "type": "record", "name": "dr", "fields": [
        {"name": "duration", "type": "string"},
        {"name": "seconds", "type": "long"},
        {"name": "months", "type": "long"},
        {"name": "days", "type": "long"},
        {"name": "nanoseconds", "type": "int"}
    ]
}}

{"name": "p", "type": {
    "type": "record", "name": "p", "fields": [
        {"name": "crs", "type": "string"},
        {"name": "x", "type": "double"},
        {"name": "y", "type": "double"},
        {"name": "srid", "type": "string"}
    ]
}}

Sink Properties

Output query additionl information

Output query is based on CQL syntax, but using CQL query with CDAP has several problem:

• neo4j-jdbc-driver can process property values only if it primitive types or arrays thereof.
• difficult to relate the output data to CQL query.

To solve these problems, the following solution was proposed:
Using next structure $(...) for identify place where properties will be inserted.
Example of using $(...):
List of output fields: ["name", "age", "profesion", "company", "rating", "position"]

Sink Data Types Mapping

Approach

Create a new maven project in it's own repository.

Pipeline Samples

Please attach one or more sample pipeline(s) and associated data. 

Releases

         Release X.Y.Z

Related Work

Database plugin enhancements