Last Updated: December 13th, 2024

Property Graphs¶

Nodes

• variety of types of entities

• variety of properties per node

Edges

• must be directed edges

• variety of types of relationships

• variety of properties per edge

At the high level, just “two representations”: nodes and edges.

• As opposed to a more restricted, multi-relational schema. In property graphs, node encompasses many entity types, edge encompasses many relation types.

• Can be represented as just two relations in RDBMS

• Easy support of graph expansion/evolution: Add new nodes, new edges, properties to existing nodes/edges

Easily supports graph traversal and path queries:

• PathStart from any node, efficiently find incoming/outgoing edges, etc.

• As opposed to pure JSON, where nested structure is difficult to query/manage

• Flexible ordering/relationships creates more execution strategies

Property Graphs in RDBMSes¶

CREATE TABLE nodes
(node_id INTEGER PRIMARY KEY,
 type STRING,
 properties JSON);

CREATE TABLE edges 
(edge_id INTEGER
   PRIMARY KEY, 
 head_id INTEGER
   REFERENCES Nodes(node_id), 
 tail_id INTEGER
   REFERENCES Nodes(node_id), 
 type STRING,
 properties JSON);

Path Traversal Queries in SQL¶

Suppose we want to find all the US locations in the graph database.

• nodes(node_id, type, properties)

• edges(edge_id, head_id, tail_id, type, properties)

WITH RECURSIVE in_usa (node_id) AS (
  ( SELECT node_id
    FROM nodes
    WHERE properties->>'name' = 'United States'
  ) UNION (
    SELECT tail_id
    FROM edges
    JOIN in_usa ON edges.head_id = in_usa.node_id
    WHERE edges.type = 'within'
  )
)
SELECT * FROM in_usa;

Path traversals of graphs in SQL are achieved using recursive CTEs.

If we wanted rapid path traversal on incoming/outgoing edges, we could build additional indexes:

CREATE INDEX edge_tails 
  ON edges(tail_id); 
CREATE INDEX edge_heads 
  ON edges(head_id);