Skip to content
Brahmand

Graph Data Modelling

Brahmand leverages ClickHouse’s powerful table engines to natively model property-graph data.

In a property graph, both nodes and relationships (edges) can have properties. Consider this example:

Graph

  • Nodes: User, Post, Vote
  • Relationships: Created, HasVote

Brahmand keeps separate tables for nodes and relationships

It uses two data structure to model the relationships.

  1. Edge List
Nodes: U = User, P = Post
Edges: U1 –[created]→ P1
       U1 –[created]→ P3
       U3 –[created]→ P2
|  U  |  P  |
|-----|-----|
|  1  |  1  |
|  1  |  3  |
|  3  |  2  |
  1. Adjacency list
Nodes: U = User, P = Post
Edges: U1 –[created]→ P1
       U1 –[created]→ P3
       U3 –[created]→ P2
|  U  |  P    |
|-----|-------|
|  1  | [1,3] |
|  3  |   2   |

Graph Modeling in Clickhouse

Section titled “Graph Modeling in Clickhouse”
Nodes:      Merge Family tables like MergeTree, ReplacingMergeTree
            Columns: (nodeId, props...)
            e.g (1, 'title', 'body')
Edges:      Merge Family tables like MergeTree, ReplacingMergeTree for edge list
            Columns: (from, to, props...)
            e.g (1, 3, props...)

Work is underway on edge indexes using materialized views and bitmaps to improve both storage efficiency and traversal performance.

Edge-Index: AggregatingMergeTree table for adjacency list.
            One for each direction.
            Columns: (from, to)
            e.g (1, [1,3])
Materialized view: for each direction (outgoing and incoming) populates its
                   corresponding Edge-Index table from the main Edge table,
                   ensuring the index stays up to date.