How Brahmand Models and Stores Graph Data

Brahmand translates high-level Cypher DDL into ClickHouse tables and materialized views, optimizing both storage and traversal performance.

Example

• Nodes: User, Post
• Relationship: Created

Cypher DDL

CREATE NODE TABLE User(
    UserId UInt64,
    Reputation String,
    <!-- other columns -->
    NODE ID(UserId),
    PRIMARY KEY(Reputation, UserId)
);

CREATE NODE TABLE Post(
    PostID UInt64,
    PostTypeId Int8,
    <!-- other columns -->
    NODE id(PostID),
    PRIMARY KEY(PostTypeId, PostID)
);

CREATE REL TABLE CREATED(From User To Post);

ClickHouse Translation

-- node User
CREATE TABLE User(
    userId UInt64,
    Reputation String,
    ...
) ENGINE = MergeTree
ORDER BY(Reputation, userId);

-- node Post
CREATE TABLE Post(
    postId UInt64,
    PostTypeId Int8,
    ...
) ENGINE = MergeTree
ORDER BY(PostTypeId, postId);

-- edge Created
CREATE TABLE CREAETED(
    from_user UInt64,
    to_post UInt64,
) ENGINE = MergeTree
ORDER BY(from_user, to_post);



-- edge index - Created outgoing direction
CREATE TABLE CREAETED_outgoing(
    from_id UInt64,
    to_id AggregateFunction(groupBitmap, UInt64)
) ENGINE = AggregatingMergeTree
ORDER BY from_id;

-- MV for outgoing edge
CREATE MATERIALIZED VIEW mv_CREAETED_outgoing TO CREAETED_outgoing AS
SELECT
    from_user AS from_id,
    groupBitmapState(to_post) AS to_id
FROM CREAETED
GROUP BY from_id;



-- edge index - Created incoming direction
CREATE TABLE CREAETED_incoming(
    from_id UInt64,
    to_id AggregateFunction(groupBitmap, UInt64)
) ENGINE = AggregatingMergeTree
ORDER BY from_id;


-- MV for incoming edge
CREATE MATERIALIZED VIEW mv_CREAETED_incoming TO CREAETED_incoming AS
SELECT
    to_post AS from_id,
    groupBitmapState(from_user) AS to_id
FROM CREAETED
GROUP BY from_id;

Bi-directional Indexes

By maintaining both outgoing and incoming edge indexes, BrahmandDB supports efficient bi-directional traversals without additional query-time joins.

Bitmap-Based Edge Index Storage

Edges are stored using ClickHouse’s Roaring Bitmap (via AggregateFunction(groupBitmap, UInt64)), offering:

• High compression of large integer sets

• Fast set operations for union/intersection during traversals

This combination of MergeTree tables, AggregatingMergeTree edge indexes, and materialized views delivers both storage efficiency and traversal speed at analytical scale.