Cube Definition Skill

This skill helps you create properly structured cube definitions for Drizzle Cube's semantic layer.

Core Concepts

Drizzle Cube is Drizzle ORM-first. All cubes reference Drizzle schema columns directly for compile-time validation and SQL injection protection.

Basic Cube Structure

import { defineCube } from 'drizzle-cube/server'
import { eq } from 'drizzle-orm'
import { employees } from './schema'

export const employeesCube = defineCube({
  name: 'Employees',

  // REQUIRED: Security context filter for multi-tenant isolation
  sql: (securityContext) => eq(employees.organisationId, securityContext.organisationId),

  measures: {
    count: {
      type: 'count',
      sql: () => employees.id
    },
    totalSalary: {
      type: 'sum',
      sql: () => employees.salary
    },
    averageSalary: {
      type: 'avg',
      sql: () => employees.salary
    }
  },

  dimensions: {
    id: {
      type: 'number',
      sql: () => employees.id,
      primaryKey: true
    },
    name: {
      type: 'string',
      sql: () => employees.name
    },
    email: {
      type: 'string',
      sql: () => employees.email
    },
    createdAt: {
      type: 'time',
      sql: () => employees.createdAt
    }
  }
})

CRITICAL: Security Context

Every cube MUST implement security filtering. This is mandatory for multi-tenant data isolation.

// REQUIRED pattern - filter by security context
sql: (securityContext) => eq(table.organisationId, securityContext.organisationId)

// For multiple conditions
sql: (securityContext) => and(
  eq(table.organisationId, securityContext.organisationId),
  eq(table.isDeleted, false)
)

The security context is passed to every query execution:

const result = await semanticLayer.execute(query, {
  organisationId: 'org-123',
  userId: 'user-456'
})

Measure Types

Type	Description	Example
count	Count rows	{ type: 'count', sql: () => table.id }
countDistinct	Count unique values	{ type: 'countDistinct', sql: () => table.userId }
sum	Sum numeric values	{ type: 'sum', sql: () => table.amount }
avg	Average numeric values	{ type: 'avg', sql: () => table.price }
min	Minimum value	{ type: 'min', sql: () => table.date }
max	Maximum value	{ type: 'max', sql: () => table.score }

Filtered Measures

Apply filters within measures:

measures: {
  activeCount: {
    type: 'count',
    sql: () => employees.id,
    filters: [{ sql: () => eq(employees.isActive, true) }]
  },
  highValueOrders: {
    type: 'sum',
    sql: () => orders.amount,
    filters: [{ sql: () => gt(orders.amount, 1000) }]
  }
}

Dimension Types

Type	Description	Example
string	Text values	{ type: 'string', sql: () => table.name }
number	Numeric values	{ type: 'number', sql: () => table.quantity }
boolean	True/false values	{ type: 'boolean', sql: () => table.isActive }
time	Date/timestamp values	{ type: 'time', sql: () => table.createdAt }

Primary Key Dimension

Mark the primary key for proper aggregations:

dimensions: {
  id: {
    type: 'number',
    sql: () => table.id,
    primaryKey: true  // Important for multi-cube queries
  }
}

Cube Joins (Relationships)

Relationship Types

Type	Description	SQL Join
belongsTo	Many-to-one	INNER JOIN
hasOne	One-to-one	LEFT JOIN
hasMany	One-to-many	LEFT JOIN (with CTE)
belongsToMany	Many-to-many	LEFT JOIN (through junction)

belongsTo Example (Many-to-One)

export const employeesCube = defineCube({
  name: 'Employees',
  sql: (ctx) => eq(employees.organisationId, ctx.organisationId),

  joins: {
    Departments: {
      targetCube: () => departmentsCube,
      relationship: 'belongsTo',
      on: [
        { source: employees.departmentId, target: departments.id }
      ]
    }
  },

  measures: { /* ... */ },
  dimensions: { /* ... */ }
})

hasMany Example (One-to-Many)

export const departmentsCube = defineCube({
  name: 'Departments',
  sql: (ctx) => eq(departments.organisationId, ctx.organisationId),

  joins: {
    Employees: {
      targetCube: () => employeesCube,
      relationship: 'hasMany',
      on: [
        { source: departments.id, target: employees.departmentId }
      ]
    }
  },

  measures: { /* ... */ },
  dimensions: { /* ... */ }
})

belongsToMany Example (Many-to-Many)

Use when relating through a junction table:

export const employeesCube = defineCube({
  name: 'Employees',
  sql: (ctx) => eq(employees.organisationId, ctx.organisationId),

  joins: {
    Projects: {
      targetCube: () => projectsCube,
      relationship: 'belongsToMany',
      on: [],  // Not used for belongsToMany
      through: {
        table: employeeProjects,  // Junction table
        sourceKey: [
          { source: employees.id, target: employeeProjects.employeeId }
        ],
        targetKey: [
          { source: employeeProjects.projectId, target: projects.id }
        ],
        // Optional: Security filter for junction table
        securitySql: (securityContext) =>
          eq(employeeProjects.organisationId, securityContext.organisationId)
      }
    }
  }
})

Star Schema Pattern

For fact-dimension-fact joins, the dimension cube MUST define hasMany relationships back to all fact cubes:

// Dimension cube - MUST define hasMany to both facts
export const productsCube = defineCube({
  name: 'Products',
  sql: (ctx) => eq(products.organisationId, ctx.organisationId),

  joins: {
    Sales: {
      targetCube: () => salesCube,
      relationship: 'hasMany',
      on: [{ source: products.id, target: sales.productId }]
    },
    Inventory: {
      targetCube: () => inventoryCube,
      relationship: 'hasMany',
      on: [{ source: products.id, target: inventory.productId }]
    }
  },

  dimensions: {
    name: { type: 'string', sql: () => products.name },
    category: { type: 'string', sql: () => products.category }
  }
})

// Fact cube #1
export const salesCube = defineCube({
  name: 'Sales',
  sql: (ctx) => eq(sales.organisationId, ctx.organisationId),

  joins: {
    Products: {
      targetCube: () => productsCube,
      relationship: 'belongsTo',
      on: [{ source: sales.productId, target: products.id }]
    }
  },

  measures: {
    totalRevenue: { type: 'sum', sql: () => sales.revenue }
  }
})

// Fact cube #2
export const inventoryCube = defineCube({
  name: 'Inventory',
  sql: (ctx) => eq(inventory.organisationId, ctx.organisationId),

  joins: {
    Products: {
      targetCube: () => productsCube,
      relationship: 'belongsTo',
      on: [{ source: inventory.productId, target: products.id }]
    }
  },

  measures: {
    totalStock: { type: 'sum', sql: () => inventory.stockLevel }
  }
})

Registering Cubes

import { SemanticLayerCompiler } from 'drizzle-cube/server'

const semanticLayer = new SemanticLayerCompiler({
  drizzle: db,
  schema: schema
})

// Register cubes
semanticLayer.registerCube(employeesCube)
semanticLayer.registerCube(departmentsCube)

// Execute queries
const result = await semanticLayer.execute({
  measures: ['Employees.count'],
  dimensions: ['Departments.name']
}, securityContext)

Common Patterns

Calculated Dimension

dimensions: {
  fullName: {
    type: 'string',
    sql: () => sql`${employees.firstName} || ' ' || ${employees.lastName}`
  }
}

Date Extraction

dimensions: {
  createdYear: {
    type: 'number',
    sql: () => sql`EXTRACT(YEAR FROM ${orders.createdAt})`
  }
}

Conditional Measure

measures: {
  completedOrders: {
    type: 'count',
    sql: () => orders.id,
    filters: [{ sql: () => eq(orders.status, 'completed') }]
  }
}

Security Best Practices

1. Always include security context filter - Never skip the sql function
2. Filter junction tables - Use securitySql in belongsToMany relationships
3. Use Drizzle query builder - Never construct raw SQL strings
4. Test security isolation - Verify queries return only authorized data

Database Support

Cubes work identically across:

• PostgreSQL
• MySQL
• SQLite

The database type is auto-detected from your Drizzle instance.