Concepts

Cube borrows a lot of terminology from OLAP theory (opens in a new tab), and this document is intended for both newcomers and regular users to refresh their understanding.

We'll use a sample e-commerce database with two tables, orders and line_items to illustrate the concepts throughout this page:

orders

line_items

Cubes

Cubes represent datasets in Cube and are conceptually similar to views in SQL (opens in a new tab). Cubes are usually declared in separate files with one cube per file. Typically, a cube points to a single table in your database using the sql_table property:

cube(`orders`, {
  sql_table: `orders`,
});

You can also use the sql property to accommodate more complex SQL queries:

cube(`orders`, {
  sql: `
    SELECT *
    FROM orders, line_items
    WHERE orders.id = line_items.order_id
  `
})

Each cube contains the definitions of its members: dimensions, measures, and segments. You can control the access to cubes and their members by configuring the member-level security.

Joins are used to define relations between cubes. Pre-aggregations are used to accelerate queries to cubes. Cubes and their members can be further referenced by views.

Note that cubes support extension, polymorphism, and data blending. Cubes can be defined statically and you can also build dynamic data models.

Views

Views sit on top of the data graph of cubes and create a facade of your whole data model with which data consumers can interact. They are useful for defining metrics, managing governance and data access, and controlling ambiguous join paths.

Views do not define their own members. Instead, they reference cubes by specific join paths and include their members. Optionally, you can also group members of a view into folders.

In the example below, we create the orders view which includes select members from base_orders, products, and users cubes:

view(`orders`, {
  cubes: [
    {
      join_path: base_orders,
      includes: [
        `status`,
        `created_date`,
        `total_amount`,
        `total_amount_shipped`,
        `count`,
        `average_order_value`,
      ],
    },
    {
      join_path: base_orders.line_items.products,
      includes: [
        {
          name: `name`,
          alias: `product`,
        },
      ],
    },
    {
      join_path: base_orders.users,
      prefix: true,
      includes: `*`,
      excludes: [`company`],
    },
  ],
});

Views do not define any pre-aggregations. Instead, they reuse pre-aggregations from underlying cubes.

View can be defined statically and you can also build dynamic data models.

Dimensions

Dimensions represent the properties of a single data point in the cube. The orders table contains only dimensions, so representing them in the orders cube is straightforward:

cube(`orders`, {
  // ...
 
  dimensions: {
    id: {
      sql: `id`,
      type: `number`,
      // Here we explicitly let Cube know this field is the primary key
      // This is required for de-duplicating results when using joins
      primary_key: true,
    },
 
    status: {
      sql: `status`,
      type: `string`,
    },
  },
});

Note that the id dimension is defined as a primary key. It is also possible to have more than one primary key dimension in a cube if you'd like them all to be parts of a composite key.

The line_items table also has a couple of dimensions which can be represented as follows:

cube(`line_items`, {
  // ...
 
  dimensions: {
    id: {
      sql: `id`,
      type: `number`,
      // Again, we explicitly let Cube know this field is the primary key
      // This is required for de-duplicating results when using joins
      primary_key: true,
    },
 
    order_id: {
      sql: `order_id`,
      type: `number`,
    },
  },
});

If needed, dimensions can be organized into hierarchies. Also, proxy dimensions are helpful for code reusability and subquery dimensions can be used to join cubes implicitly.

Dimension types

Dimensions can be of different types, e.g., string, number, or time. Often, data types in SQL are mapped to dimension types in the following way:

Time dimensions

Time-based properties are modeled using dimensions of the time type. They allow grouping the result set by a unit of time (e.g., days, weeks, month, etc.), also known as the time dimension granularity.

The following granularities are available by default for any time dimension: year, quarter, month, week (starting on Monday), day, hour, minute, second. You can also define custom granularities and optionally expose them via proxy dimensions in case you need to use weeks starting on Sunday, fiscal years, etc.

Here's how we can add time dimensions to the data model:

cube(`orders`, {
  // ...
 
  dimensions: {
    created_at: {
      sql: `created_at`,
      type: `time`
      // You can use this time dimension with all default granularities:
      // year, quarter, month, week, day, hour, minute, second
    },
 
    completed_at: {
      sql: `completed_at`,
      type: `time`,
      // You can use this time dimension with all default granularities
      // and an additional custom granularity defined below
      granularities: {
        fiscal_year_starting_on_february_01: {
          interval: `1 year`,
          offset: `1 month`
        }
      }
    }
  }
})

Time dimensions are essential to enabling performance boosts such as partitioned pre-aggregations and incremental refreshes.

Measures

Measures represent the properties of a set of data points in the cube. To add a measure called count to our orders cube, for example, we can do the following:

cube(`orders`, {
  // ...
 
  measures: {
    count: {
      type: `count`,
    },
  },
});

In our LineItems cube, we can also create a measure to sum up the total value of line items sold:

cube(`line_items`, {
  // ...
 
  measures: {
    total: {
      sql: `price`,
      type: `sum`,
    },
  },
});

Calculated measures and subquery dimensions can be used for measure composition. Multi-stage calculations enable data modeling of more sophisticated measures.

Measure types

Measures can be of different types, e.g., count, sum, or number. Often, aggregate functions in SQL are mapped to measure types in the following way:

Measure additivity

Additivity is a property of measures that detemines whether measure values, once calculated for a set of dimensions, can be further aggregated to calculate measure values for a subset of these dimensions.

Measure additivity has an impact on pre-aggregation matching.

Additivity of a measure depends on its type. Only measures with the following types are considered additive: count, count_distinct_approx, min, max, sum. Measures with all other types are considered non-additive.

Example

Consider the following cube:

cubes:
  - name: employees
    sql: >
      SELECT 1 AS id, 'Ali' AS first_name, 20 AS age, 'Los Gatos' AS city UNION ALL
      SELECT 2 AS id, 'Bob' AS first_name, 30 AS age, 'San Diego' AS city UNION ALL
      SELECT 3 AS id, 'Eve' AS first_name, 40 AS age, 'San Diego' AS city
 
    measures:
      - name: count
        type: count
 
      - name: avg_age
        sql: age
        type: avg
 
    dimensions:
      - name: city
        sql: city
        type: string

If we run a query that includes city as a dimension and count and avg_age as measures, we'll get the following results:

Then, if we remove the city dimension from the query, we'll get the following results:

As you can see, the value of the count measure that we've got for the second query could have been calculated based on the results of the first one: 1 + 2 = 3. It explains why the count measure, having the count type, is considered additive.

However, the value of the avg_age measure that we've got for the second query can't be calculated based on the results of the first one: there's no way to derive 30 from 20 and 35. This is why the avg_age measure, having the avg type, is considered non-additive.

Leaf measures

Measures that do not reference other measures are considered leaf measures.

By definition, all measures that only reference SQL columns and expressions are leaf measures. On the other hand, calculated measures might not necessarily be leaf measures because they can reference other measures.

Whether a query contains only additive leaf measures has an impact on pre-aggregation matching.

Joins

Joins define the relationships between cubes, which then allows accessing and comparing properties from two or more cubes at the same time. In Cube, all joins are LEFT JOINs.

In the following example, we are left-joining the line_items cube onto our orders cube:

cube(`orders`, {
  // ...
 
  joins: {
    line_items: {
      relationship: `many_to_one`,
      // Here we use the `CUBE` global to refer to the current cube,
      // so the following is equivalent to `orders.id = line_items.order_id`
      sql: `${CUBE}.id = ${line_items.order_id}`,
    },
  },
});

There are three types of join relationships (one_to_one, one_to_many, and many_to_one) and a few other concepts such as the direction of joins and trasitive joins pitfalls.

Segments

Segments are pre-defined filters that are kept within the data model instead of a Cube query. They help to simplify queries and make it easy to reuse common filters across a variety of queries.

To add a segment which limits results to completed orders, we can do the following:

cube(`orders`, {
  // ...
 
  segments: {
    only_completed: {
      sql: `${CUBE}.status = 'completed'`,
    },
  },
});

Pre-aggregations

Pre-aggregations provide a powerful way to accelerate frequently used queries and keep the cache up-to-date. Within a data model, they are defined using the pre_aggregations property:

cube(`orders`, {
  // ...
 
  pre_aggregations: {
    main: {
      measures: [CUBE.count],
      dimensions: [CUBE.status],
      timeDimension: CUBE.created_at,
      granularity: `day`,
    },
  },
});

A more thorough introduction can be found in Getting Started with Pre-Aggregations.