Dimension Groups

[diannewood]

Problem

Dimensions are currently defined as fields within datasets. This creates limitations:

• Same dataset may need different dimension attributes in different models
• Conformed dimensions can't be shared between models
• Conformed dimensions can't span datasets (snowflake dimension)
• No hierarchical structure support

Proposal

Add a new dimension_group as a top-level object

dimension_groups:
  - name: Product
    dataset: Item  # Can be overridden
    dimensions:
      - name: Product Name
        columns: [i_product_name]
        is_unique_key: true
      - name: Product Category
        columns: [i_category]
      - name: Product Color
        columns: [i_color]

Notes

• A single dataset is used as the default. It can be overridden by individual dimensions
• Hierarchies can be optionally defined (see separate proposal)

[siljamardla]

How is this different from having a dimension table for product, with primary key = product_id, with relations to other tables where product_id would be foreign key?

[diannewood]

You're right, it can be defined how you described. The example was used to show a structure that is most similar to the existing OSI spec. But where it falls short is when you have something like a conformed dimension for time or geography that uses multiple tables and needs to be sharable. This structure separates the dataset/table that could be used in different ways in different models, from the conformed dimension definition that needs to be locked down.