Working with partially materialized semantic models

Applies to: Kyvos Enterprise  Kyvos Cloud (SaaS on AWS) Kyvos AWS Marketplace

Kyvos Azure Marketplace   Kyvos GCP Marketplace Kyvos Single Node Installation (Kyvos SNI)

Kyvos supports creating and working with partially materialized or HOLAP semantic models. A HOLAP semantic model is a hybrid semantic model that combines features of MOLAP and ROLAP semantic models and allows you to keep some dimensions or measures non-materialized. See  MOLAP, ROLAP, and HOLAP for the distinction between the different semantic model types. 

When you use HOLAP semantic models, and a query asks for a materialized dimension or measure, it is delegated to Kyvos query engines, but when a query asks for a non-materialized dimension or measure, it is delegated to an external SQL engine (PRESTO, Spark, Hive, or Snowflake). To learn more about non-materialized or raw data, see Working with non-materialized or raw data semantic models.

You can filter based on the Materialize property value to see all items that are set to Yes or No.

Consider using partially materialized semantic models when:

• When some dimensions or measures have high cardinality and pre-aggregating the data is unlikely to improve performance

• All underlying data is based on a register file using Table or SQL

• If the source dataset uses transformations, they must be materialized.
  (A semantic model cannot have non-materialized dimensions or measure columns if the source dataset uses any transformations.)

• You are not using MDX-enabled BI tools (such as MicroStrategy)

• You are not using Sentry or Ranger external column-level security tools

There are certain limitations to using a partially materialized semantic model:

• Browsing a HOLAP semantic model will not work with external tools that use MDX connectors (like MicroStrategy), and such semantic models are not listed.

• There may be a data mismatch if the underlying data has changed since the semantic model was last built.

• You cannot browse MDX-based calculated measures created in the semantic model.

• A partition cannot be created on any level of a hierarchy of a dimension that is non-materialized. (This means the sliding window is not supported.)

• You cannot add new materialized columns in between incremental data processes.

• HOLAP-enabled browsing is only supported for a single file, star schemas, snowflake schemas, and multifact schemas.

• Crosstab visualization cannot be viewed in worksheets built on a partially materialized semantic model.

• Due to the floating-point precision issue, the aggregations done by Kyvos may not match exactly with the aggregation done by PRESTO or Hive.

• FirstChild and LastChild semi-additive measures are not supported.

To create a partially materialized semantic model, perform the following steps: 

1. From the Toolbox, click Semantic Models.

2. Click the Actions menu (   ⋮  )  at the top of the Semantic Models column and click Add Semantic Model.

3. Enter a name for the semantic model.

4. Select a Relationship from the list and click Add.  

   1. Use a Kyvos dataset created using Table or SQL as the source. You can enter a term in the search box to quickly find a name.

5. Do one of the following:

   1. Right-click to add dimensions or measures.

   2. Drag fields to the semantic model design worksheet to the areas for measures and dimensions. Use Ctrl+click to select multiple fields. 
      To learn more about each field type, refer to the other topics in this guide.

   3. Click the plus buttons in the Dimension or Measure columns to add additional dimensions or measures, or select candidate measures or dimensions from the candidate lists on the left.

6. In the Properties pane, set Raw Data Querying to Enable.

7. Select the desired dimension or measure you do not want to materialize, and set the Materialize property to No.

8. Setting the value of materialize property for any hierarchy level as No, automatically sets the same for lower levels. 

9. Click Save.

Related topics

• Materialize property

• Working with non-materialized or raw data semantic models