Filtering and Sampling

Honeydew pushes filters down to data, in order to reduce data scans.

Use domain filters to reduce scanned data in ad-hoc exploration

Domain filters limit the data that is accessed by a user query, which improves performance. For example, can add a filter to a domain, that limits data in a fact table to the last week only. Any user query, regardless of the choice of attributes and metrics, will then be limited to the last week’s data. Honeydew engine will push down the filter to reduce data access and improve performance. In some cases of logical data partitioning (for example, having a column indicating a tenant in a multi-tenant environment, or a column indicating data sampling level) a domain source filter can be further applied to improve performance.

Enforce performance filters with conditional filtering

Domain filters can be conditional, applying unless a user overrides them. For example, may by default filter to last week only, unless the user asked for more data. To create -

Create an entity dim_data_cutoff using Custom SQL with possible cutoff values:

 SELECT * FROM VALUES ('week'), ('month'), ('year'),('all_data') AS t(cutoff);

Add a source filter to your domain filtering the data:

type: domain
name: fast

# ...

source_filters:
  # Ensure that data is filtered according to chosen cutoff date
  # By default, filter to last week only ('week')
  - name: dim_data_cutoff_filter
    sql: |
      huge_fact.date >=
          CASE GET_FIELD_SELECTION(dim_data_cutoff.cutoff, 'week')
              WHEN 'week'  THEN CURRENT_DATE - INTERVAL '7' DAY
              WHEN 'month' THEN CURRENT_DATE - INTERVAL '1' MONTH
              WHEN 'year'  THEN CURRENT_DATE - INTERVAL '1' YEAR
              ELSE              DATE '1900-01-01'
          END

When source filters are used, calculations will only see reduced data. If there is a calculation that uses data that is out of the filtering boundaries (i.e revenue_ytd calculating year-to-date revenue) it will return wrong values.

Use domain filters to sample data

Domain source filters can be used to implement data sampling:

In a large fact table to be sampled, create a column with sampling level in the fact table. Filtering on that field will allow to select a data sample - for example, 0.1% of events will have the value “0.1”. Related events are commonly sampled together (i.e. events from the same user/session).
Add the sampling level column to Snowflake clustering settings of the fact table.
Create a “sampled data domain” with a source filter on the sampling level column.

Use parameters to reduced scanned data in dynamic datasets deployed as VIEWs

Dynamic Datasets can be deployed as a Snowflake VIEW. When they are deployed as a VIEW, all metrics and attributes are part of VIEW query, so improvements such as reducing JOINs or pushing down filters are not possible. However, performance of VIEWs can be increased by combining parameters and filters. If a dynamic dataset is defined with a parameterized filter, any access to the view can be controlled with the view parameter. For example, if a view with the following filter orders.o_orderdate = $date_to_explore was defined, then can do:

-- snowflake query
SET date_to_explore='1998-01-01'::DATE;
-- will only process a single chosen date
SELECT * FROM honeydew_view;

Quick Start

Semantic Ecosystem

Modeling Reference

Context Layer

Management

Integrations

Security

Release Notes

Filtering and Sampling