Data Catalog methods to select data from a data source

In Collibra, the following Data Catalog features require data from a data source:

• Data sampling: Collibra needs data to show examples on an asset page.
• Data classification: Collibra needs data to verify it against available data classes and suggest the best classification for the asset.
• Data profiling: Collibra needs data to calculate descriptive statistics, which are shown on an asset page.

Data Catalog can use the following methods to select data from a data source:

• Source-driven random sampling: This method is where the data source itself returns a set of random rows, improving performance for features that need data.
  Note 

  This selection method is also referred to as Push down sampling (Jobserver), partial scan (Edge), or Random Rows (Edge).

   

  Show the data sources for which source-driven random sampling is available

  Amazon Athena (only on Edge)

  Amazon Redshift

  Apache Hive

  Apache Impala

  Apache Spark SQL

  Cloudera Hive

  Databricks

  Exasol

  Google BigQuery, except if you have unnested the array columns via connection property Other=UnnestArrays=Data.

  Oracle (Thin client-side)

  PostgreSQL

  Snowflake

  SQL Server

  Teradata

  
  

  
  

• Platform-driven random sampling: This method is where all rows are selected from the data source and only a random set of them are used by the Edge capability.
• All rows: This method is where all rows are selected in the data source and all of them are used by the Edge capability.

Below, you find which selection methods are used by the Data Catalog features.

Selection method	Source-driven random sampling	Platform-driven random sampling	All rows
Data Sampling	Automatically used if available for the data.	Used if source-driven random sampling isn't available for the data source.	Not used
Data Classification	Automatically used if available for the data.	Used if source-driven random sampling isn't available for the data source.	Not used
Data Profiling	Used if available for the data source and if the Random rows option is selected when configuring the profiling options.	Not used	Used if source-driven random sampling isn't available for the data source.