Table creation script¶
The process of ingesting a file into Databricks is performed by an Azure Data Factory (ADF) pipeline that generates and executes two scripts:
- The table creation script creates the necessary database and tables in the Databricks cluster.
- The transfer query script reads the raw copy of the Asset and inserts the information in the tables created by the previous script.
The table creation is a Spark SQL script specifically generated for each Entity and when it is executed it creates in the Databricks cluster the following structures:
- A database for the Provider
- A table for the Entity
- A validation error table for the Entity
That means that all the Entities associated to the same Provider will share the same database but each Entity will have its own table and validation error table.
The table creation script is generated and executed by the Python Notebook
CreateTables.py , which is stored in the
Shared folder in Databricks.
In previous versions of Sidra, the table creation script was generated by a custom activity named
CreateTableScript which stores it in Databricks File System (DBFS) and then it was executed by a Notebook activity.
The following pseudo-code is a sample of the part of the Spark SQL query that creates the database:
So, it will use the value in the
DatabaseName column of the Provider as the database name.
This is a sample of the part of the Spark SQL query that creates the table:
So, a table is created with the name configured in the
TableName column of the Entity.
That table will have a column for each Attribute associated to the Entity.
The name of the column will be the
Name of the Attribute. The type will be the
HiveType of the Attribute.
As format, the
Format of the Entity will be used. Currently,
Format can be ORC, Parquet or Delta.
Finally, for each Attribute associated to the Entity that has the
IsPartitionColumn to true, a partition will be be generated with the
Name of the Attribute.
Entity validation error table¶
The validation error table is quite similar to the cluster table except for the following differences:
- The name of the table has the suffix validationerrors appended, e.g. if the Entity
TableNameis sample the validation error table name will be samplevalidationerrors.
- The type of the non-calculated columns is
STRINGinstead of the
HiveTypeof the Attribute. Non-calculated columns can be identified for having the
IsCalculatedcolumn configured to
- The validation error table contains an additional column:
ViolatedConstraintsof the Hive type
STRING. This table helps to debug which constraints have been violated during the load of the asset and report the issue.
An Entity sample¶
The relation between the information stored in Core database and the data structures generated in Databricks can be seen in the following simplified sample:
CREATE DATABASE IF NOT EXISTS sales; CREATE TABLE IF NOT EXISTS sales.employee ( 'Name' STRING, 'Age' INT, 'FileDate' DATE) USING PARQUET PARTITIONED BY ('filedate') CREATE TABLE IF NOT EXISTS sales.employeevalidationerrors ( 'Name' STRING, 'Age' STRING, 'ViolatedConstraints' STRING, 'FileDate' DATE) USING PARQUET PARTITIONED BY ('filedate')