แก้ไข

แชร์ผ่าน


Using column patterns in mapping data flows

APPLIES TO: Azure Data Factory Azure Synapse Analytics

Tip

Try out Data Factory in Microsoft Fabric, an all-in-one analytics solution for enterprises. Microsoft Fabric covers everything from data movement to data science, real-time analytics, business intelligence, and reporting. Learn how to start a new trial for free!

Several mapping data flows transformations allow you to reference template columns based on patterns instead of hard-coded column names. This matching is known as column patterns. You can define patterns to match columns based on name, data type, stream, origin, or position instead of requiring exact field names. There are two scenarios where column patterns are useful:

  • If incoming source fields change often such as the case of changing columns in text files or NoSQL databases. This scenario is known as schema drift.
  • If you wish to do a common operation on a large group of columns. For example, wanting to cast every column that has 'total' in its column name into a double.

Column patterns in derived column and aggregate

To add a column pattern in a derived column, aggregate, or window transformation, click on Add above the column list or the plus icon next to an existing derived column. Choose Add column pattern.

Screenshot shows the plus icon to Add column pattern.

Use the expression builder to enter the match condition. Create a boolean expression that matches columns based on the name, type, stream, origin, and position of the column. The pattern will affect any column, drifted or defined, where the condition returns true.

Screenshot shows the Derived column's settings tab.

The above column pattern matches every column of type double and creates one derived column per match. By stating $$ as the column name field, each matched column is updated with the same name. The value of each column is the existing value rounded to two decimal points.

To verify your matching condition is correct, you can validate the output schema of defined columns in the Inspect tab or get a snapshot of the data in the Data preview tab.

Screenshot shows the Output schema tab.

Hierarchical pattern matching

You can build pattern matching inside of complex hierarchical structures as well. Expand the section Each MoviesStruct that matches where you will be prompted for each hierarchy in your data stream. You can then build matching patterns for properties within that chosen hierarchy.

Screenshot shows hierarchical column pattern.

Flattening structures

When your data has complex structures like arrays, hierarchical structures, and maps, you can use the Flatten transformation to unroll arrays and denormalize your data. For structures and maps, use the derived column transformation with column patterns to form your flattened relational table from the hierarchies. You can use the column patterns that would look like this sample, which flattens the geography hierarchy into a relational table form:

Screenshot shows the Derived column's flatten structure.

Rule-based mapping in select and sink

When mapping columns in source and select transformations, you can add either fixed mapping or rule-based mappings. Match based on the name, type, stream, origin, and position of columns. You can have any combination of fixed and rule-based mappings. By default, all projections with greater than 50 columns will default to a rule-based mapping that matches on every column and outputs the inputted name.

To add a rule-based mapping, click Add mapping and select Rule-based mapping.

Screenshot shows Rule-based mapping selected from Add mapping.

Each rule-based mapping requires two inputs: the condition on which to match by and what to name each mapped column. Both values are inputted via the expression builder. In the left expression box, enter your boolean match condition. In the right expression box, specify what the matched column will be mapped to.

Screenshot shows a mapping.

Use $$ syntax to reference the input name of a matched column. Using the above image as an example, say a user wants to match on all string columns whose names are shorter than six characters. If one incoming column was named test, the expression $$ + '_short' will rename the column test_short. If that's the only mapping that exists, all columns that don't meet the condition will be dropped from the outputted data.

Patterns match both drifted and defined columns. To see which defined columns are mapped by a rule, click the eyeglasses icon next to the rule. Verify your output using data preview.

Regex mapping

If you click the downward chevron icon, you can specify a regex-mapping condition. A regex-mapping condition matches all column names that match the specified regex condition. This can be used in combination with standard rule-based mappings.

Screenshot shows the regex-mapping condition with Hierarchy level and Name matches.

The above example matches on regex pattern (r) or any column name that contains a lower case r. Similar to standard rule-based mapping, all matched columns are altered by the condition on the right using $$ syntax.

Rule-based hierarchies

If your defined projection has a hierarchy, you can use rule-based mapping to map the hierarchies subcolumns. Specify a matching condition and the complex column whose subcolumns you wish to map. Every matched subcolumn will be outputted using the 'Name as' rule specified on the right.

Screenshot shows a rule-based mapping using for a hierarchy.

The above example matches on all subcolumns of complex column a. a contains two subcolumns b and c. The output schema will include two columns b and c as the 'Name as' condition is $$.

Pattern matching expression values

  • $$ translates to the name or value of each match at run time. Think of $$ as equivalent to this
  • $0 translates to the current column name match at run time for scalar types. For hierarchical types, $0 represents the current matched column hierarchy path.
  • name represents the name of each incoming column
  • type represents the data type of each incoming column. The list of data types in the data flows type system can be found here.
  • stream represents the name associated with each stream, or transformation in your flow
  • position is the ordinal position of columns in your data flow
  • origin is the transformation where a column originated or was last updated