Compartilhar via


Aggregate Transformation

The Aggregate transformation applies aggregate functions, such as Average, to column values and copies the results to the transformation output. Besides aggregate functions, the transformation provides the GROUP BY clause, which you can use to specify groups to aggregate across.

You configure the Aggregate transformation at the transformation, output, and column levels.

  • At the transformation level, you configure the Aggregate transformation for performance by specifying the number of keys and distinct count keys the transformation is expected to handle and the percentage by which memory can be extended during the aggregation. The Aggregate transformation can also be configured to generate a warning instead of failing when the value of a divisor is zero.
  • At the output level, you configure the Aggregate transformation for performance by specifying the number of keys the output is expected to contain. The Aggregate transformation supports multiple outputs, and each can be configured differently.
  • At the column level, you specify the aggregation that the column performs and the comparison options of the aggregation. You can also configure the Aggregate transformation for performance by specifying the number of keys and distinct count keys that each column contains, and identifying columns as IsBig if a column contains large numeric values or numeric values with high precision.

The Aggregate transformation is asynchronous, which means that it does not consume and publish data row by row. Instead it consumes the whole rowset, performs its groupings and aggregations, and then publishes the results.

This transformation does not pass through any columns, but creates new columns in the data flow for the data it publishes. Only the input columns to which aggregate functions apply or the input columns the transformation uses for grouping are copied to the transformation output. For example, an Aggregate transformation input might have three columns: CountryRegion, City, and Population. The transformation groups by the CountryRegion column and applies the Sum function to the Population column. Therefore the output does not include the City column.

You can also add multiple outputs to the Aggregate transformation and direct each aggregation to a different output. For example, if the Aggregate transformation applies the Sum and the Average functions, each aggregation can be directed to a different output.

You can apply multiple aggregations to a single input column. For example, if you want the sum and average values for an input column named Sales, you can configure the transformation to apply both the Sum and Average functions to the Sales column.

The Aggregate transformation has one input and one or more outputs. It does not support an error output.

Operations

The Aggregate transformation supports the following operations.

Operation Description

Group by

Divides datasets into groups. Columns of any data type can be used for grouping. For more information, see GROUP BY (Transact-SQL).

Sum

Sums the values in a column. Only columns with numeric data types can be summed. For more information, see SUM (Transact-SQL).

Average

Returns the average of the column values in a column. Only columns with numeric data types can be averaged. For more information, see AVG (Transact-SQL).

Count

Returns the number of items in a group. For more information, see COUNT (Transact-SQL).

Count distinct

Returns the number of unique nonnull values in a group. For more information, see Eliminating Duplicates with DISTINCT.

Minimum

Returns the minimum value in a group. For more information, see MIN (Transact-SQL). In contrast to the Transact-SQL MIN function, this operation can be used only with numeric, date, and time data types.

Maximum

Returns the maximum value in a group. For more information, see MAX (Transact-SQL). In contrast to the Transact-SQL MAX function, this operation can be used only with numeric, date, and time data types.

The Aggregate transformation handles null values in the same way as the SQL Server relational database engine. The behavior is defined in the SQL-92 standard. The following rules apply:

  • In a GROUP BY clause, nulls are treated like other column values. If the grouping column contains more than one null value, the null values are put into a single group.
  • In the COUNT (column name) and COUNT (DISTINCT column name) functions, nulls are ignored and the result excludes rows that contain null values in the named column.
  • In the COUNT (*) function, all rows are counted, including rows with null values.

Handling Big Numbers in Aggregates

A column may contain numeric values that require special consideration because of their large value or precision requirements. The Aggregation transformation includes the IsBig property, which you can set on output columns to invoke special handling of big or high-precision numbers. If a column value may exceed 4 billion or a precision beyond a float data type is required, IsBig should be set to 1.

Setting the IsBig property to 1 affects the output of the aggregation transformation in the following ways:

  • The DT_R8 data type is used instead of the DT_R4 data type.
  • Count results are stored as the DT_UI8 data type.
  • Distinct count results are stored as the DT_UI4 data type.

Note

You cannot set IsBig to 1 on columns that are used in the GROUP BY, Maximum, or Minimum operations.

Performance Considerations

The Aggregate transformation includes a set of properties that you can set to enhance the performance of the transformation.

  • Set the Keys and KeysScale properties of the component and the component outputs. Using Keys, you can specify the exact number of keys the transformation is expected to handle, and using KeysScale, you can specify an approximate number of keys. When you specify a value for Keys, which is the value the transformation will receive when the package runs, the transformation avoids reorganizing cached totals, improving performance.
  • Set the CountDistinctKeys and CountDistinctScale properties of the component. Using CountDistinctKeys, you can specify the exact number of keys the transformation is expected to handle for a count distinct operation. Using CountDistinctScale, you can specify an approximate number of keys for a count distinct operation. When you specify a value for CountDistinctScale, which is the value the transformation will receive when the package runs, the transformation also avoids reorganizing cached totals, improving performance.

Configuring the Aggregate Transformation

You can set properties through SSIS Designer or programmatically.

For more information about the properties that you can set in the Aggregate Transformation Editor dialog box, click one of the following topics:

The Advanced Editor dialog box reflects the properties that can be set programmatically. For more information about the properties that you can set in the Advanced Editor dialog box or programmatically, click one of the following topics:

For more information about how to set properties, click one of the following topics:

See Also

Concepts

Creating Package Data Flow
Integration Services Transformations

Help and Information

Getting SQL Server 2005 Assistance