Data Sources and Bindings (ASSL)
Cubes, dimensions, and other Microsoft SQL Server 2005 Analysis Services (SSAS) objects can be bound to a data source. A data source can be one of the following objects:
- A relational data source.
- A SQL Server 2005 Analysis Services (SSAS) pipeline that outputs a rowset (or chaptered rowsets).
The means of expressing the data source varies by the type of data source. For example, a relational data source is distinguished by the connection string. For more information about data sources, see Data Sources (Analysis Services).
Regardless of the data source used, the data source view (DSV) contains the metadata for the data source. Thus, the bindings for a cube or other Analysis Services objects are expressed as bindings to the DSV. These bindings can include bindings to logical objects—objects such as views, calculated columns, and relationships that do not physically exist in the data source. Using the DSV to bind logical objects is new to SQL Server 2005. For instance, in Microsoft SQL Server 2000, an OLAP measure could specify an expression as its data source. However, in SQL Server 2005, Analysis Services adds a calculated column that encapsulates the expression to the DSV, and then binds the corresponding OLAP measure to that column in the DSV. For more information about DSVs, see Data Source Views (Analysis Services).
Each Analysis Services object binds to the data source in its own way. In addition, the data bindings for these objects and the definition of the data source can be provided inline with the definition of the databound object (for example, the dimension), or out-of-line as a separate set of definitions.
Bindings for Dimensions
Each attribute of a dimension is bound to a column in a DSV. All the attributes of a dimension must come from a single data source. However, the attributes can be bound to columns in different tables. The relationships between the tables are defined in the DSV. In the case where more than one set of relationships exists to the same table, it might be necessary to introduce a named query in the DSV to act as an ‘alias’ table. Expressions and filters are defined in the DSV by using named calculations and named queries.
Bindings for MeasureGroups, Measures, and Partitions
Each measure group has the following default bindings:
- The measure group is bound to a table in a DSV (for example, MeasureGroup.Source).
- Each measure is bound to a column in that table (for example, Measure.ValueColumn.Source).
- Each measure group dimension has a set of granularity attributes that define the granularity of the measure group. Each of these attributes must be bound to the column or columns in the fact table that contain the attribute key. (For more information about granularity attributes, see MeasureGroup Granularity Attributes later in this topic.)
These default bindings can be selectively overridden per partition. Each partition can specify a different data source, table or query name, or filter expression. The most common partitioning strategy is to override the table per partition, by using the same data source. Alternatives include applying a different filter per partition or changing the data source.
The default data source must be defined in the DSV, thereby providing the schema information, including the details of relationships. Any additional tables or queries specified at the partition level do not need to be listed in the DSV, but they must have the same schema as the default table defined for the measure group, or at least they must contain all the columns used by the measures or granularity attributes. The detailed bindings per measure and granularity attribute cannot be overridden at the partition level, and they are assumed to be to the same columns as defined for the measure group. Therefore, if the partition uses a data source that does in fact have a different schema, the TableDefinition query defined for the partition must result in the same schema as the schema used by the measure group.
MeasureGroup Granularity Attributes
When the granularity of a measure group matches the granularity known in the database, and there is a direct relationship from the fact table to the dimension table, the granularity attribute only needs to be bound to the appropriate foreign key column or columns on the fact table. For example, consider the following fact and dimension tables:
Sales(RequestedDate, OrderedProductID, ReplacementProductID, Qty)
Relation: Sales.OrderedProductID -> Product.ProductID
Relation: Sales.ReplacementProductID -> Product.ProductID
If you analyze by the ordered product, for the Ordered Product on Sales dimension role, the Product granularity attribute would be bound to Sales.OrderedProductID.
However, there may be times when the GranularityAttributes might not exist as columns on the fact table. For example, the GranularityAttributes might not exist as columns in the following circumstances:
- The OLAP granularity is coarser than the granularity in the source.
- An intermediate table interposes between the fact table and the dimension table.
- The dimension key is not the same as the primary key in the dimension table.
In all such cases, the DSV must be defined so that the GranularityAttributes exist on the fact table. For example, a named query or calculated column can be introduced.
For example, in the same example tables as above, if the granularity were by Category, then a view of the Sales could be introduced:
SalesWithCategory(RequestedDate, OrderedProductID, ReplacementProductID, Qty, OrderedProductCategory)
SELECT Sales.*, Product.Category AS OrderedProductCategory
FROM Sales INNER JOIN Product
ON Sales.OrderedProductID = Product.ProductID
In this case, the GranularityAttribute Category is bound to SalesWithCategory.OrderedProductCategory.
Migrating from Decision Support Objects
Decision Support Objects (DSO) 8.0 allows PartitionMeasures to be rebound. Therefore, the migration strategy in these cases is to construct the appropriate query.
Similarly, it is not possible to rebind the dimension attributes within a partition, although DSO 8.0 allows this rebinding also. The migration strategy in these cases is to define the necessary named queries in the DSV so that the same tables and columns exist in the DSV for the partition as the tables and columns that are used for the dimension. These cases may require the adoption of the simple migration, in which the From/Join/Filter clause is mapped to a single named query rather than to a structured set of related tables. As DSO 8.0 allows PartitionDimensions to be rebound even when the partition is using the same data source, migration may also require multiple DSVs for the same data source.
In DSO 8.0, the corresponding bindings can be expressed in two different ways, depending on whether optimized schemas are employed, by binding to either the primary key on the dimension table or the foreign key on the fact table. In ASSL, the two different forms are not distinguished.
The same approach to bindings applies even for a partition using a data source that does not contain the dimension tables, because the binding is made to the foreign key column in the fact table, not to the primary key column in the dimension table.
Bindings for Mining Models
A mining model is either relational or OLAP. The data bindings for a relational mining model are considerably different than the bindings for an OLAP mining model.
Bindings for a Relational Mining Model
A relational mining model relies on the relationships defined in the DSV to resolve any ambiguity regarding which columns are bound to which data sources. In a relational mining model, the data bindings follow these rules:
- Each non-nested table column is bound to a column either on the case table or a table related to the case table (following a many-to-one or one-to-one relationship). The DSV defines the relationships between the tables.
- Each nested-table column is bound to a source table. The columns owned by the nested-table column are then bound to columns on that source table or a table related to the source table. (Again, the binding follows a many-to-one or one-to-one relationship.) The mining model bindings do not provide the join path to the nested table. Instead, the relationships defined in the DSV provide this information.
Bindings for an OLAP Mining Model
OLAP mining models do not have the equivalent of a DSV. Therefore, the data bindings must provide any disambiguation between columns and data sources. For example, a mining model can be based on the Sales cube, and columns can be based on Qty, Amount, and Product Name. Alternatively, a mining model can be based on Product, and columns can be based on Product Name, Product Color, and a nested table with Sales Qty.
In an OLAP mining model, the data bindings follow these rules:
- Each non-nested table column is bound to a measure on a cube, to an attribute on a dimension of that cube (specifying the CubeDimension to disambiguate in the case of dimension roles), or to an attribute on a dimension.
- Each nested table column is bound to a CubeDimension. That is, it defines how to navigate from a dimension to a related cube or (in the less common case of nested tables) from a cube to one of its dimensions.
Out-of-line bindings refer to bindings that are included in a command and are not persisted. Out-of-line bindings apply only while that particular command executes. In contrast, inline bindings are contained in the ASSL object definition, and persist with the object definition within server metadata.
ASSL allows out-of-line bindings to be specified on either a Process command, if it is not in a batch, or on a Batch command. If out-of-line bindings are specified on the Batch command, all bindings specified in the Batch command create a new binding context in which all Process commands of the batch run. This new binding context includes objects that are indirectly processed because of the Process command.
When out-of-line bindings are specified on a command, they override the inline bindings contained in the persisted DDL for the specified objects. These processed objects may include the object directly named in the Process command, or they may include other objects whose processing is automatically initiated as a part of the processing.
Out-of-line bindings are specified by including the optional Bindings collection object with the processing command. The optional Bindings collection contains the following elements.
Provides a collection of new bindings.
Replaces DataSource from server that would have been used.
Replaces the DataSourceView from the
server that would have been used.
All elements that relate to out-of-line bindings are optional. For any elements not specified, ASSL uses the specification contained in the DDL of the persisted object. Specification of DataSource or DataSourceView in the Process command is optional. If DataSource or DataSourceView are specified, they are not instantiated and do not persist after the Process command has completed.
Definition of the Out-of-line Binding Type
Inside the out-of-line Bindings collection, ASSL allows a collection of bindings for multiple objects, each a Binding. Each Binding has an extended object reference, which is similar to the object reference, but it can refer to minor objects as well (for example, dimension attributes and measure group attributes). This object takes the flat form typical of the Object element in Process commands, except that the <Object></Object> tags are not present.
Each object for which the binding is specified is identified by an XML element of the form <object>ID (for example, DimensionID). After you have identified the object as specifically as possible with the form <object>ID, then you identify the element for which the binding is being specified, which is usually Source. A common case to note is that in which Source is a property on the DataItem, which is the case for column bindings in an attribute. In this case, you do not specify the DataItem tag; instead, you simply specify the Source property, as if it were directly on the column to be bound.
KeyColumns are identified by their ordering inside the KeyColumns collection. There it is not possible to specify, for example, only the first and third key columns of an attribute, because there is no way to indicate that the second key column is to be skipped. All of the key columns must be present in the out-of-line binding for a dimension attribute.
Translations, although they have no ID, are semantically identified by their language. Therefore, the Translations inside a Binding need to include their language identifier.
One additional element allowed within a Binding that does not exist directly in the DDL is ParentColumnID, which is used for nested tables for data mining. In this case, it is necessary to identify the parent column in the nested table for which the binding is being provided.