Using IDENTITY to create surrogate keys using dedicated SQL pool in Azure Synapse Analytics

In this article, you'll find recommendations and examples for using the IDENTITY property to create surrogate keys on tables in dedicated SQL pool.

What is a surrogate key

A surrogate key on a table is a column with a unique identifier for each row. The key is not generated from the table data. Data modelers like to create surrogate keys on their tables when they design data warehouse models. You can use the IDENTITY property to achieve this goal simply and effectively without affecting load performance.

Note

In Azure Synapse Analytics:

  • The IDENTITY value increases on its own in each distribution and does not overlap with IDENTITY values in other distributions. The IDENTITY value in Synapse is not guaranteed to be unique if the user explicitly inserts a duplicate value with “SET IDENTITY_INSERT ON” or reseeds IDENTITY. For details, see CREATE TABLE (Transact-SQL) IDENTITY (Property).
  • UPDATE on distribution column does not guarantee IDENTITY value to be unique. Use DBCC CHECKIDENT (Transact-SQL) after UPDATE on distribution column to verify uniqueness.

Creating a table with an IDENTITY column

The IDENTITY property is designed to scale out across all the distributions in the dedicated SQL pool without affecting load performance. Therefore, the implementation of IDENTITY is oriented toward achieving these goals.

You can define a table as having the IDENTITY property when you first create the table by using syntax that is similar to the following statement:

CREATE TABLE dbo.T1
(    C1 INT IDENTITY(1,1) NOT NULL
,    C2 INT NULL
)
WITH
(   DISTRIBUTION = HASH(C2)
,   CLUSTERED COLUMNSTORE INDEX
)
;

You can then use INSERT..SELECT to populate the table.

This remainder of this section highlights the nuances of the implementation to help you understand them more fully.

Allocation of values

The IDENTITY property doesn't guarantee the order in which the surrogate values are allocated due to the distributed architecture of the data warehouse. The IDENTITY property is designed to scale out across all the distributions in the dedicated SQL pool without affecting load performance.

The following example is an illustration:

CREATE TABLE dbo.T1
(    C1 INT IDENTITY(1,1)    NOT NULL
,    C2 VARCHAR(30)                NULL
)
WITH
(   DISTRIBUTION = HASH(C2)
,   CLUSTERED COLUMNSTORE INDEX
)
;

INSERT INTO dbo.T1
VALUES (NULL);

INSERT INTO dbo.T1
VALUES (NULL);

SELECT *
FROM dbo.T1;

DBCC PDW_SHOWSPACEUSED('dbo.T1');

In the preceding example, two rows landed in distribution 1. The first row has the surrogate value of 1 in column C1, and the second row has the surrogate value of 61. Both of these values were generated by the IDENTITY property. However, the allocation of the values is not contiguous. This behavior is by design.

Skewed data

The range of values for the data type are spread evenly across the distributions. If a distributed table suffers from skewed data, then the range of values available to the datatype can be exhausted prematurely. For example, if all the data ends up in a single distribution, then effectively the table has access to only one-sixtieth of the values of the data type. For this reason, the IDENTITY property is limited to INT and BIGINT data types only.

SELECT..INTO

When an existing IDENTITY column is selected into a new table, the new column inherits the IDENTITY property, unless one of the following conditions is true:

  • The SELECT statement contains a join.
  • Multiple SELECT statements are joined by using UNION.
  • The IDENTITY column is listed more than one time in the SELECT list.
  • The IDENTITY column is part of an expression.

If any one of these conditions is true, the column is created NOT NULL instead of inheriting the IDENTITY property.

CREATE TABLE AS SELECT

CREATE TABLE AS SELECT (CTAS) follows the same SQL Server behavior that's documented for SELECT..INTO. However, you can't specify an IDENTITY property in the column definition of the CREATE TABLE part of the statement. You also can't use the IDENTITY function in the SELECT part of the CTAS. To populate a table, you need to use CREATE TABLE to define the table followed by INSERT..SELECT to populate it.

Explicitly inserting values into an IDENTITY column

Dedicated SQL pool supports SET IDENTITY_INSERT <your table> ON|OFF syntax. You can use this syntax to explicitly insert values into the IDENTITY column.

Many data modelers like to use predefined negative values for certain rows in their dimensions. An example is the -1 or "unknown member" row.

The next script shows how to explicitly add this row by using SET IDENTITY_INSERT:

SET IDENTITY_INSERT dbo.T1 ON;

INSERT INTO dbo.T1
(   C1
,   C2
)
VALUES (-1,'UNKNOWN')
;

SET IDENTITY_INSERT dbo.T1 OFF;

SELECT     *
FROM    dbo.T1
;

Loading data

The presence of the IDENTITY property has some implications to your data-loading code. This section highlights some basic patterns for loading data into tables by using IDENTITY.

To load data into a table and generate a surrogate key by using IDENTITY, create the table and then use INSERT..SELECT or INSERT..VALUES to perform the load.

The following example highlights the basic pattern:

--CREATE TABLE with IDENTITY
CREATE TABLE dbo.T1
(    C1 INT IDENTITY(1,1)
,    C2 VARCHAR(30)
)
WITH
(   DISTRIBUTION = HASH(C2)
,   CLUSTERED COLUMNSTORE INDEX
)
;

--Use INSERT..SELECT to populate the table from an external table
INSERT INTO dbo.T1
(C2)
SELECT     C2
FROM    ext.T1
;

SELECT *
FROM   dbo.T1
;

DBCC PDW_SHOWSPACEUSED('dbo.T1');

Note

It's not possible to use CREATE TABLE AS SELECT currently when loading data into a table with an IDENTITY column.

For more information on loading data, see Designing Extract, Load, and Transform (ELT) for dedicated SQL pool and Loading best practices.

System views

You can use the sys.identity_columns catalog view to identify a column that has the IDENTITY property.

To help you better understand the database schema, this example shows how to integrate sys.identity_column` with other system catalog views:

SELECT  sm.name
,       tb.name
,       co.name
,       CASE WHEN ic.column_id IS NOT NULL
             THEN 1
        ELSE 0
        END AS is_identity
FROM        sys.schemas AS sm
JOIN        sys.tables  AS tb           ON  sm.schema_id = tb.schema_id
JOIN        sys.columns AS co           ON  tb.object_id = co.object_id
LEFT JOIN   sys.identity_columns AS ic  ON  co.object_id = ic.object_id
                                        AND co.column_id = ic.column_id
WHERE   sm.name = 'dbo'
AND     tb.name = 'T1'
;

Limitations

The IDENTITY property can't be used:

  • When the column data type is not INT or BIGINT
  • When the column is also the distribution key
  • When the table is an external table

The following related functions are not supported in dedicated SQL pool:

Common tasks

This section provides some sample code you can use to perform common tasks when you work with IDENTITY columns.

Column C1 is the IDENTITY in all the following tasks.

Find the highest allocated value for a table

Use the MAX() function to determine the highest value allocated for a distributed table:

SELECT MAX(C1)
FROM dbo.T1

Find the seed and increment for the IDENTITY property

You can use the catalog views to discover the identity increment and seed configuration values for a table by using the following query:

SELECT  sm.name
,       tb.name
,       co.name
,       ic.seed_value
,       ic.increment_value
FROM        sys.schemas AS sm
JOIN        sys.tables  AS tb           ON  sm.schema_id = tb.schema_id
JOIN        sys.columns AS co           ON  tb.object_id = co.object_id
JOIN        sys.identity_columns AS ic  ON  co.object_id = ic.object_id
                                        AND co.column_id = ic.column_id
WHERE   sm.name = 'dbo'
AND     tb.name = 'T1'
;

Next steps