GROUP BY options in Synapse SQL

Artikkel
02/18/2022

Synapse SQL allows for developing solutions by implementing different GROUP BY options.

What GROUP BY does

The GROUP BY T-SQL clause aggregates data to a summary set of rows.

Serverless SQL pool doesn't support GROUP BY options. Dedicated SQL pool supports a limited number of GROUP BY options.

GROUP BY options supported in dedicated SQL pool

GROUP BY has some options that dedicated SQL pool doesn't support. These options have workarounds, which are as follows:

GROUP BY with ROLLUP
GROUPING SETS
GROUP BY with CUBE

Rollup and grouping sets options

The simplest option here is to use UNION ALL to execute the rollup rather than relying on the explicit syntax. The result is exactly the same

The following example uses the GROUP BY statement with the ROLLUP option:

SELECT [SalesTerritoryCountry]
,      [SalesTerritoryRegion]
,      SUM(SalesAmount)             AS TotalSalesAmount
FROM  dbo.factInternetSales s
JOIN  dbo.DimSalesTerritory t       ON s.SalesTerritoryKey       = t.SalesTerritoryKey
GROUP BY ROLLUP (
                        [SalesTerritoryCountry]
                ,       [SalesTerritoryRegion]
                )
;

By using ROLLUP, the preceding example requests the following aggregations:

Country and Region
Country
Grand Total

To replace ROLLUP and return the same results, you can use UNION ALL and explicitly specify the required aggregations:

SELECT [SalesTerritoryCountry]
,      [SalesTerritoryRegion]
,      SUM(SalesAmount) AS TotalSalesAmount
FROM  dbo.factInternetSales s
JOIN  dbo.DimSalesTerritory t     ON s.SalesTerritoryKey       = t.SalesTerritoryKey
GROUP BY
       [SalesTerritoryCountry]
,      [SalesTerritoryRegion]
UNION ALL
SELECT [SalesTerritoryCountry]
,      NULL
,      SUM(SalesAmount) AS TotalSalesAmount
FROM  dbo.factInternetSales s
JOIN  dbo.DimSalesTerritory t     ON s.SalesTerritoryKey       = t.SalesTerritoryKey
GROUP BY
       [SalesTerritoryCountry]
UNION ALL
SELECT NULL
,      NULL
,      SUM(SalesAmount) AS TotalSalesAmount
FROM  dbo.factInternetSales s
JOIN  dbo.DimSalesTerritory t     ON s.SalesTerritoryKey       = t.SalesTerritoryKey;

To replace GROUPING SETS, the sample principle applies. You only need to create UNION ALL sections for the aggregation levels you want to see.

Cube options

It's possible to create a GROUP BY WITH CUBE using the UNION ALL approach. The problem is that the code can quickly become cumbersome and unwieldy. To mitigate this issue, you can use this more advanced approach.

The first step is to define the 'cube' that defines all the levels of aggregation that we want to create. Take note of the CROSS JOIN of the two derived tables as it generates all levels. The rest of the code is there for formatting.

CREATE TABLE #Cube
WITH
(   DISTRIBUTION = ROUND_ROBIN
,   LOCATION = USER_DB
)
AS
WITH GrpCube AS
(SELECT    CAST(ISNULL(Country,'NULL')+','+ISNULL(Region,'NULL') AS NVARCHAR(50)) as 'Cols'
,          CAST(ISNULL(Country+',','')+ISNULL(Region,'') AS NVARCHAR(50))  as 'GroupBy'
,          ROW_NUMBER() OVER (ORDER BY Country) as 'Seq'
FROM       ( SELECT 'SalesTerritoryCountry' as Country
             UNION ALL
             SELECT NULL
           ) c
CROSS JOIN ( SELECT 'SalesTerritoryRegion' as Region
             UNION ALL
             SELECT NULL
           ) r
)
SELECT Cols
,      CASE WHEN SUBSTRING(GroupBy,LEN(GroupBy),1) = ','
            THEN SUBSTRING(GroupBy,1,LEN(GroupBy)-1)
            ELSE GroupBy
       END AS GroupBy  --Remove Trailing Comma
,Seq
FROM GrpCube;

The following image shows the results of CREATE TABLE AS SELECT:

Group by cube

The second step is to specify a target table for storing interim results:

DECLARE
 @SQL NVARCHAR(4000)
,@Columns NVARCHAR(4000)
,@GroupBy NVARCHAR(4000)
,@i INT = 1
,@nbr INT = 0
;
CREATE TABLE #Results
(
 [SalesTerritoryCountry] NVARCHAR(50)
,[SalesTerritoryRegion]  NVARCHAR(50)
,[TotalSalesAmount]      MONEY
)
WITH
(   DISTRIBUTION = ROUND_ROBIN
,   LOCATION = USER_DB
)
;

The third step is to loop over the cube of columns performing the aggregation. The query will run once for every row in the #Cube temporary table. Results are stored in the #Results temp table:

SET @nbr =(SELECT MAX(Seq) FROM #Cube);

WHILE @i<=@nbr
BEGIN
    SET @Columns = (SELECT Cols    FROM #Cube where seq = @i);
    SET @GroupBy = (SELECT GroupBy FROM #Cube where seq = @i);

    SET @SQL ='INSERT INTO #Results
              SELECT '+@Columns+'
              ,      SUM(SalesAmount) AS TotalSalesAmount
              FROM  dbo.factInternetSales s
              JOIN  dbo.DimSalesTerritory t  
              ON s.SalesTerritoryKey = t.SalesTerritoryKey
              '+CASE WHEN @GroupBy <>''
                     THEN 'GROUP BY '+@GroupBy ELSE '' END

    EXEC sp_executesql @SQL;
    SET @i +=1;
END

Lastly, you can return the results by reading from the #Results temporary table:

SELECT *
FROM #Results
ORDER BY 1,2,3
;

By breaking up the code into sections and generating a looping construct, the code becomes more manageable and maintainable.

Next steps

For more development tips, see development overview.

Jagamisviis: