我可以在SQL Server 2016中提高此JSON转换的性能吗？

Question

我有一个类似于下面的表（下面的代码创建一个名为#Temp的表。这有160,000行，这与我在真实数据集中使用的行数大致相同，但真实数据集中有更多列）：

/* Create dummy employees*/

;WITH employeeNumbers
AS ( SELECT 1 AS employeeId
     UNION ALL
     SELECT employeeNumbers.employeeId + 1
     FROM   employeeNumbers
     WHERE  employeeNumbers.employeeId < 16000 )
SELECT *
INTO   #employeeId
FROM   employeeNumbers
OPTION ( MAXRECURSION 16000 )


/*Create saleItems*/
CREATE TABLE #SalesItems
    (
        category VARCHAR(100)
      , subCategory VARCHAR(100)
      , productName VARCHAR(1000)
    )
INSERT INTO #SalesItems ( category
                        , subCategory
                        , productName )
VALUES ( 'Furniture', 'Bookcases', 'Bush Somerset Collection Bookcase' )
     , ( 'Furniture', 'Chairs', 'Hon Deluxe Fabric Upholstered Stacking Chairs, Rounded Back' )
     , ( 'Office Supplies', 'Labels', 'Self-Adhesive Address Labels for Typewriters by Universal' )
     , ( 'Furniture', 'Tables', 'Bretford CR4500 Series Slim Rectangular Table' )
     , ( 'Office Supplies', 'Storage', 'Eldon Fold n Roll Cart System' )
     , ( 'Furniture', 'Furnishings', 'Eldon Expressions Wood and Plastic Desk Accessories, Cherry Wood' )
     , ( 'Office Supplies', 'Art', 'Newell 322' )
     , ( 'Technology', 'Phones', 'Mitel 5320 IP Phone VoIP phone' )
     , ( 'Office Supplies', 'Binders', 'DXL Angle-View Binders with Locking Rings by Samsill' )
     , ( 'Technology', 'Phones', 'Samsung Galaxy S8' )

-- Create some random sales figures between 10 and 100
SELECT employeeId
     , category
     , subCategory
     , productName
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Jul 2017]
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Aug 2017]
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Sep 2017]
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Oct 2017]
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Nov 2017]
     , CONVERT(DECIMAL(13, 2), 10 + ( 100 - 10 ) * RAND(CHECKSUM(NEWID()))) [Dec 2017]
INTO   #Temp
FROM   #employeeId
JOIN   #SalesItems ON 1 = 1

CREATE INDEX empId
    ON #Temp ( employeeId )

SELECT *
FROM   #Temp

我正在做的是将这些结果转换为表中每个员工ID的单个json字符串。我的查询如下：

SELECT DISTINCT x.employeeId
              , (   SELECT y.category
                         , y.subCategory
                         , y.productName
                         , [Jul 2017] AS 'salesAmounts.Jul 2017'
                         , [Aug 2017] AS 'salesAmounts.Aug 2017'
                         , [Sep 2017] AS 'salesAmounts.Sep 2017'
                         , [Oct 2017] AS 'salesAmounts.Oct 2017'
                         , [Nov 2017] AS 'salesAmounts.Nov 2017'
                         , [Dec 2017] AS 'salesAmounts.Dec 2017'
                    FROM   #Temp y
                    WHERE  y.employeeId = x.employeeId
                    FOR JSON PATH, INCLUDE_NULL_VALUES ) data
FROM   #Temp x

哪个有效，但它的表现并不好。在此示例中，执行此操作需要25秒，但在我的真实数据集中需要更长时间。返回#Temp表中的所有结果需要1秒钟。无论如何我可以在这里重新设计我的查询以改善查询时间吗？我确实尝试使用游标迭代每个employeeId并以这种方式生成json字符串，但它仍然很糟糕。

Answer 1

Read "Performance Surprises and Assumptions : GROUP BY vs. DISTINCT" by Aaron Bertrand

尝试使用GROUP BY而不是DISTINCT。在创建结果集之后，DISTINCT会抛出重复项，从而比需要更频繁地调用JSON。 GROUP BY应该首先将设置减少到不同的employeeId值，并且每次只执行一次JSON。

目前无法测试，但这应该做同样的，只是更快：

SELECT x.employeeId
              , (   SELECT y.category
                         , y.subCategory
                         , y.productName
                         , [Jul 2017] AS 'salesAmounts.Jul 2017'
                         , [Aug 2017] AS 'salesAmounts.Aug 2017'
                         , [Sep 2017] AS 'salesAmounts.Sep 2017'
                         , [Oct 2017] AS 'salesAmounts.Oct 2017'
                         , [Nov 2017] AS 'salesAmounts.Nov 2017'
                         , [Dec 2017] AS 'salesAmounts.Dec 2017'
                    FROM   #Temp y
                    WHERE  y.employeeId = x.employeeId
                    FOR JSON PATH, INCLUDE_NULL_VALUES ) data
FROM   #Temp x
GROUP BY x.EmployeeId