SQL Server查询性能降低和优化

Question

I've inherited several data sources and I've been tasked with integration and report. Nothing new.

The issue is that while I have a working integrated query (as a view), performance degrades daily (from 8 to 12 seconds).

Here is a slightly simplified version of the table schema and a working query to get the list of tickets.

/*  Table Schema  */
CREATE TABLE [dbo].[KioskImport](
    [ImportID] [bigint] IDENTITY(1,1) NOT NULL,
    [ImportDate] [datetime] NOT NULL,
    [TICKET_NBR] [varchar](25) NULL,
    [TTL_AMT] [decimal](18, 2) NOT NULL,
    [GROUP_ID] [varchar](15) NULL,
    [SALE_DT] [date] NULL,
    [SALE_TMS] [time](0) NULL,
    [BADGE_NO] [varchar](25) NULL
) ON [PRIMARY]

/*  Working Query  */
SELECT 
    [KioskImport].TICKET_NBR
    ,[KioskImport].GROUP_ID
    ,[KioskImport].TTL_AMT
    ,CONVERT(datetime, [KioskImport].SALE_DT) + CONVERT(datetime, [KioskImport].SALE_TMS) AS [SALE_DATE]
    ,[KioskImport].BADGE_NO
FROM [KioskImport]
INNER JOIN 
(
    SELECT MAX(ImportID) MaxID, MAX(ImportDate) MaxDate, TICKET_NBR
    FROM [KioskImport]
    GROUP BY TICKET_NBR
) [MaxRevenueImport]
ON [MaxRevenueImport].MaxID = [KioskImport].ImportID
AND [MaxRevenueImport].MaxDate = [KioskImport].ImportDate

From here I have a view that joins several queries like this to create a product result set.

Does anyone have good suggestions on how to optimize this query so as to mitigate the ongoing performance degradation (buying enough time to redesign...everything).

SQL Version is 2008 R2

Answer 1

As some of the other folks pointed out, you should note what version of SQL you are using. That said, this might perform better as a CROSS APPLY . Try changing your query to:

SELECT 
    [KioskImport].TICKET_NBR
    ,[KioskImport].GROUP_ID
    ,[KioskImport].TTL_AMT
    ,CONVERT(datetime, [KioskImport].SALE_DT) + CONVERT(datetime, [KioskImport].SALE_TMS) AS [SALE_DATE]
    ,[KioskImport].BADGE_NO
FROM [KioskImport] kiOuter
CROSS APPLY (
    SELECT
        kiInner.TICKET_NBR
    FROM [KioskImport] kiInner
    GROUP BY
        kiInner.TICKET_NBR
    HAVING 
        MAX(kiInner.ImportID) = kiOuter.ImportID
        AND MAX(kiInner.ImportDate) = kiOuter.ImportDate
    ) kiInner

Also, if you're seeing query performance change this dramatically over the course of a single day, you likely either are having fragmentation from heavy table churn, are comparing cached vs. uncached query performance, or the server is running other workloads.

Answer 2

I bet 'cross apply' run slower. My question for you is when you have MAX(kiInner.ImportID) what is MAX(kiInner.ImportDate) for that ImportID, or when you get MAX(kiInner.ImportDate) what is MAX(kiInner.ImportID) for that ImportDate. They maybe not in the same row. I suggest to use CTE.

;WITH cte AS (SELECT TICKET_NBR
               , ImportID
               , ImportDate
               , ROW_NUMBER() OVER(PARTITION BY TICKET_NBR              
                    ORDER BY ImportDate DESC, ImportID DESC) AS Rownum
          FROM KioskImport
)
SELECT [KioskImport].TICKET_NBR
    ,[KioskImport].GROUP_ID
    ,[KioskImport].TTL_AMT
    ,CONVERT(datetime, [KioskImport].SALE_DT) + CONVERT(datetime,  [KioskImport].SALE_TMS) AS [SALE_DATE]
    ,[KioskImport].BADGE_NO
FROM [KioskImport] k
INNER JOIN cte cte
     ON cte.ImportID = k.ImportID
     AND cte.ImportDate = k.ImportDate
     AND cte.Rownum = 1;

Answer 3

use this before the select :

SELECT MAX(ImportID) MaxID, MAX(ImportDate) MaxDate, TICKET_NBR
into #MaxRevenueImport
    FROM [KioskImport]
    GROUP BY TICKET_NBR

then use #MaxRevenueImport at your join. If #MaxRevenueImport has more than 10000 rows create index on the temp at TICKET_NBR before the join in the main select. Do not use table variable use temp #.