SQL Server 子查询/CTE 建议

Posted 2023-04-14

【中文标题】SQL Server 子查询/CTE 建议

【英文标题】：SQL Server subquery/CTE advice

【发布时间】：2021-02-22 10:23:52

【问题描述】：

我有一个运行不到 1 秒的 SQL 查询：

SELECT
    GB_BR_NAME AS 'Branch',
    VM_VT_DESC AS 'VType',
    S.VM_SK_VEHICLE AS 'Vehicle',
    VM_SK_SEQUENCE AS 'Vehicle Sequence',
    CASE
        WHEN VM_VT_NU = 'N' THEN VM_SK_CONSTOTAL
        WHEN VM_VT_NU = 'U' THEN VM_SK_ADJSTOCK
    END AS 'Stock Value',
    VM_SK_CONSTOTAL AS 'Cons Total',
    VM_SK_ADJSTOCK AS 'Adj Total',
    DD_MA_DESC AS 'Manufacturer',
    DD_CL_DESCRIPTION AS 'Variant Class',
    VM_VT_NU AS 'New/Used',
    CASE
        WHEN VM_EN_PROGRESS = 'Inv' THEN 'Y'
        WHEN VM_EN_PROGRESS = 'Con' THEN 'Y'
        ELSE 'N'
    END AS 'Sold',
    VM_SK_INTFREEDATE AS 'Interest Free Date',
    CASE
        WHEN VM_SK_INTFREEDATE <= GETDATE() AND DD_MA_DESC NOT IN ('Toyota', 'Skoda', 'Lexus', 'Honda') THEN 1
        ELSE 0
    END AS 'Interest Bearing',
    VM_SK_STATUS AS 'Status'
FROM
    VM_01_STOCK S
    LEFT OUTER JOIN GB_00_BRANC B ON S.VM_SK_BRANCH = B.GB_BR_BRANCH
    LEFT OUTER JOIN VM_01_VTYPE V ON S.VM_SK_VTYPE = V.VM_VT_VTYPE
    LEFT OUTER JOIN VM_00_VEHICLE VE ON S.VM_SK_VEHICLE = VE.VM_VE_VEHICLE
    LEFT OUTER JOIN DD_00_MANUF DDM ON VE.VM_VE_MANUFACT = DDM.DD_MA_MANUFACT
    LEFT OUTER JOIN VM_01_SORDER SO ON S.VM_SK_VEHICLE = SO.VM_SO_VEHICLE AND S.VM_SK_SEQUENCE = SO.VM_SO_SEQUENCE
    LEFT OUTER JOIN DD_00_VARCLASS CL ON VE.VM_VE_VARCLASS = CL.DD_CL_CODE
    LEFT OUTER JOIN VM_01_ENQUIRY E ON S.VM_SK_ENQUIRY = E.VM_EN_ENQUIRY

WHERE
    (VM_SK_STATUS = 'A' OR VM_SK_STATUS = 'C')

这会输出 ~8000 行。

然后我有一个输出约 40 行的 CTE。这也运行不到 1 秒 -

WITH cte (VM_SK_VEHICLE) AS
    (SELECT VM_SK_VEHICLE
    FROM VM_01_STOCK SS LEFT OUTER JOIN VM_01_VTYPE V ON SS.VM_SK_VTYPE = V.VM_VT_VTYPE LEFT OUTER JOIN VM_00_VEHICLE VE ON SS.VM_SK_VEHICLE = VE.VM_VE_VEHICLE LEFT OUTER JOIN DD_00_VARCLASS CL ON VE.VM_VE_VARCLASS = CL.DD_CL_CODE LEFT OUTER JOIN NL_01_TRANS T ON LTRIM(RTRIM(str(VM_SK_VEHICLE))) + LTRIM(RTRIM(str(VM_SK_SEQUENCE))) = NL_TR_SUFFIX LEFT OUTER JOIN PL_01_TRANS PT ON NL_TR_DAYBOOK = PL_TX_DBNUM AND NL_TR_ITEM = PL_TX_DBITEM LEFT OUTER JOIN VM_01_VSPEC VP ON VM_SK_VEHICLE = VM_VS_VEHICLE 
    WHERE (VM_SK_STATUS = 'A' OR VM_SK_STATUS = 'C') AND VM_VT_NU = 'N' AND DD_CL_CODE IN ('CAR', 'LCV', 'HGV') AND VM_VS_DESC = 'Vehicle Price' AND VM_VS_ACTCOST <> 0.00
    GROUP BY VM_SK_VEHICLE, VM_SK_SEQUENCE, VM_SK_CONSTOTAL 
    HAVING SUM(PL_TX_BASEVAL - PL_TX_BASEPAID) < (VM_SK_CONSTOTAL/2))

当我添加到我的主查询时

CASE
    WHEN S.VM_SK_VEHICLE IN (SELECT cte.VM_SK_VEHICLE FROM cte) THEN 1
    ELSE 0
END AS 'Fully Paid',

开始显示结果大约需要 15 分钟。我正在执行 LEFT JOIN 以将 CTE 加入我的主查询。

我还尝试使用子查询进行此操作，大约需要 37 分钟才能开始显示结果。

我需要做什么才能让它以适当的速度运行？

谢谢。

【问题讨论】：

检查您的执行计划并查看花费的时间以及是否缺少索引。然而，有时可以通过转换为将执行计划拆分为 2 的临时表来优化 CTE。

【参考方案1】：

这个谓词似乎出现了问题：

LTRIM(RTRIM(STR(VM_SK_VEHICLE))) + LTRIM(RTRIM(STR(VM_SK_SEQUENCE))) = NL_TR_SUFFIX

这根本不是 sargable。

用这个表达式创建一个计算的持久列：

ALTER TABLE ??? ADD VM_SK_VEHICLE_SEQUENCE
   AS LTRIM(RTRIM(STR(VM_SK_VEHICLE))) + LTRIM(RTRIM(STR(VM_SK_SEQUENCE))) PERSISTED

如果您使用的是最新版本的 SQL Server，请将“LTRIM(RTRIM(”) 替换为“TRIM(”

另外，为连接部分创建一个新索引...

【讨论】：

您好 SQLpro，我已将修剪后的 concat 添加为计算列，并为这个新列和 NL_TR_SUFFIX 编制索引，并将执行时间减少到 ~1min30！这是一个了不起的结果，谢谢！现在“执行计划”中“成本最高”的部分是 Row Count Spool (Lazy Spool)，占 37%。我还可以实施其他优化吗？

要做到这一点，我需要所有表的完整 DDL 和索引的完整列表...当然查询是对的....但是今天我没有时间备用吧！