SQL：使用冗余联合代码优化嵌套查询？

Question

虽然下面的查询非常适合我的使用，但我试图了解是否有优化的方法，因为它在 UNION 的两个语句中使用相同的嵌套子查询。我的直觉认为应该有一种方法可以查找子查询一次并将该结果用于 UNION 的两个部分，但是在尝试跨 UNION 使用 join 时我遇到了语法问题。

SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
    FROM dbo.ClientLocal
    WHERE clientLocalID = 
        ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
UNION
SELECT ClientMasterName, '0', clientMasterID, 2
    FROM dbo.ClientMaster
    WHERE clientMasterID = 
        ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
ORDER BY HierarchyType ASC;

为了帮助解释为什么需要原始查询，请想象一个“本地”客户端和“主”客户端的概念——本地客户端与主客户端绑定，但主客户端可能有也可能没有本地客户。同样，与每个相关联的电话号码只能与本地客户端或主客户端绑定（因此在第二个语句中，本地客户端 ID 硬编码为“0”——因为没有关联）。

顺便说一下，这将针对 Microsoft SQL Server 2008 R2 运行。

Answer 1

如果您不喜欢重复代码，您可以使用公用表表达式：

WITH cte AS (
  SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789'
)
SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
FROM dbo.ClientLocal
WHERE clientLocalID IN (SELECT HierarchyItem FROM cte)
UNION
SELECT ClientMasterName, '0', clientMasterID, 2
FROM dbo.ClientMaster
WHERE clientMasterID IN (SELECT HierarchyItem FROM cte)
ORDER BY HierarchyType ASC;

Answer 2

因为似乎没有人提到它：代码越少并不一定意味着它的执行速度也会更快。人们似乎常常认为首先执行 CTE（或“按顺序”，以防有多个），然后应用最终查询；他们不是。事实上，服务器几乎会处理 cte，就好像它是 FROM 或 JOIN 子句中某处的派生表一样，因此选择完全相同的查询计划来执行它们。为了避免系统必须扫描ClientPhone 表两次（LIKE '%0123...' 是一个非常“烦人”的操作），您必须预先执行一次，将结果存储在临时表中，然后使用所述临时表- 实际查询中的表。

为了提供一些背景信息，我冒昧地创建了一些示例数据（参见下面的代码）。我的测试结果是：

• 原始查询耗时 71ms
• CTE 版本耗时 71 毫秒，使用完全相同的查询计划
• 临时表的创建需要 26 毫秒，使用它的查询需要 23 毫秒

所以即使服务器需要创建临时表并填充它，组合后者仍然只需要 49 毫秒而不是原来的 71 毫秒！

当然，您的里程可能会根据所涉及的记录数量和重复查询的复杂性而有所不同，就像我说的那样，LIKE '%blah' 是一件烦人的事情，因为它需要对表进行全面扫描（或使用（覆盖）索引的一点运气）。如果是WHERE pk_field = @value，效果可能会大不相同）

查询愉快...

IF DB_ID('test') IS NULL CREATE DATABASE test
GO
USE test
GO

-- setup
IF OBJECT_ID('ClientPhone') IS NOT NULL DROP TABLE ClientPhone 
IF OBJECT_ID('ClientLocal') IS NOT NULL DROP TABLE ClientLocal
IF OBJECT_ID('ClientMaster') IS NOT NULL DROP TABLE ClientMaster
GO
SELECT TOP 50000 
       HierarchyItem = IDENTITY(int, 1, 1),
       PhoneNumber   = Convert(varchar(100), NewID())
  INTO dbo.ClientPhone 
  FROM sys.objects x1, sys.columns x2, sys.objects x3, sys.columns x4

SELECT ClientLocalName = 'client dummy', 
       ClientLocalID   = Convert(int, Rand(HierarchyItem * 37) * 50000), 
       ClientMasterID   = Convert(int, Rand(HierarchyItem * 47) * 50000), 
       HierarchyType   = 1
  INTO dbo.ClientLocal
  FROM dbo.ClientPhone 

SELECT ClientMasterName = 'master dummy', 
       ClientLocalID    = Convert(int, Rand(HierarchyItem * 51) * 50000), 
       ClientMasterID   = Convert(int, Rand(HierarchyItem * 53) * 50000), 
       HierarchyType    = 2
  INTO dbo.ClientMaster
  FROM dbo.ClientPhone 

GO
-- original 
SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
  FROM dbo.ClientLocal
 WHERE ClientLocalID = ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%01')
UNION
SELECT ClientMasterName, '0', ClientMasterID, 2
  FROM dbo.ClientMaster
 WHERE ClientMasterID = ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%01')
ORDER BY HierarchyType ASC;

-- CTE
;WITH cte
   AS (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%01')

SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
  FROM dbo.ClientLocal
 WHERE ClientLocalID = ANY (SELECT HierarchyItem FROM cte)
UNION
SELECT ClientMasterName, '0', ClientMasterID, 2
  FROM dbo.ClientMaster
 WHERE ClientMasterID = ANY (SELECT HierarchyItem FROM cte)
ORDER BY HierarchyType ASC;

-- temp table
IF OBJECT_ID('tempdb..#ClientPhone') IS NOT NULL DROP TABLE #ClientPhone

SELECT HierarchyItem INTO #ClientPhone FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%01'

SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
  FROM dbo.ClientLocal
 WHERE ClientLocalID = ANY (SELECT HierarchyItem FROM #ClientPhone)
UNION
SELECT ClientMasterName, '0', ClientMasterID, 2
  FROM dbo.ClientMaster
 WHERE ClientMasterID = ANY (SELECT HierarchyItem FROM #ClientPhone)
ORDER BY HierarchyType ASC;

Answer 3

;with subQry as (
    SELECT ClientLocalName, ClientLocalID, ClientMasterID
        FROM dbo.ClientLocal
        WHERE clientLocalID = 
            ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
)
SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
    FROM subQry
UNION
SELECT ClientMasterName, '0', clientMasterID, 2
    FROM subQry
ORDER BY HierarchyType ASC;

Answer 4

我能看到的唯一明显优化是使用union all 而不是union：

SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
FROM dbo.ClientLocal
WHERE clientLocalID = ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
UNION ALL
SELECT ClientMasterName, '0', clientMasterID, 2
FROM dbo.ClientMaster
WHERE clientMasterID = ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
ORDER BY HierarchyType ASC;

您显然在子选择之间没有重复项（因为最后一列）。这假设您在每个子选择中也没有重复项。

我不知道 SQL Server 对= ANY 的处理方式，但我认为它与IN 基本相同。 = ANY 子查询将运行一次（对于每个子查询），但第二次将缓存结果。您可以将其简化为：

SELECT ClientLocalName, ClientLocalID, ClientMasterID, 1 HierarchyType
FROM ((SELECT ClientLocalName, ClientLocalID, ClientMasterID
       FROM dbo.ClientLocal
      ) UNION ALL
      (SELECT ClientLocalName, '0', ClientMasterID
       FROM dbo.ClientMasterName
      ) 
     ) c        
WHERE c.clientLocalID = ANY (SELECT HierarchyItem FROM dbo.ClientPhone WHERE PhoneNumber LIKE '%0123456789')
ORDER BY HierarchyType ASC;

我不知道这是否会有很大的改进。

LIKE 模式在数字开头有一个通配符，因此很难优化。