我有以下查询来过滤给定城市和横幅中的活跃商店
SELECT ShopName,CityName
FROM Shop S INNER JOIN City C ON C.CityId=S.CityId
WHERE S.Active=1 AND S.Banner like '%res%'
SELECT ShopName,CityName
FROM Shop S INNER JOIN City C ON C.CityId=S.CityId AND S.Active=1
WHERE S.Banner like '%res%'
如果您查看两个查询，它们会给出相同的结果，但在不同的地方使用 S.Active。一个是 IN JOIN 另一个是 WHERE 子句。
哪个查询会执行得更好？
我在 City 表中有 3000000 个城市，在 Shops 表中有 40000000 个商店。我为 (City,Active) 创建了非聚集索引
您应该阅读 Conor Cunningham 撰写的这篇博文：如何编写非 JOIN WHERE 子句。
如果连接类型是INNER[short] answer usually no。如果您在 AdventureWorks 数据库 (SQL 2008) 上运行这两个查询：
SET STATISTICS PROFILE ON
GO
SELECT h.SalesOrderID, h.OrderDate, d.LineTotal
FROM Sales.SalesOrderDetail d
INNER JOIN Sales.SalesOrderHeader h ON d.SalesOrderID = h.SalesOrderID
WHERE h.OrderDate = '20010707'
AND d.LineTotal < 100
SELECT h.SalesOrderID, h.OrderDate, d.LineTotal
FROM Sales.SalesOrderDetail d
INNER JOIN Sales.SalesOrderHeader h ON h.OrderDate = '20010707' AND d.SalesOrderID = h.SalesOrderID
WHERE d.LineTotal < 100
GO
SET STATISTICS PROFILE OFF
GO
然后你会得到这些执行计划：
Rows Executes StmtText StmtId NodeId Parent PhysicalOp
-------------------- -------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------- ----------- ----------- --------------------
5 1 SELECT h.SalesOrderID, h.OrderDate, d.LineTotal
FROM Sales.SalesOrderDetail d
INNER JOIN Sales.SalesOrderHeader h ON d.SalesOrderID = h.SalesOrderID
WHERE h.OrderDate = '20010707'
AND d.LineTotal < 100 1 1 0 NULL NULL NULL
0 0 |--Compute Scalar(DEFINE:([d].[LineTotal]=[AdventureWorks2008].[Sales].[SalesOrderDetail].[LineTotal] as [d].[LineTotal])) 1 2 1 Compute Scalar
5 1 |--Nested Loops(Inner Join, OUTER REFERENCES:([h].[SalesOrderID])) 1 3 2 Nested Loops
4 1 |--Index Seek(OBJECT:([AdventureWorks2008].[Sales].[SalesOrderHeader].[IX_SalesOrderHeader_OrderDate] AS [h]), SEEK:([h].[OrderDate]='2001-07-07 00:00:00.000') ORDERED FORWARD) 1 4 3 Index Seek
0 0 |--Compute Scalar(DEFINE:([d].[LineTotal]=isnull((CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPrice] as [d].[UnitPrice],0)*((1.0)-CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPriceDiscount] as [d].[UnitPriceDiscount],0)))*CONVERT_IMPLICIT(numeric(5,0),[AdventureWorks2008].[Sales].[SalesOrderDetail].[OrderQty] as [d].[OrderQty],0),(0.000000)))) 1 5 3 Compute Scalar
5 4 |--Clustered Index Seek(OBJECT:([AdventureWorks2008].[Sales].[SalesOrderDetail].[PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID] AS [d]), SEEK:([d].[SalesOrderID]=[AdventureWorks2008].[Sales].[SalesOrderHeader].[SalesOrderID] as [h].[SalesOrderID]), WHERE:([AdventureWorks2008].[Sales].[SalesOrderDetail].[SalesOrderID] as [d].[SalesOrderID]>=(1) AND [AdventureWorks2008].[Sales].[SalesOrderDetail].[SalesOrderID] as [d].[SalesOrderID]<=(999999) AND isnull((CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPrice] as [d].[UnitPrice],0)*((1.0)-CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPriceDiscount] as [d].[UnitPriceDiscount],0)))*CONVERT_IMPLICIT(numeric(5,0),[AdventureWorks2008].[Sales].[SalesOrderDetail].[OrderQty] as [d].[OrderQty],0),(0.000000))<(100.000000)) ORDERED FORWARD) 1 6 5 Clustered Index Seek
and
Rows Executes StmtText StmtId NodeId Parent PhysicalOp
-------------------- -------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------- ----------- ----------- --------------------
5 1 SELECT h.SalesOrderID, h.OrderDate, d.LineTotal
FROM Sales.SalesOrderDetail d
INNER JOIN Sales.SalesOrderHeader h ON h.OrderDate = '20010707' AND d.SalesOrderID = h.SalesOrderID
WHERE d.LineTotal < 100 1 1 0 NULL NULL NULL
0 0 |--Compute Scalar(DEFINE:([d].[LineTotal]=[AdventureWorks2008].[Sales].[SalesOrderDetail].[LineTotal] as [d].[LineTotal])) 1 2 1 Compute Scalar
5 1 |--Nested Loops(Inner Join, OUTER REFERENCES:([h].[SalesOrderID])) 1 3 2 Nested Loops
4 1 |--Index Seek(OBJECT:([AdventureWorks2008].[Sales].[SalesOrderHeader].[IX_SalesOrderHeader_OrderDate] AS [h]), SEEK:([h].[OrderDate]='2001-07-07 00:00:00.000') ORDERED FORWARD) 1 4 3 Index Seek
0 0 |--Compute Scalar(DEFINE:([d].[LineTotal]=isnull((CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPrice] as [d].[UnitPrice],0)*((1.0)-CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPriceDiscount] as [d].[UnitPriceDiscount],0)))*CONVERT_IMPLICIT(numeric(5,0),[AdventureWorks2008].[Sales].[SalesOrderDetail].[OrderQty] as [d].[OrderQty],0),(0.000000)))) 1 5 3 Compute Scalar
5 4 |--Clustered Index Seek(OBJECT:([AdventureWorks2008].[Sales].[SalesOrderDetail].[PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID] AS [d]), SEEK:([d].[SalesOrderID]=[AdventureWorks2008].[Sales].[SalesOrderHeader].[SalesOrderID] as [h].[SalesOrderID]), WHERE:([AdventureWorks2008].[Sales].[SalesOrderDetail].[SalesOrderID] as [d].[SalesOrderID]>=(1) AND [AdventureWorks2008].[Sales].[SalesOrderDetail].[SalesOrderID] as [d].[SalesOrderID]<=(999999) AND isnull((CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPrice] as [d].[UnitPrice],0)*((1.0)-CONVERT_IMPLICIT(numeric(19,4),[AdventureWorks2008].[Sales].[SalesOrderDetail].[UnitPriceDiscount] as [d].[UnitPriceDiscount],0)))*CONVERT_IMPLICIT(numeric(5,0),[AdventureWorks2008].[Sales].[SalesOrderDetail].[OrderQty] as [d].[OrderQty],0),(0.000000))<(100.000000)) ORDERED FORWARD) 1 6 5 Clustered Index Seek
如果您将这些计划与 WinMerge 进行比较，您将看到不同之处：
对于NodeId=1（代表最终结果集的最后一步SELECT ...），您会得到差异，因为 SQL 语句不同。但是，对于前面的步骤 ( NodeId=2..6)，有no differences.
注意 1：在一些奇怪的情况下（ANSI_NULSS ON/OFF），你可能会得到 diff。结果和差异。执行计划：
SET ANSI_NULLS ON;
SELECT *
FROM (VALUES (1,100),(2,200),(3,NULL)) AS Customer(CustomerID,Limit)
INNER JOIN (VALUES (11,1),(12,1),(13,2),(14,3)) AS [Order](OrderID,CustomerID)
ON Customer.CustomerID=[Order].CustomerID
AND Customer.Limit=NULL
SET ANSI_NULLS OFF;
SELECT *
FROM (VALUES (1,100),(2,200),(3,NULL)) AS Customer(CustomerID,Limit)
INNER JOIN (VALUES (11,1),(12,1),(13,2),(14,3)) AS [Order](OrderID,CustomerID)
ON Customer.CustomerID=[Order].CustomerID
WHERE Customer.Limit=NULL
SET ANSI_NULLS ON;
结果：
CustomerID Limit OrderID CustomerID
----------- ----------- ----------- -----------
(0 row(s) affected)
CustomerID Limit OrderID CustomerID
----------- ----------- ----------- -----------
3 NULL 14 3
(1 row(s) affected)
注2：ANSI_NULLS设置应始终为ON。
不要强调哪个会表现更好。您必须了解 SQL 是一种声明性语言而不是一种过程性语言。简单来说，您的查询将由查询优化器转换为将要运行的实际代码。在所有可能的情况下，这将是完全相同的。
您可以查看查询计划确定，但即使它们不同，这种过早的优化也是浪费时间。当您知道自己有问题时进行优化，而不是在编写查询之前进行优化。
它们是相同的，但应该在 JOIN 中。
在过去，JOIN 没有 ON，因此连接在 WHERE 中。但是 ANSI 在 JOIN 中引入了 ON，现代 SQL 引擎会自动将 WHERE 转换为 ON。
在您的场景中，重要的是您在 C.CityId 上有一个索引