使用复合聚集索引时 SQL Server 会跳转吗？

Question

考虑以下复合聚集索引：

CREATE UNIQUE CLUSTERED INDEX ix_mytable ON mytable(a, b)

显然，b 上的单独索引将使搜索 b 的特定值更快。

但是，如果 b 上的单独索引不 被使用，在我看来，复合索引仍然可以用于查找具有特定值的元组 b 不是表扫描，而是通过遍历 a 的离散值树并在本地搜索 b，跳转到下一个值一个等

这就是 SQL Server 的工作方式吗？ （例如，如果 MSSQL 对具有多列的索引使用单个哈希值，则不会。）

确实如此，并且由于其他原因已经需要复合索引，并且 a 的离散值的数量足够少，性能/空间的权衡可能会偏离拥有 a b 的单独索引。

（上面的 UNIQUE 和 CLUSTERED 约束对于这个例子来说并不是真正需要的，但它们代表了对 b 的最快检索，它不涉及 b的单独索引>--前者为a的每个循环提供了一个快捷方式，后者在查找中删除了一个间接度）。

Answer 1

USE AdventureWorks2008R2;
-- Source: http://msftdbprodsamples.codeplex.com/releases/view/59211
GO

SET NOCOUNT ON;
GO

CREATE NONCLUSTERED INDEX IX_SalesOrderHeader_OrderDate_#_ShipDate_SubTotal
ON [Sales].[SalesOrderHeader] ([OrderDate])
INCLUDE (ShipDate,SubTotal)
-- WITH(DROP_EXISTING=ON);
GO

-- Test 1
SET STATISTICS IO ON;
SELECT  COUNT(*)
FROM    Sales.SalesOrderHeader h -- Index Seek on IX_SalesOrderHeader_OrderDate_#_ShipDate_SubTotal
WHERE   h.OrderDate BETWEEN '2008-07-01T00:00:00.000' AND '2008-07-15T23:59:59.997';
SET STATISTICS IO OFF;
GO
-- End of Test 1
-- Results:
-- Table 'SalesOrderHeader'. Scan count 1, logical reads 5, physical reads 0

DROP INDEX IX_SalesOrderHeader_OrderDate_#_ShipDate_SubTotal
ON [Sales].[SalesOrderHeader]
GO
CREATE NONCLUSTERED INDEX [IX_SalesOrderHeader_ShipMethodID_OrderDate_#_ShipDate_SubTotal] 
ON Sales.SalesOrderHeader 
(
    ShipMethodID ASC,
    OrderDate ASC
)
INCLUDE (ShipDate,SubTotal);
GO

-- Test 2
SET STATISTICS IO ON;
SELECT  COUNT(*)
FROM    Sales.SalesOrderHeader h -- Index Scan on IX_SalesOrderHeader_ShipMethodID_OrderDate_#_ShipDate_SubTotal
WHERE   h.OrderDate BETWEEN '2008-07-01T00:00:00.000' AND '2008-07-15T23:59:59.997';
SET STATISTICS IO OFF;
GO
-- End of Test 2
-- Results:
-- Table 'SalesOrderHeader'. Scan count 1, logical reads 150, physical reads 0

-- Test 3
SET STATISTICS IO ON;
SELECT  COUNT(*)
FROM    Purchasing.ShipMethod sm
INNER JOIN Sales.SalesOrderHeader h ON h.ShipMethodID=sm.ShipMethodID -- FK elimination + Index Scan on IX_SalesOrderHeader_ShipMethodID_OrderDate_#_ShipDate_SubTotal
WHERE   h.OrderDate BETWEEN '2008-07-01T00:00:00.000' AND '2008-07-15T23:59:59.997';
SET STATISTICS IO OFF;
GO
-- End of Test 3
-- Results:
-- Table 'SalesOrderHeader'. Scan count 1, logical reads 150, physical reads 0

-- Test 4
SET STATISTICS IO ON;
SELECT  MIN(sm.ShipMethodID) AS DummnyCol, -- To prevent FK elimination 
        COUNT(*)
FROM    Purchasing.ShipMethod sm
INNER JOIN Sales.SalesOrderHeader h ON h.ShipMethodID=sm.ShipMethodID -- Index Seek on IX_SalesOrderHeader_ShipMethodID_OrderDate_#_ShipDate_SubTotal
WHERE   h.OrderDate BETWEEN '2008-07-01T00:00:00.000' AND '2008-07-15T23:59:59.997';
SET STATISTICS IO OFF;
GO
-- End of Test 4
-- Results:
-- Table 'SalesOrderHeader'. Scan count 5, logical reads 13, physical reads 0
-- Table 'ShipMethod'. Scan count 1, logical reads 2, physical reads 0

DROP INDEX [IX_SalesOrderHeader_ShipMethodID_OrderDate_#_ShipDate_SubTotal] 
ON Sales.SalesOrderHeader;
GO
SET NOCOUNT OFF;
GO

Answer 2

不，没有跳过“a”簇。只有指定了最左边的列才能使用索引，否则需要使用全扫描。

Oracle 有所谓的'Index Skip Scan' 运算符。