sql - SQL Server 2008 包含的列

Question

我最近发现 SQL Server 索引中包含的列。索引中包含的列是否会占用额外的内存，或者它们是否存储在磁盘上？

也有人可以指出将不同数据类型的列包含在主键中的性能影响，在我的情况下，这通常是一个 in？

谢谢。

Answer 1

I don't fully understand the question: "Do included columns in an index take up extra memory or are they stored on disk?" Indexes are both stored on disk (for persistence) and in memory (for performance when being used).

The answer to your question is that the non-key columns are stored in the index and hence are stored both on disk and memory, along with the rest of the index. Included columns do have a significant performance advantage over key columns in the index. To understand this advantage, you have to understand the key values may be stored more than once in a b-tree index structure. They are used both as "nodes" in the tree and as "leaves" (the latter point to the actual records in the table). Non-key values are stored only in leaves, providing potentially a big savings in storage.

Such a savings means that more of the index can be stored in memory in a memory-limited environment. And that an index takes up less memory, allowing memory to be used for other things.

The use of included columns is to allow the index to be a "covering" index for queries, with a minimum of additional overhead. An index "covers" a query when all the columns needed for the query are in the index, so the index can be used instead of the original data pages. This can be a significant performance savings.

The place to go to learn more about them is the Microsoft documentation.

Answer 2

在 SQL Server 2005 或更高版本中，您可以通过在非聚集索引的叶级别添加非键列来扩展非聚集索引的功能。

通过包含非键列，您可以创建涵盖更多查询的非聚集索引。这是因为非键列具有以下好处：

• 它们可以是不允许作为索引键列的数据类型。（除了text、ntext和image之外的所有数据类型都是允许的。）

• 在计算索引键列数或索引键大小时，数据库引擎不会考虑它们。您可以在非聚集索引中包含非键列，以避免超出当前索引大小限制，即最多 16 个键列和最大索引键大小900字节。

当查询中的所有列都作为键或非键列包含在索引中时，包含非键列的索引可以显着提高查询性能。性能提升是因为查询优化器可以定位索引中的所有列值；不访问表或聚集索引数据，从而减少磁盘 I/O 操作。

例子：

Create Table Script
CREATE TABLE [dbo].[Profile](
    [EnrollMentId] [int] IDENTITY(1,1) NOT NULL,
    [FName] [varchar](50) NULL,
    [MName] [varchar](50) NULL,
    [LName] [varchar](50) NULL,
    [NickName] [varchar](50) NULL,
    [DOB] [date] NULL,
    [Qualification] [varchar](50) NULL,
    [Profession] [varchar](50) NULL,
    [MaritalStatus] [int] NULL,
    [CurrentCity] [varchar](50) NULL,
    [NativePlace] [varchar](50) NULL,
    [District] [varchar](50) NULL,
    [State] [varchar](50) NULL,
    [Country] [varchar](50) NULL,
    [UIDNO] [int] NOT NULL,
    [Detail1] [varchar](max) NULL,
    [Detail2] [varchar](max) NULL,
    [Detail3] [varchar](max) NULL,
    [Detail4] [varchar](max) NULL,
PRIMARY KEY CLUSTERED 
(
    [EnrollMentId] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY] TEXTIMAGE_ON [PRIMARY]

GO

SET ANSI_PADDING OFF
GO

存储过程脚本

CREATE Proc [dbo].[InsertIntoProfileTable]
As
BEGIN
SET NOCOUNT ON
Declare @currentRow int
Declare @Details varchar(Max)
Declare @dob Date
set @currentRow =1;
set @Details ='Let''s think about the book. Every page in the book has the page number. All information in this book is presented sequentially based on this page number. Speaking in the database terms, page number is the clustered index. Now think about the glossary at the end of the book. This is in alphabetical order and allow you to quickly find the page number specific glossary term belongs to. This represents non-clustered index with glossary term as the key column.        Now assuming that every page also shows "chapter" title at the top. If you want to find in what chapter is the glossary term, you have to lookup what page # describes glossary term, next - open corresponding page and see the chapter title on the page. This clearly represents key lookup - when you need to find the data from non-indexed column, you have to find actual data record (clustered index) and look at this column value. Included column helps in terms of performance - think about glossary where each chapter title includes in addition to glossary term. If you need to find out what chapter the glossary term belongs - you don''t need to open actual page - you can get it when you lookup the glossary term.      So included column are like those chapter titles. Non clustered Index (glossary) has addition attribute as part of the non-clustered index. Index is not sorted by included columns - it just additional attributes that helps to speed up the lookup (e.g. you don''t need to open actual page because information is already in the glossary index).'
while(@currentRow <=200000)
BEGIN
insert into dbo.Profile values( 'FName'+ Cast(@currentRow as varchar), 'MName' + Cast(@currentRow as varchar), 'MName' + Cast(@currentRow as varchar), 'NickName' + Cast(@currentRow as varchar), DATEADD(DAY, ROUND(10000*RAND(),0),'01-01-1980'),NULL, NULL, @currentRow%3, NULL,NULL,NULL,NULL,NULL, 1000+@currentRow,@Details,@Details,@Details,@Details)
set @currentRow +=1;
END

SET NOCOUNT OFF
END

GO

使用上面的SP可以一次插入200000条记录。

您可以看到“EnrollMentId”列上有一个聚集索引。

现在在“UIDNO”列上创建一个非聚集索引。

脚本

CREATE NONCLUSTERED INDEX [NonClusteredIndex-20140216-223309] ON [dbo].[Profile]
(
    [UIDNO] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

现在运行以下查询

select UIDNO,FName,DOB, MaritalStatus, Detail1 from dbo.Profile --Takes about 30-50 seconds and return 200,000 results.

查询 2

select UIDNO,FName,DOB, MaritalStatus, Detail1 from dbo.Profile
where DOB between '01-01-1980' and '01-01-1985'
 --Takes about 10-15 seconds and return 36,479 records.

现在删除上面的非聚集索引并使用以下脚本重新创建

CREATE NONCLUSTERED INDEX [NonClusteredIndex-20140216-231011] ON [dbo].[Profile]
(
    [UIDNO] ASC,
    [FName] ASC,
    [DOB] ASC,
    [MaritalStatus] ASC,
    [Detail1] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

它将引发以下错误 Msg 1919, Level 16, State 1, Line 1 表 'dbo.Profile' 中的列 'Detail1' 的类型不能用作索引中的键列。

因为我们不能使用 varchar(Max) 数据类型作为键列。

现在使用以下脚本创建包含列的非聚集索引

CREATE NONCLUSTERED INDEX [NonClusteredIndex-20140216-231811] ON [dbo].[Profile]
(
    [UIDNO] ASC
)
INCLUDE (   [FName],
    [DOB],
    [MaritalStatus],
    [Detail1]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

现在运行以下查询

select UIDNO,FName,DOB, MaritalStatus, Detail1 from dbo.Profile --Takes about 20-30 seconds and return 200,000 results.

查询 2

select UIDNO,FName,DOB, MaritalStatus, Detail1 from dbo.Profile
where DOB between '01-01-1980' and '01-01-1985'
 --Takes about 3-5 seconds and return 36,479 records.

Answer 3

包含列提供类似于聚集索引的功能，其中行内容保存在主索引的叶节点中。除了索引中的键列之外，索引表叶节点中还保留了其他属性。

这允许立即访问列值，而无需访问数据库中的另一个页面。增加索引大小和一般存储与不必通过索引中的页面引用间接实现的改进响应之间存在折衷。向表中添加多个索引可能会产生类似的影响。

Answer 4

从这里：-

当查询中的所有列都作为键或非键列包含在索引中时，具有非键列的索引可以显着提高查询性能。性能提升是因为查询优化器可以定位索引中的所有列值；不访问表或聚集索引数据，从而减少磁盘 I/O 操作。