SQLServer地址搜索性能优化

2017-02-18 17:02:11 SQLserver
网上关于SQL优化的教程很多,但是比较杂乱。今天我们来从一个简单的例子来探讨下SQLServer地址搜索性能优化\

这是一个很久以前的例子,现在在整理资料时无意发现,就拿出来再改写分享。

1.需求

 1.1 基本需求: 根据输入的地址关键字,搜索出完整的地址路径,耗时要控制在几十毫秒内。

 1.2 数据库地址表结构和数据:

 表TBAddress

表数据

1.3 例子:

 

 e.g. 给出一个字符串如“广 大”,找出地址全路径中包含有“广” 和“大”的所有地址,結果如下:

下面将通过4个方法来实现,再分析其中的性能优劣,然后选择一个比较优的方法。

 2.创建表和插入数据

 

 2.1 创建数据表TBAddress

use test;
go
/* create table */
if object_id('TBAddress') is not null
  drop table TBAddress;
go
create table TBAddress
(
 ID int ,
 Parent int not null ,
 LevelNo smallint not null ,
 Name nvarchar(50) not null ,
 constraint PK_TBAddress primary key ( ID )
);
go
create nonclustered index ix_TBAddress_Parent on TBAddress(Parent,LevelNo) include(Name) with(fillfactor=80,pad_index=on);
create nonclustered index ix_TBAddress_Name on TBAddress(Name)include(LevelNo)with(fillfactor=80,pad_index=on);
go

create table

 

2.2 插入数据

use test
go
/*insert data*/
set nocount on
Begin Try
  Begin Tran
  Insert Into TBAddress ([ID],[Parent],[LevelNo],[Name])
    Select 1,0,0,N'中国' Union All 
    Select 2,1,1,N'直辖市' Union All 
    Select 3,1,1,N'辽宁省' Union All 
    Select 4,1,1,N'广东省' Union All 
    ... ...
    Select 44740,930,4,N'奥依塔克镇' Union All 
    Select 44741,932,4,N'巴音库鲁提乡' Union All 
    Select 44742,932,4,N'吉根乡' Union All 
    Select 44743,932,4,N'托云乡'
  Commit Tran
End Try
Begin Catch
  throw 50001,N'插入數據過程中發生錯誤.' ,1
Rollback Tran
End Catch
go

附件: insert Data

 Note: 数据有44700条,insert代码比较长,所以采用附件形式。

3.测试,方法1

3.1 分析:

a. 先搜索出包字段Name中含有“广”、“大”的所有地址记录存入临时表#tmp。

  b. 再找出#tmp中各个地址到Level 1的全路径。

    c. 根据步骤2所得的结果,筛选出包含有“广”和“大”的地址路径。

      d. 根据步骤3筛选的结果,查询所有到Level n(n为没有子地址的层编号)的地址全路径。

3.2 存储过程代码:

Use test
Go
if object_ID('[up_SearchAddressByNameV0]') is not null
  Drop Procedure [up_SearchAddressByNameV0]
Go
create proc up_SearchAddressByNameV0 
(
  @Name nvarchar(200)
)
As
set nocount on
declare @sql nvarchar(max)
 
declare @tmp Table (Name nvarchar(50))
 
set @Name=@Name+' '
 
while patindex('% %',@Name)>0
begin
  set @Name=replace(@Name,' ',' ')  
end
 
set @sql ='select ''' +replace(@Name,' ',''' union all select ''')+''''
insert into @tmp(Name) exec(@sql)
 
if object_id('tempdb..#tmp') is not null drop table #tmp
if object_id('tempdb..#') is not null drop table #
 
create table #tmp(ID int )
 
 
while @Name>''
begin
  insert into #tmp(ID)
  select a.ID from TBAddress a where a.Name like '%'+substring(@Name,1,patindex('% %',@Name)-1)+'%' 
 
  set @Name=Stuff(@Name,1,patindex('% %',@Name),'')
end
 
 
;with cte_SearchParent as
(
  select a.ID,a.Parent,a.LevelNo,convert(nvarchar(500),a.Name) as AddressPath from TBAddress a where exists(select 1 from #tmp x where a.ID=x.ID) 
  union all
  select a.ID,b.Parent,b.LevelNo,convert(nvarchar(500),b.Name+'/'+a.AddressPath) as AddressPath
    from cte_SearchParent a
    inner join TBAddress b on b.ID=a.Parent
      --and b.LevelNo=a.LevelNo -1
      and b.LevelNo>=1
)
select a.ID,a.AddressPath 
  into #
  from cte_SearchParent a 
  where a.LevelNo=1 and exists(select 1 from @tmp x where a.AddressPath like '%'+x.Name+'%' having count(1)=(select count(1) from @tmp))
 
;with cte_result as
(
  select a.ID,a.LevelNo,b.AddressPath
    from TBAddress a 
      inner join # b on b.ID=a.ID
  union all
  select b.ID,b.LevelNo,convert(nvarchar(500),a.AddressPath+'/'+b.Name) As AddressPath
    from cte_result a
      inner join TBAddress b on b.Parent=a.ID
        --and b.LevelNo=a.LevelNo+1
            
)
select distinct a.ID,a.AddressPath 
  from cte_result a 
  where not exists(select 1 from TBAddress x where x.Parent=a.ID)
  order by a.AddressPath 
Go
procedure:up_SearchAddressByNameV0
3.3 执行查询:
exec up_SearchAddressByNameV0 '广 大'

共返回195行记录。

3.4 客户端统计信息:

平均的执行耗时:  244毫秒

4.测试,方法2

 方法2是参照方法1,并借助全文索引来优化方法1中的步骤1。也就是在name列上建立全文索引,在步骤1中,通过全文索引搜索出包字段Name中含有“广”、“大”的所有地址记录存入临时表#tmp,其他步骤保持不变。

 

 4.1 创建全文索引

use test
go
/*create fulltext index*/
if not exists(select 1 from sys.fulltext_catalogs a where a.name='ftCatalog')
begin
create fulltext catalog ftCatalog As default;
end
go
--select * From sys.fulltext_languages    
create fulltext index on TBAddress(Name language 2052 ) key index PK_TBAddress
go   
alter fulltext index on dbo.TBAddress add(Fullpath language 2052)
go
Note:  在Name列上创建全文索引使用的语言是简体中文(Simplified Chinese)
4.2 存储过程代码:
Use test
Go
if object_ID('[up_SearchAddressByNameV1]') is not null
  Drop Procedure [up_SearchAddressByNameV1]
Go
create proc up_SearchAddressByNameV1 
(
  @Name nvarchar(200)
)
As
set nocount on
declare @sql nvarchar(max),@contains nvarchar(500)
 
declare @tmp Table (Name nvarchar(50))
 
while patindex('% %',@Name)>0
begin
  set @Name=replace(@Name,' ',' ')  
end
 
set @sql ='select ''' +replace(@Name,' ',''' union all select ''')+''''
set @contains='"'+replace(@Name,' ','*" Or "')+'*"'
 
insert into @tmp(Name) exec(@sql)
 
if object_id('tempdb..#') is not null drop table #
 
;with cte_SearchParent as
(
  select a.ID,a.Parent,a.LevelNo,convert(nvarchar(2000),a.Name) as AddressPath from TBAddress a where exists(select 1 from TBAddress x where contains(x.Name,@contains) And x.ID=a.ID) 
  union all
  select a.ID,b.Parent,b.LevelNo,convert(nvarchar(2000),b.Name+'/'+a.AddressPath) as AddressPath
    from cte_SearchParent a
    inner join TBAddress b on b.ID=a.Parent
      --and b.LevelNo=a.LevelNo -1
      and b.LevelNo>=1
)
select a.ID,a.AddressPath 
  into #
  from cte_SearchParent a 
  where a.LevelNo=1 and exists(select 1 from @tmp x where a.AddressPath like '%'+x.Name+'%' having count(1)=(select count(1) from @tmp))
 
;with cte_result as
(
  select a.ID,a.LevelNo,b.AddressPath
    from TBAddress a 
      inner join # b on b.ID=a.ID
  union all
  select b.ID,b.LevelNo,convert(nvarchar(2000),a.AddressPath+'/'+b.Name) As AddressPath
    from cte_result a
      inner join TBAddress b on b.Parent=a.ID
        --and b.LevelNo=a.LevelNo+1
            
)
select distinct a.ID,a.AddressPath 
  from cte_result a 
  where not exists(select 1 from TBAddress x where x.Parent=a.ID)
  order by a.AddressPath 
Go

procedure:up_SearchAddressByNameV1

4.3测试存储过程:

exec up_SearchAddressByNameV1 '广 大'

从平均耗时方面分析,一眼就知道方法3比较符合开始的需求(耗时要控制在几十毫秒内)。

当然还有其他的方法,如通过程序实现,把数据一次性加载至内存中,再通过程序写的算法进行搜索,或通过其他工具如Lucene来实现。不管哪一种方法,我们都是选择最优的方法。实际的工作经验告诉我们,在实际应用中,多选择和测试不同的方法来,选择其中一个满足我们环境的,而且是最优的方法。