(
有产品表，包含id，name，city，addtime四个字段，因报表需要按城市分组，统计每个城市的最新10个产品，便向该表中插入了100万数据，做了如下系列测试：
drop table #Top
cross apply (select top 10 * from products where city = a.city order by addtime desc) b
select top 10 id,name,city,addtime from products where city = @city
select no, id,name,city
where ( select count(city) from products where a.city = city and addtime>a.addtime) < 10
open mycursor
2、采用cross apply方法，执行了3次，基本都在3分5秒以上，已经很慢了。
select id,name,city from products a
[id] [int] IDENTITY(1,1) NOT NULL,
fetch next from mycursor into @city
deallocate mycursor
select distinct city from products order by city asc

[name] [nvarchar](50) NULL,
begin
create table #Top(id int,name nvarchar(50),city nvarchar(10),addtime datetime)

复制代码 代码如下:

[addtime] [datetime] NULL,

while @@fetch_status =0
3、采用Count查询，只执行了两次，第一次执行到5分钟时，取消任务执行了；第二次执行到13分钟时，没有hold住又直接停止了，实在无法忍受。

end

from (select no =row_number() over (partition by city order by addtime desc), * from products)t

复制代码 代码如下:

1、采用row_number方法，执行5次，平均下来8秒左右，速度最快。

CONSTRAINT [PK_products] PRIMARY KEY CLUSTERED
通过上述对比不难发现，在面临Group获取Top N场景时，可以首选row_number，游标cursor其次，另外两个就基本不考虑了，数据量大的时候根本没法使用。

复制代码 代码如下:

复制代码 代码如下:

order by city asc,addtime desc
select distinct b.id,b.name,b.city from products a
4、采用游标方法，这个最后测试的，执行了5次，每次都是10秒完成，感觉还不错。

close mycursor

insert into #Top

复制代码 代码如下:

declare @city nvarchar(10)
) ON [PRIMARY]
Select * from #Top order by city asc,addtime desc
[city] [nvarchar](10) NULL,
CREATE TABLE [dbo].[products](
fetch next from mycursor into @city
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
where no< 11 order by city asc,addtime desc
declare mycursor cursor for
[id] ASC