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：
复制代码 代码如下:
create table [dbo].[products](
[id] [int] identity(1,1) not null,
[name] [nvarchar](50) null,
[addtime] [datetime] null,
[city] [nvarchar](10) null,
constraint [pk_products] primary key clustered
(
[id] asc
)with (pad_index = off, statistics_norecompute = off, ignore_dup_key = off, allow_row_locks = on, allow_page_locks = on) on [primary]
) on [primary]
1、采用row_number方法，执行5次，平均下来8秒左右，速度最快。
复制代码 代码如下:
select no, id,name,city
from (select no =row_number() over (partition by city order by addtime desc), * from products)t
where no< 11 order by city asc,addtime desc
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。
复制代码 代码如下:
select distinct b.id,b.name,b.city from products a
cross apply (select top 10 * from products where city = a.city order by addtime desc) b
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。
复制代码 代码如下:
select id,name,city from products a
where ( select count(city) from products where a.city = city and addtime>a.addtime) < 10
order by city asc,addtime desc
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。
复制代码 代码如下:
declare @city nvarchar(10)
create table #top(id int,name nvarchar(50),city nvarchar(10),addtime datetime)
declare mycursor cursor for
select distinct city from products order by city asc
open mycursor
fetch next from mycursor into @city
while @@fetch_status =0
begin
insert into #top
select top 10 id,name,city,addtime from products where city = @city
fetch next from mycursor into @city
end
close mycursor
deallocate mycursor
select * from #top order by city asc,addtime desc
drop table #top
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。