SQL server join与子查询性能问题

Question

I discovered that in some cases a query like

select 
   usertable.userid,
   (select top 1 name from nametable where userid = usertable.userid) as name 
from usertable 
where active = 1

takes an order of magnitude longer to complete in SS2008R2 than the equivalent join query

select 
   usertable.userid,
   nametable.name 
from usertable 
left join nametable on nametable.userid = usertable.userid 
where usertable.active = 1

where both tables are indexed and have over 100k rows. Interestingly, inserting a top clause into the original query makes it perform on par with the join query:

select 
    top (select count(*) from usertable where active = 1) usertable.userid,
    (select top 1 name from nametable where userid = usertable.userid) as name 
from usertable 
where active = 1

Does anyone have any idea why the original query performs so poorly?

Answer 1

Well, the queries are different - unless the userid column is a primary key or has a uniqueness constraint then the second query could return more rows than the first.

That said, with the assumption that userid is a primary key / unique try removing the TOP 1 part of the first subquery:

select 
   usertable.userid,
   (select name from nametable where userid = usertable.userid) as name 
from usertable 
where active = 1

Answer 2

It's a correlated subquery, which means it needs to execute once per return row of the outer query since it references a field in the outer query.

A JOIN runs once for the entire result set and gets merged. Your subquery runs the outer query, then for each returned row it runs the subquery again.

Answer 3

The original query will execute the sub select as many times as there are rows, thus the poor performance.

When you JOIN you get the whole result set at once.