如何在SQL Server中使用Forceseek扫描

Question

I am using SQL Server 2012.

My queries are:

create table tab (id int, nm varchar(10))

insert into tab values (1, 'a'), (2, 'b'), (3, 'c')

create index IX_nm on tab(id)
go

create table tab2(id1 int, age int)

insert into tab2 values (1, 21), (2, 22), (3, 54)

create index IX_nm2 on tab2(id1)
go

select * 
from tab a 
inner join tab2 b on a.id = b.id1

select * 
from tab a 
inner join tab2 b with(forceseek) on a.id = b.id1

I have included the execution plans of them.

I am not able to differentiate between the 2 queries, and they produce the same execution plan.

Answer 1

There is no difference between the two plans because SQL Server has already chosen to do an Index Seek on tab2.IX_nm2 without the need for the FORCESEEK hint.

Take a look at the plan produced:

You want all rows from tab , so SQL Server does a table scan. It won't use IX_nm because it doesn't contain nm and would have to do a lookup to retrieve this column from the heap. However, you could use a CLUSTERED index which will remove that table scan:

create CLUSTERED index IX_nm on tab(id);

The Nested Loop between The Scan and the Seek basically says, For every row returned by the Table Scan on tab, get me the corresponding row from tab2 based on the ID . The underlying statistics on IX_nm2 indicates to SQL Server that there is likely to be only one or two rows returned and so it does a Seek for those rows, rather than scans the whole table.

However, because your index is NONCLUSTERED and doesn't include age , SQL Server also has to do a RID Lookup to get the age . Again, you can avoid this by making the index IX_nm2 a CLUSTERED index.

CREATE clustered index CIX_nm2 on tab2(id1);

This would result in a much simpler plan.