Question

IMHO SQL Server can choose itself (unless being told) what is the best index to use for the query.

Ok

What about something like this (pseudo code):

select __a from tbl where __a not in
 (
   select __b  from tbl 

 ) 

(let’s say we have index_1 which is for (__a) and index_2 which is for (__b)

Will SQL Server still use one index at execution or multiple indexes together…?

Answer 1

First, create your tables:

USE tempdb;
GO
CREATE TABLE dbo.tbl(__a INT, __b INT);

Then create two indexes:

CREATE INDEX a_index ON dbo.tbl(__a);
CREATE INDEX b_index ON dbo.tbl(__b);

Now populate with some data:

INSERT dbo.tbl(__a, __b)
  SELECT [object_id], column_id
    FROM sys.all_columns;

Now run your query and turn actual execution plan on. You will see something like this, showing that yes, both indexes are used (in fact the index on __b is used both for data retrieval in the subquery and as a seek to eliminate rows):

A more efficient way to write your query would be:

select __a from dbo.tbl AS t where not exists
 (
   select 1 from dbo.tbl AS t2 
     where t2.__b = t.__a
 );

Now here’s your whole plan (again, both indexes are used, but notice how there are much fewer operations):