Question

I have a sql query,

SELECT 
  col1, 
  col2, 
  col3, (
    SELECT COUNT(id) 
    FROM B 
    WHERE B.aid = A.id
  ) AS col4 
FROM A

What is the performance impact of having a select as a column? Will that statement be executed for every row that is returned?

I am really just interested in the performance of this query. I know there are other ways getting the count can be accomplished. But in this case I am only trying to understand how sql works with an inline select.

Apologies if this question is a dupe, I have looked through stackoverflow and I have not been able to find this anywhere.

Answer 1

This depends upon your SQL implementation.

SQL is declarative and it is up to the optimiser to construct a physical plan from the logical specification. SQL Server can decorrelate this sub query and turn into an OUTER JOIN

CREATE TABLE A(id INT PRIMARY KEY, col1 INT, col2 INT, col3 INT)

CREATE TABLE B(aid INT)

CREATE CLUSTERED INDEX ix ON B(aid)

/*Fool optimiser into thinking tables aren't empty*/
update statistics A with rowcount = 1000000, pagecount = 100000

update statistics B with rowcount = 1000000, pagecount = 100000

SELECT col1,
       col2,
       col3,
       (SELECT count( B.aid)
        FROM   B
        WHERE  B.aid = A.id) AS col4
FROM   A 

DROP TABLE A, B 

Gives plan

Which is basically the same as

SELECT col1,
       col2,
       col3,
       Cnt
FROM   (SELECT COUNT(aid) AS Cnt,
               aid
        FROM   B
        GROUP  BY aid) T
       RIGHT OUTER JOIN A
         ON A.id = T.aid 

The sub query is logically represented as a RIGHT OUTER JOIN with MERGE JOIN as the physical implementation. The merge join processes each input once rather than the row by row behaviour of a nested loops join.