Optimizing SELECT COUNT to EXISTS

Question

I have a query to find certain customers from a table.

SELECT COUNT(*)
  FROM CUSTOMER
 WHERE amount <> 0
   AND customerid = 22

There is an index on customerid, so the DB scans all rows with customerid = 22.

Since the result is processed by checking whether the count returns zero or more than zero, how can I optimize the query? I.e. such that at the first customer row with amount <> 0 the query returns 0 else if all rows are = 0, then return 1.

Related: http://stackoverflow.com/questions/2759756/is-it-possible-to-select-exists-directly-as-a-bit — RQDQ, Feb 17 '11 at 16:16

Martin Smith · Accepted Answer · 2022-02-15T13:17:44.350

34

select case
         when exists (select *
                      from   customer
                      where  amount <> 0
                             and customerid = 22) then 1
         else 0
       end  as non_zero_exists

edited Feb 15 '22 at 13:17

answered Feb 17 '11 at 16:14

Martin Smith

438,706
87
741
845

4

Would it be more efficient to replace 'select *' with 'select customerid' or maybe 'select 1' ? – Carlo V. Dango Feb 17 '11 at 22:17
1

@CarloV.Dango from memory SQL Server 2005 and up has added optimizations that makes exists and count checks like above the same speed. – Bertus Kruger Mar 04 '14 at 01:03
@BertusKruger That only applies if the `COUNT` test against `0` is in the query, which it is not in the original question. I wonder if `TOP 1` would optimize it. – NetMage Mar 26 '19 at 17:19

Chris Shain · Answer 2 · 2011-02-17T16:49:47.453

8

First index on customerid and amount

CREATE INDEX customer_idx ON customer(customerid, amount);

then rewrite your query as

IF EXISTS (SELECT customerid
    FROM customer
    WHERE amount > 0 -- I am assuming here that amount cannot be a negative number.
    AND customerid = 22)
   SELECT 1
ELSE
   SELECT 0

This should result in an index seek on customer_idx. Otherwise you'll need to scan all rows for that customer (which your question seems to imply could be a lot).

edited Feb 17 '11 at 16:49

answered Feb 17 '11 at 16:16

Chris Shain

50,833
6
93
125

1

+1 the solution to *every* SQL performance question is first and foremost to create an appropriate index. – Remus Rusanu Feb 17 '11 at 16:26
2

@Chris this produces `Incorrect syntax near the keyword 'exists although the index bit is nice – Conrad Frix Feb 17 '11 at 16:28
Sorry, was coding in the SO window, always a bad decision. Updated to use something more like your query. +1 to your answer. – Chris Shain Feb 17 '11 at 16:48
1

Yes, but so can long running, blocking queries that are not operating on properly indexed data. – Chris Shain Feb 17 '11 at 22:20
And for what it's worth, they don't cause deadlocks. They may exacerbate deadlocks caused by insert-heavy workloads, however. – Chris Shain Feb 17 '11 at 22:21

score 4 · Answer 3 · answered Feb 17 '11 at 16:22

4

Seems straight forward enough

IF EXISTS ( SELECT customerid
            FROM   customer
            WHERE  amount <> 0
                  and customerid = 22))
   SELECT 1
ELSE
   SELECT 0

answered Feb 17 '11 at 16:22

Conrad Frix

51,984
12
96
155

score 2 · Answer 4 · answered Feb 17 '11 at 18:13

2

An alternative to EXISTS

select ISNULL((select TOP 1 1
               from customer
               where amount <> 0
                 and customerid = 22),0)

I already assumed that you will have an index on (customerid) or better (customerid,amount).

answered Feb 17 '11 at 18:13

RichardTheKiwi

105,798
26
196
262

Optimizing SELECT COUNT to EXISTS

4 Answers4

Linked