SQL Tip of the Day: Be Wary of SELECT COUNT(*)


Recently, I’ve encountered this sort of query all over the place at a customer site:

DECLARE
  v_var NUMBER(10);
BEGIN
  SELECT COUNT(*)
  INTO   v_var
  FROM   table1
  JOIN   table2 ON table1.t1_id = table2.t1_id
  JOIN   table3 ON table2.t2_id = table3.t2_id
  ...
  WHERE  some_predicate;

  IF (v_var = 0) THEN
    do_something
  ELSE
    do_something_else
  END IF;
END;

Unfortunately, COUNT(*) is often the first solution that comes to mind when we want to check our relations for some predicate. But COUNT() is expensive, especially if all we’re doing is checking our relations for existence. Does the word ring a bell? Yes, we should use the EXISTS predicate, because if we don’t care about the exact number of records that return true for a given predicate, we shouldn’t go through the complete data set to actually count the exact number. The above PL/SQL block can be rewritten trivially to this one:

DECLARE
  v_var NUMBER(10);
BEGIN
  SELECT CASE WHEN EXISTS (
    SELECT 1
    FROM   table1
    JOIN   table2 ON table1.t1_id = table2.t1_id
    JOIN   table3 ON table2.t2_id = table3.t2_id
    ...
    WHERE  some_predicate
  ) THEN 1 ELSE 0 END
  INTO   v_var
  FROM   dual;

  IF (v_var = 0) THEN
    do_something
  ELSE
    do_something_else
  END IF;
END;

Let’s measure!

Query 1 yields this execution plan:

-----------------------------------------------
| Id  | Operation           | E-Rows | A-Rows |
-----------------------------------------------
|   0 | SELECT STATEMENT    |        |      1 |
|   1 |  SORT AGGREGATE     |      1 |      1 |
|*  2 |   HASH JOIN         |      4 |      4 |
|*  3 |    TABLE ACCESS FULL|      2 |      2 |
|*  4 |    TABLE ACCESS FULL|      6 |      6 |
-----------------------------------------------

Query 2 yields this execution plan:

----------------------------------------------
| Id  | Operation          | E-Rows | A-Rows |
----------------------------------------------
|   0 | SELECT STATEMENT   |        |      1 |
|   1 |  NESTED LOOPS      |      4 |      1 |
|*  2 |   TABLE ACCESS FULL|      2 |      1 |
|*  3 |   TABLE ACCESS FULL|      2 |      1 |
|   4 |  FAST DUAL         |      1 |      1 |
----------------------------------------------

You can ignore the TABLE ACCESS FULL operations, the actual query was executed on a trivial database with no indexes.

What’s essential, however, are the much improved E-Rows values (E = Estimated) and even more importantly the optimal A-Rows values (A = Actual). As you can see, the EXISTS predicate could be aborted early, as soon as the first record that matches the predicate is encountered – in this case immediately.

See this post about more details of how to collect Oracle Execution plans

Conclusion

Whenever you encounter a COUNT(*) operation, you should ask yourself if it is really needed. Do you really need to know the exact number of records that match a predicate? Or are you already happy knowing that any record matches the predicate?

Answer: It’s probably the latter.

Tags: , ,

2 responses to “SQL Tip of the Day: Be Wary of SELECT COUNT(*)”

  1. Steve Jakob says :

    Please note, the example SQL query at the beginning of this post does NOT test for existence; it tests that there is one AND ONLY ONE record that meets the query criteria. If there are 2 matching records then the original example would call “do_something_else” whereas your PL/SQL code would call “do_something”.

    Had the “IF” statement in the original example started with:

    IF (v_var > 0) THEN

    … then your premise would have been correct.

    Love your articles. I’ve a learned a lot reading them, and you’re one of the few authors worth reading on the “java.dzone.com” site.

    • lukaseder says :

      You’re right of course. That was a “copy paste” error on my part. The original code checked for v_var = 0. Will fix that in the article.

      Thanks for your nice words. I’ll certainly continue blogging about the Java / SQL experience made here at Data Geekery.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 2,084 other followers

%d bloggers like this: