为什么联接受限表比限制联接表快得多？

Question

I'm new in database algorithms and don't know how it works internally.

So, I tried to write the following two queries to compare the performance:

SELECT * 
FROM (
    SELECT *
    FROM account
    WHERE id < 20000
) as dpa_r
FULL OUTER JOIN (
    SELECT *
    FROM transactions
    WHERE id < 20000
) as pet_r
ON SOME_CONDITION
FULL OUTER JOIN (
    SELECT *
    FROM profit
    WHERE id < 20000
) as dp_r
ON SOME_CONDITION

OFFSET 200000 LIMIT 20

The query execution time = 940ms.

I rewrote the query as follows:

SELECT * 
FROM (
    SELECT *
    FROM account
) as dpa_r
FULL OUTER JOIN (
    SELECT *
    FROM transactions
) as pet_r
ON SOME_CONDITION
FULL OUTER JOIN (
    SELECT *
    FROM profit
) as dp_r
ON SOME_CONDITION
WHERE dp_r.id < 20000 AND dpa_r.id < 20000 AND pet_r.id < 20000
OFFSET 200000 LIMIT 20

The query execution time is 17321ms.

Why is there so difference in time?

Answer 1

You're two queries give different results, one of them is an outer join query, but the other one is an inner join query. Probably the reason why you get different execution times.

With column conditions in the WHERE clause, your outer joins perform as regular inner joins:

SQL>create table t1 (c1 int, c2 char(10));
SQL>create table t2 (c1 int, c2 char(10));
SQL>insert into t1 values (1,'One');
SQL>insert into t1 values (2,'Two');
SQL>insert into t2 values (1,'One');
SQL>insert into t2 values (5,'Five');
SQL>select * from t1 left join t2 on t1.c1 = t2.c1
SQL&where t1.c1 < 10 and t2.c1 < 10;
         c1 c2                  c1 c2
=========== ========== =========== ==========
          1 One                  1 One

                  1 row found

SQL>select * from t1 left join t2 on t1.c1 = t2.c1 
SQL&  and t1.c1 < 10 and t2.c1 < 10;
         c1 c2                  c1 c2
=========== ========== =========== ==========
          1 One                  1 One
          2 Two                  - -

                  2 rows found

When those conditions were moved to the ON clause, you get true OUTER JOIN behavoir!