提高删除查询的性能

Question

I want to delete entries from table B so there is only a single entry per A_id (the one with highest ID)

Table A :

+----+------------+
| id |    name    |
+----+------------+
|  1 | Some name  |
|  2 | Other name |
+----+------------+

Table B :

+----+-------+------+
| id | stuff | A_id |
+----+-------+------+
|  1 | aab   |    1 |
|  2 | aac   |    1 |
|  3 | aad   |    2 |
|  4 | aae   |    1 |
|  5 | aak   |    1 |
|  6 | aal   |    2 |
+----+-------+------+

My current query (that works fine):

DELETE FROM B 
WHERE id NOT IN (SELECT MAX(id)
                 FROM B
                 GROUP BY A_id)

Which results in the correct result:

+----+-------+------+
| id | stuff | A_id |
+----+-------+------+
|  5 | aak   |    1 |
|  6 | aal   |    2 |
+----+-------+------+

But it is very very slow when there are many rows in table B. Is there any way to improve the performance of the query (or perhaps do it in an entirely different way?)

Answer 1

You are deleting a large number of rows. That is the problem. There is lots of overhead in deletions.

If you are deleting a significant number of rows in a table -- and significant might only be a few percent -- then it is often faster to recreate the table:

select b.*
into temp_b  -- actually, I wouldn't use a temporary table in case the server goes down
from b
where b.id = (select max(a.id) from b b2 where b2.id = b.a_id);

truncate table b;

insert into b
    select *
    from temp_b;

Before attempting this, be sure that you have backed up b or at least stashed a copy of it somewhere.

Note that I changed the structure of the NOT IN . I strongly discourage the use of NOT IN , because the semantics are not intuitive when the subquery returns NULL values. If there were a single NULL value, then the WHERE would never evaluate to TRUE. Even if NULL values are not a problem in this case, I strongly recommend using other alternatives so you won't have a problem when NULL s are a possibility.

For performance on the SELECT , you want an index on b(a_id, id) . You might find that such an index helps on your original query.

Answer 2

Your query looks fine to me.

Your problem seems to be that you have a very large amount of data and need ways to optimize performance.

What you can do is materialize your subquery, and make sure max_id is indexed, for example by making it a primary key.

So create a temporary table Max_B , and store the results of your sub query in this table. Then perform the delete and drop the temp table afterwards.