MySQl查询：加入和分组，性能降低

Question

I have a problem with a MySQL Query and I'm not able to optimize it.

SELECT 
  p.id, 
  p.name,
  p.sku, 
  p.type
FROM 
  xm_products p 
  LEFT JOIN xm_store_product sp ON p.id = sp.product_id 
  LEFT JOIN xm_store s ON sp.store_id = s.id 
WHERE 
  s.id = 1 
ORDER BY 
  p.type, 
  p.name asc 
LIMIT 
  20 OFFSET 0

This Query is very slow: Querytime 2.532s

If I remove the Order By clause, the query is very fast: 0.0001s

Explain show the following information:

+----+-------------+-------+--------+-------------------------------------------+----------------------+---------+---------------------------------------+--------+----------------------------------------------+
| id | select_type | table | type   | possible_keys                             | key                  | key_len | ref                                   | rows   | Extra                                        |
+----+-------------+-------+--------+-------------------------------------------+----------------------+---------+---------------------------------------+--------+----------------------------------------------+
|  1 | SIMPLE      | s     | const  | PRIMARY                                   | PRIMARY              | 4       | const                                 |      1 | Using index; Using temporary; Using filesort |
|  1 | SIMPLE      | sp    | ref    | IDX_CA42254AB092A811,IDX_CA42254A4584665A | IDX_CA42254AB092A811 | 5       | const                                 | 102157 | Using where                                  |
|  1 | SIMPLE      | p     | eq_ref | PRIMARY                                   | PRIMARY              | 4       | model.sp.product_id                   |      1 |                                              |
+----+-------------+-------+--------+-------------------------------------------+----------------------+---------+---------------------------------------+--------+----------------------------------------------+
3 rows in set

I have 3 tables:

• xm_product with primary key id
• xm_store with primary key id
• xm_store_product with index on store_id and product_id

I tried adding an index on p.name and p.type and also a combined index (p.name, p.type) but it did not help.

How can I optimize the performance of this query?

EDIT:

I took me 2 hours to create a sqlfidle. But here it is with some data in it. http://sqlfiddle.com/#!2/2bf8d/1

The problem is the "Group By" causing "Using index; Using temporary; Using filesort".

How can I correct my example?

EDIT 2:

De xm_products table has about 200'000 records. The xm_store_products about 400'000. The Query is for a pager, the limit to 20 ist per page.

Answer 1

I'll put it as an answer after the positive feedback :)

That there is a (noticable) difference in performance makes a lot of sense.

I don't know how big your tables are, but a LIMIT 20 will only increase performance without an order by. With the order by, all records will first have to be retrieved before the order by can happen, whereas without the order by, execution stops after 20 matches.

Maybe with a combined index on your order-by-clause and a view on that you could gain performance.

Answer 2

Your problem is that the MySQL optimizer is cost based.

So it calculates the best plan for accessing your tables in this case MySQL has choosen the wrong order.. because disk i/o is more expensive vs CPU cycles based on wait times..

You sort on table product (p.type, p.name asc) but this table is accessed last in the explain plan so MySQL needs to build an temporary table "Using temporary" to hold the results.. and because the results are not correctly sorted an extra quicksort "using filesort" is needed..

You can test with STRAIGHT_JOIN so the MySQL optimizer is "bypassed" note this may give negative performance results if s.id = 1 not exists in the table...

SELECT 
  STRAIGHT_JOIN
  p.id, 
  p.name,
  p.sku, 
  p.type
FROM 
  xm_products p 
  LEFT JOIN xm_store_product sp ON p.id = sp.product_id 
  LEFT JOIN xm_store s ON sp.store_id = s.id 
WHERE 
  s.id = 1 
ORDER BY 
  p.type, 
  p.name asc 
LIMIT 
  20 OFFSET 0

Note there may be better options if you also share the create table statement output see MySQL slow query using filesort ... If s.id is always 1 record you can use the derived table approach note you may need add an new index