Percona 5.6 InnoDB issue not using indexes correctly

女生的网名这么多〃 提交于 2019-12-24 17:00:56

问题


I just installed Percona 5.6 on my new CentOS 6.4 server. It's a fast machine 32 core xenon, 72GB ram, 8x SAS RAID 10 setup. So far so good

My old server is a bit less powerful, and was running MySQL 5.1 still. So this was quite an upgrade. But I'm having some issues with InnoDB, it is not using the indexes correctly on some tables it seems. Where on my old machine the same queries were running fine.

Both servers have the same database. I did a mysqldump on the old machine and imported it onto the new Percona 5.6 server. Indexes stayed the same. Both servers use the same my.cnf config settings.

Table items has indexes on: item_id, item_format, item_private and contains about 40 million rows. Table formats has index on: format_id and contains about 250 rows.

SELECT 
i.item_name, i.item_key, i.item_date, f.format_long
FROM
items i, formats f
WHERE 
i.item_format = f.format_id
AND
i.item_private = 0 
ORDER BY 
i.item_id DESC LIMIT 8

On my old server this query takes about 0.0003 seconds. On the new server it takes over 100 seconds.

Query with EXPLAIN on OLD server.

+----+-------------+-------+--------+---------------+---------+---------+----------------------+------+-------------+
| id | select_type | table |  type  | possible_keys |   key   | key_len |         ref          | rows |    Extra    |
+----+-------------+-------+--------+---------------+---------+---------+----------------------+------+-------------+
|  1 | SIMPLE      | i     | index  | item_format   | PRIMARY |       4 | NULL                 |    8 | Using where |
|  1 | SIMPLE      | f     | eq_ref | PRIMARY       | PRIMARY |       4 | dbname.i.item_format |    1 |             |
+----+-------------+-------+--------+---------------+---------+---------+----------------------+------+-------------+

Query with EXPLAIN on NEW [problem] server.

+----+-------------+-------+------+---------------+-------------+---------+--------------------+------+---------------------------------+
| id | select_type | table | type | possible_keys |     key     | key_len |        ref         | rows |              Extra              |
+----+-------------+-------+------+---------------+-------------+---------+--------------------+------+---------------------------------+
|  1 | SIMPLE      | f     | ALL  | PRIMARY       | NULL        | NULL    | NULL               |  219 | Using temporary; Using filesort |
|  1 | SIMPLE      | i     | ref  | item_format   | item_format | 4       | dbname.f.format_id | 3026 | Using where                     |
+----+-------------+-------+------+---------------+-------------+---------+--------------------+------+---------------------------------+

You can see that it's using temporary and filesort. This seems to be the reason for the slowness.

Any idea how I could resolve this issue?


回答1:


This sounds like: Bug #70617 Default persistent stats can cause unexpected long query times

For what it's worth, this is not a Percona bug, it's also present in MySQL 5.6 community edition.

There are three possible workarounds:

  1. Use STRAIGHT_JOIN to give a hint to the optimizer not to reorder table references.

    SELECT STRAIGHT_JOIN
      i.item_name, i.item_key, i.item_date, f.format_long
    FROM items i
    INNER JOIN formats f
      ON i.item_format = f.format_id
    WHERE i.item_private = 0 
    ORDER BY i.item_id DESC LIMIT 8
    

    I've rewritten your JOIN to use SQL-92 syntax, which I recommend.

  2. Disable the new InnoDB persistent stats feature, reverting to pre-5.6 behavior.

    In your my.cnf file:

    innodb_stats_persistent=0
    
  3. Refresh InnoDB optimizer stats manually after you do a significant change to the data (for example, after loading a mysqldump):

    ANALYZE TABLE items;
    ANALYZE TABLE formats;
    

PS: I work at Percona, and this bug was discovered by my colleague Justin Swanhart.



来源:https://stackoverflow.com/questions/19961013/percona-5-6-innodb-issue-not-using-indexes-correctly

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!