减少不必要的group by字段

我们可以看到没有去掉多余字段前 Sorting for group和Copying to tmp table占用了大部分io和cpu，而去掉后Copying to tmp table占

后台某维度统计语句
SELECT products_id, sku, px_id, sj_id, cat_path, COUNT(*) AS pv, COUNT(DISTINCT ip) AS ip_numbers, SUM(is_bounce) AS bounce_numbers, SUM(remain_time) AS remain_time
FROM dm_pv_records_search
WHERE 1 AND add_date >= '2014-02-26 10:00:00' AND products_id > 0
GROUP BY products_id,sku, px_id, sj_id, cat_path
order by ip_numbers desc limit 0,20;

每次有同事在后台查询这类数据时，都反应数据非常慢,偶尔不能出来数据。

explain
+----+-------------+----------------------+-------+----------------------+----------+---------+------+---------+----------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------------------+-------+----------------------+----------+---------+------+---------+----------------------------------------------+
| 1 | SIMPLE | dm_pv_records_search | range | products_id,add_date | add_date | 8 | NULL | 12831019 | Using where; Using temporary; Using filesort |
+----+-------------+----------------------+-------+----------------------+----------+---------+------+---------+----------------------------------------------+

这条语句索引使用是正确的，不过结果集及其大，而且需要对结果集group by操作。

观察该sql语句发现，sku,px_id,sj_id,cat_path四个字段为冗余字段，，都可以从其他表连表获得。

而改sql又存在order by .. limit m,n，可知道后期结果集只有20个，更加适合后期连接相关表获取其他必要的字段sku,px_id,sj_id,cat_path。
如此一转换，可以将group by中的sku,px_id,sj_id,cat_path去掉(其中sku和cat_path为字符串)，节省后期大结果集中排序的内存。
而group by和distinct优化的一种方式是，尽量减少不必要的字段，可以参考简朝阳写的《MySQL性能调优和架构设计》的第8.6小结，或者参考以下连接： 。

我们可以尝试将sql改成
SELECT products_id, COUNT(*) AS pv, COUNT(DISTINCT ip) AS ip_numbers, SUM(is_bounce) AS bounce_numbers, SUM(remain_time) AS remain_time
FROM dm_pv_records_search
WHERE 1 AND add_date >= '2014-02-26 10:00:00' AND products_id > 0
GROUP BY products_id
order by ip_numbers desc limit 0,20;

对比两次profiling,前者执行时间是58s,后者执行时间是5s.
前者profiling如下
+----------------------+-----------+----------+------------+--------------+---------------+
| Status | Duration | CPU_user | CPU_system | Block_ops_in | Block_ops_out |
+----------------------+-----------+----------+------------+--------------+---------------+
| starting | 0.000133 | 0.000000 | 0.000000 | 0 | 0 |
| checking permissions | 0.000014 | 0.000000 | 0.000000 | 0 | 0 |
| Opening tables | 0.000031 | 0.000000 | 0.000000 | 0 | 0 |
| System lock | 0.000018 | 0.000000 | 0.000000 | 0 | 0 |
| init | 0.000061 | 0.000000 | 0.000000 | 0 | 0 |
| optimizing | 0.000019 | 0.000000 | 0.000000 | 0 | 0 |
| statistics | 0.000073 | 0.000000 | 0.000000 | 0 | 0 |
| preparing | 0.000051 | 0.000000 | 0.000000 | 0 | 0 |
| Creating tmp table | 0.000051 | 0.000000 | 0.000000 | 0 | 0 |
| Sorting for group | 47.735389 | 4.614299 | 10.773362 | 3632 | 2811456 |
| executing | 0.000010 | 0.000000 | 0.000000 | 0 | 0 |
| Copying to tmp table | 11.566292 | 0.910861 | 0.586911 | 256 | 4408 |
| Sorting result | 0.030459 | 0.025996 | 0.004000 | 0 | 0 |
| Sending data | 0.000057 | 0.000000 | 0.000000 | 0 | 0 |
| end | 0.000005 | 0.000000 | 0.000000 | 0 | 0 |
| removing tmp table | 0.008139 | 0.000000 | 0.008998 | 0 | 0 |
| end | 0.000007 | 0.000000 | 0.000000 | 0 | 0 |
| query end | 0.000004 | 0.000000 | 0.000000 | 0 | 0 |
| closing tables | 0.000011 | 0.000000 | 0.000000 | 0 | 0 |
| freeing items | 0.000071 | 0.000000 | 0.000000 | 0 | 0 |
| removing tmp table | 0.000006 | 0.000000 | 0.000000 | 0 | 0 |
| freeing items | 0.000331 | 0.000000 | 0.000000 | 0 | 0 |
| logging slow query | 0.000006 | 0.000000 | 0.000000 | 0 | 0 |
| logging slow query | 0.000047 | 0.000000 | 0.000000 | 0 | 8 |
| cleaning up | 0.000005 | 0.000000 | 0.000000 | 0 | 0 |
+----------------------+-----------+----------+------------+--------------+---------------+