全文索引对索引选择的干扰
mysql全文索引使用得到，对性能提升有一定帮助；但是，若使用不得到，将会是异常灾难；mysql全文索引对整个优化器的索引选择都有干扰。看我生产环境下优化过的一条sql
SELECT DISTINCT pc.products_id, pd.products_name,p.products_date_added,pso.products_id
FROM products_to_categories AS pc
LEFT JOIN products_description AS pd ON pd.products_id=pc.products_id
LEFT JOIN products AS p ON p.products_id=pd.products_id
LEFT JOIN specials AS sps ON sps.products_id=p.products_id
LEFT JOIN temp_products_7days_orders_amount AS 7days ON 7days.products_id=pc.products_id
LEFT JOIN products_realtime_quantity AS prq ON prq.sku_or_poa = p.products_model
LEFT JOIN products_stockout AS pso ON pso.products_id=pd.products_id
WHERE p.products_status=1 AND (prq.msg != 'Temporary out stock.' OR ISNULL(prq.msg))
AND pc.categories_id IN ( 153,323,1055,1241,1431)
AND MATCH(pd.products_name) AGAINST('*iphone*' IN BOOLEAN MODE) AND MATCH(pd.products_name)
AGAINST('*c*' IN BOOLEAN MODE) ORDER BY 7days.orders_sum DESC
这条语句执行非常慢，经常出现卡住情况,有时候发现执行需要几分钟,而结果才几条，该语句也为涉及到大结果运算，各种连表条件上上都有索引。唯一特殊的地方就是pd.products_name为全文索引，而且执行的过程中pc.categories_id优先级高于pd.products_name全文索引，导致使用了pc.categories_id索引。按理来讲，这样也没有多大关系。但是explain后发现了问题
+----+-------------+-------+----------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-------+----------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+ | 1 | SIMPLE | pc | range | PRIMARY,categories_id | categories_id | 4 | NULL | 307 | Using where; Using temporary; Using filesort | | 1 | SIMPLE | pd | fulltext | PRIMARY,products_name | products_name | 0 | | 1 | Using where | | 1 | SIMPLE | p | eq_ref | PRIMARY | PRIMARY | 4 | banggood.pd.products_id | 1 | Using where | | 1 | SIMPLE | sps | ref | products_id | products_id | 4 | banggood.pd.products_id | 16 | Using index | | 1 | SIMPLE | 7days | ref | PRIMARY | PRIMARY | 4 | banggood.p.products_id | 1032 | | | 1 | SIMPLE | prq | ref | ix_prg_sku_or_poa | ix_prg_sku_or_poa | 152 | banggood.p.products_model | 10 | Using where | | 1 | SIMPLE | pso | eq_ref | PRIMARY | PRIMARY | 4 | banggood.pd.products_id | 1 | Using index | +----+-------------+-------+----------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+
我们发现驱动表示pc表，使用了categories_id索引，可能优化器优先选择了它，但是再看pd表，
按理来讲，这个时候pd表应该使用products_id索引，也就是这个表的primary key，但是优化器却选择了products_name全文索引，坑爹了！
profiling这条语句，执行时间为2分钟以上
+-------------------------+------------+-----------+------------+--------------+---------------+ | Status | Duration | CPU_user | CPU_system | Block_ops_in | Block_ops_out | +-------------------------+------------+-----------+------------+--------------+---------------+ | starting | 0.000415 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000011 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000004 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000002 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000003 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000056 | 0.001000 | 0.000000 | 0 | 0 | | checking permissions | 0.000006 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000009 | 0.000000 | 0.000000 | 0 | 0 | | Opening tables | 0.000225 | 0.000000 | 0.000000 | 0 | 0 | | System lock | 0.000029 | 0.000000 | 0.000000 | 0 | 0 | | init | 0.000138 | 0.000000 | 0.000000 | 0 | 0 | | optimizing | 0.000046 | 0.000000 | 0.000000 | 0 | 0 | | statistics | 0.001115 | 0.001000 | 0.000000 | 0 | 0 | | preparing | 0.001246 | 0.002000 | 0.000000 | 0 | 0 | | FULLTEXT initialization | 0.000088 | 0.000000 | 0.000000 | 0 | 0 | | Creating tmp table | 0.000057 | 0.000000 | 0.000000 | 0 | 0 | | executing | 0.000005 | 0.000000 | 0.000000 | 0 | 0 | | Copying to tmp table | 120.430834 | 81.227651 | 38.749110 | 1112 | 0 | | Sorting result | 0.000058 | 0.000000 | 0.000000 | 0 | 0 | | Sending data | 0.000026 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | | removing tmp table | 0.000015 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000041 | 0.001000 | 0.000000 | 0 | 0 | | query end | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | | closing tables | 0.000023 | 0.000000 | 0.000000 | 0 | 0 | | freeing items | 0.008546 | 0.000000 | 0.007999 | 0 | 0 | | logging slow query | 0.000008 | 0.000000 | 0.000000 | 0 | 0 | | logging slow query | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | | cleaning up | 0.000008 | 0.000000 | 0.000000 | 0 | 0 | +-------------------------+------------+-----------+------------+--------------+---------------+
看到Copying to tmp table占据了大量的cpu运算。
看来，mysql优化器太弱了，又要我们强制使用索引了！force index(primary) ,强制使用pd表的主键
改成如下
SELECT DISTINCT pc.products_id, pd.products_name,p.products_date_added,pso.products_id
FROM products_to_categories AS pc
LEFT JOIN products_description AS pd FORCE INDEX(PRIMARY) ON pd.products_id=pc.products_id
LEFT JOIN products AS p ON p.products_id=pd.products_id
LEFT JOIN specials AS sps ON sps.products_id=p.products_id
LEFT JOIN temp_products_7days_orders_amount AS 7days ON 7days.products_id=pc.products_id
LEFT JOIN products_realtime_quantity AS prq ON prq.sku_or_poa = p.products_model
LEFT JOIN products_stockout AS pso ON pso.products_id=pd.products_id
WHERE p.products_status=1 AND (prq.msg != 'Temporary out stock.' OR ISNULL(prq.msg))
AND pc.categories_id IN ( 153,323,1055,1241,1431)
AND MATCH(pd.products_name) AGAINST('*iphone*' IN BOOLEAN MODE) AND MATCH(pd.products_name)
AGAINST('*c*' IN BOOLEAN MODE) ORDER BY 7days.orders_sum DESC
explain如下
+----+-------------+-------+--------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-------+--------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+ | 1 | SIMPLE | pc | range | PRIMARY,categories_id | categories_id | 4 | NULL | 307 | Using where; Using temporary; Using filesort | | 1 | SIMPLE | p | eq_ref | PRIMARY | PRIMARY | 4 | banggood.pc.products_id | 1 | Using where | | 1 | SIMPLE | pd | ref | PRIMARY | PRIMARY | 4 | banggood.p.products_id | 764 | Using where | | 1 | SIMPLE | sps | ref | products_id | products_id | 4 | banggood.pc.products_id | 16 | Using index | | 1 | SIMPLE | 7days | ref | PRIMARY | PRIMARY | 4 | banggood.p.products_id | 1032 | | | 1 | SIMPLE | prq | ref | ix_prg_sku_or_poa | ix_prg_sku_or_poa | 152 | banggood.p.products_model | 10 | Using where | | 1 | SIMPLE | pso | eq_ref | PRIMARY | PRIMARY | 4 | banggood.pd.products_id | 1 | Using index | +----+-------------+-------+--------+-----------------------+-------------------+---------+---------------------------+------+----------------------------------------------+
pd表索引已经使用正确；
profiling如下
+-------------------------+----------+----------+------------+--------------+---------------+ | Status | Duration | CPU_user | CPU_system | Block_ops_in | Block_ops_out | +-------------------------+----------+----------+------------+--------------+---------------+ | starting | 0.000191 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000010 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000003 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000002 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000002 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000003 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000002 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | | Opening tables | 0.000184 | 0.000000 | 0.000000 | 0 | 8 | | System lock | 0.000021 | 0.000000 | 0.000000 | 0 | 0 | | init | 0.000074 | 0.000000 | 0.000000 | 0 | 0 | | optimizing | 0.000034 | 0.000000 | 0.000000 | 0 | 0 | | statistics | 0.000191 | 0.000999 | 0.000000 | 0 | 0 | | preparing | 0.000052 | 0.000000 | 0.000000 | 0 | 0 | | FULLTEXT initialization | 0.000019 | 0.000000 | 0.000000 | 0 | 0 | | Creating tmp table | 0.000041 | 0.000000 | 0.000000 | 0 | 0 | | executing | 0.000005 | 0.000000 | 0.000000 | 0 | 0 | | Copying to tmp table | 0.006914 | 0.000000 | 0.005999 | 0 | 0 | | Sorting result | 0.000021 | 0.000000 | 0.000000 | 0 | 0 | | Sending data | 0.000021 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000004 | 0.000000 | 0.000000 | 0 | 0 | | removing tmp table | 0.000008 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000004 | 0.000000 | 0.000000 | 0 | 0 | | query end | 0.000004 | 0.000000 | 0.000000 | 0 | 0 | | closing tables | 0.000020 | 0.000000 | 0.000000 | 0 | 0 | | freeing items | 0.000100 | 0.000000 | 0.001000 | 0 | 0 | | logging slow query | 0.000006 | 0.000000 | 0.000000 | 0 | 0 | | cleaning up | 0.000006 | 0.000000 | 0.000000 | 0 | 0 | +-------------------------+----------+----------+------------+--------------+---------------+
执行时间已经下降为0.01以下！可以看到Creating tmp table性能损耗选项已经不见！
注意事项：
使用全文索引时，如果全文索引不能为驱动索引，那么会导致mysql优化器在选择其他索引为驱动索引的同时，对全文索引所在表的索引选择出现错误的判断，个人认为这个问题可以视为一个bug。