前幾天統計一個sql，是一個人提交了多少工單，順便做了相關sql優化。數據大概2000多w。
select CustName,count(1) c from WorkOrder where CreateDate>'2016-5-1' and CreateDate<'2017-1-1'
group by CustName having c>100
order by c desc;
為了實驗最少受其他因素干擾，將生產庫的200多w數據導出來，用測試服務器進行測試。
導出來的數據是一個堆表，沒有主鍵，沒有索引。
mysql> show index from WorkOrder; 查詢index方法1 Empty set (0.00 sec) mysql> show keys from WorkOrder; 查詢index方法2 Empty set (0.00 sec)
1.堆表的情況
這時候就在這時候，用執行計划分析下語句。
mysql> explain select CustName,count(1) c from WorkOrder where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+------+---------------+------+---------+------+---------+----------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+------+---------------+------+---------+------+---------+----------------------------------------------+ | 1 | SIMPLE | WorkOrder | ALL | NULL | NULL | NULL | NULL | 2528727 | Using where; Using temporary; Using filesort | +----+-------------+-----------+------+---------------+------+---------+------+---------+----------------------------------------------+ 1 row in set
select_type的值為SIMPLE，表示簡單的select查詢，不使用union或子查詢。
type的值為ALL，表示要對表進行表掃描。
possible_keys 表示能使用哪個索引找到行記錄。
key 表示Mysql決定使用的索引（鍵）。
key_len 表示Mysql決定使用索引的長度。
ref 表示使用哪個列和key一起從表中選擇行。
rows 表示Mysql認為它執行查詢時必須檢查的行數。
extra 表示查詢的詳情信息，用到where，臨時表，排序。
執行下該語句三次，發現執行了16.30 sec、16.34 sec、16.24 sec。
2.有索引的情況
建了四個索引，分別以custname，CreateDate建兩個單列索引，另外兩個是聯合索引，只是最左邊列不一樣。
alter table WorkOrder add index ix_name(custname) alter table WorkOrder add index ix_date(CreateDate) alter table WorkOrder add index ix_namedate(custname,CreateDate) alter table WorkOrder add index ix_datename(CreateDate,custname)
mysql> show keys from WorkOrder; +-----------+------------+-------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment | +-----------+------------+-------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ | WorkOrder | 1 | ix_name | 1 | CustName | A | 1264363 | NULL | NULL | YES | BTREE | | | | WorkOrder | 1 | ix_date | 1 | CreateDate | A | 2528727 | NULL | NULL | | BTREE | | | | WorkOrder | 1 | ix_namedate | 1 | CustName | A | 1264363 | NULL | NULL | YES | BTREE | | | | WorkOrder | 1 | ix_namedate | 2 | CreateDate | A | 2528727 | NULL | NULL | | BTREE | | | | WorkOrder | 1 | ix_datename | 1 | CreateDate | A | 2528727 | NULL | NULL | | BTREE | | | | WorkOrder | 1 | ix_datename | 2 | CustName | A | 2528727 | NULL | NULL | YES | BTREE | | | +-----------+------------+-------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ 6 rows in set (0.00 sec)
之后，用執行計划分析下sql查詢語句。
mysql> explain select CustName,count(1) c from WorkOrder where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+-------+-----------------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+-------+-----------------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ | 1 | SIMPLE | WorkOrder | range | ix_name,ix_date,ix_namedate,ix_datename | ix_datename | 4 | NULL | 824372 | Using where; Using index; Using temporary; Using filesort | +----+-------------+-----------+-------+-----------------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ 1 row in set (0.01 sec)
從執行計划可以看出，Mysql從四個索引中選取了ix_datename這個索引，type為range表示索引范圍掃描。rows的數量值是沒堆表的1/3。
執行語句三次，時間是 8.64 sec、8.61sec、8.55 sec。
我建了三個索引，那么我想用下另外三個索引怎么辦？
這里可以用force index()，這個指令可以指定本次查詢強制使用哪個索引，因為Mysql優化器的選擇並不是最優的索引。
mysql> explain select CustName,count(1) c from WorkOrder force index(ix_namedate)
where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+---------+-----------------------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+---------+-----------------------------------------------------------+ | 1 | SIMPLE | WorkOrder | index | ix_name,ix_namedate,ix_datename | ix_namedate | 307 | NULL | 2528727 | Using where; Using index; Using temporary; Using filesort | +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+---------+-----------------------------------------------------------+
選用另一個聯合索引 ix_namedate，這次type變為index，可以這樣理解，根據索引的順序進行全表掃描，比ALL效率要高些，rows的值和堆表的值差不多。
執行語句三次，時間是 7.84 sec、7.92 sec、7.84 sec。
mysql> explain select CustName,count(1) c from WorkOrder force index(ix_name) where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+-------+---------------------------------+---------+---------+------+---------+----------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+-------+---------------------------------+---------+---------+------+---------+----------------------------------------------+ | 1 | SIMPLE | WorkOrder | index | ix_name,ix_namedate,ix_datename | ix_name | 303 | NULL | 2528727 | Using where; Using temporary; Using filesort | +----+-------------+-----------+-------+---------------------------------+---------+---------+------+---------+----------------------------------------------+ 1 row in set
選用另一個聯合索引 ix_name，這次type是index，可以這樣理解，根據索引的順序進行全表掃描，比ALL效率要高些，rows的值和堆表的值差不多。
執行語句三次，時間是 1 min 28.17 sec、1 min 27.64 sec、1 min 27.58 sec。
mysql> explain select CustName,count(1) c from WorkOrder force index(ix_date) where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+-------+-----------------------------------------+---------+---------+------+--------+-------------------------------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+-------+-----------------------------------------+---------+---------+------+--------+-------------------------------------------------------------------+ | 1 | SIMPLE | WorkOrder | range | ix_name,ix_date,ix_namedate,ix_datename | ix_date | 4 | NULL | 921062 | Using index condition; Using MRR; Using temporary; Using filesort | +----+-------------+-----------+-------+-----------------------------------------+---------+---------+------+--------+-------------------------------------------------------------------+
選用另一個聯合索引 ix_date，這次type是range，表示索引范圍掃描，rows的值是堆表的1/3多些 。
執行語句三次，時間是 9.55 sec、9.52 sec、9.39 sec。
假如我不想用索引了怎么辦？
可以使用ignore index()，這個指令可以強制Mysql在查詢時，不使用某索引。
mysql> explain select CustName,count(1) c from WorkOrder ignore index(ix_date) where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ | 1 | SIMPLE | WorkOrder | range | ix_name,ix_namedate,ix_datename | ix_datename | 4 | NULL | 824372 | Using where; Using index; Using temporary; Using filesort | +----+-------------+-----------+-------+---------------------------------+-------------+---------+------+--------+-----------------------------------------------------------+ mysql> explain select CustName,count(1) c from WorkOrder ignore index(ix_date,ix_name,ix_namedate,ix_datename) where CreateDate>'2016-5-1' and CreateDate<'2017-1-1' group by CustName having c>100 order by c desc; +----+-------------+-----------+------+---------------------------------+------+---------+------+---------+----------------------------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-----------+------+---------------------------------+------+---------+------+---------+----------------------------------------------+ | 1 | SIMPLE | WorkOrder | ALL | ix_name,ix_namedate,ix_datename | NULL | NULL | NULL | 2528727 | Using where; Using temporary; Using filesort | +----+-------------+-----------+------+---------------------------------+------+---------+------+---------+----------------------------------------------+
上面第一個強制不使用ix_date索引，那么就Mysql就從剩下的三個索引中，選取他認為是最優的索引。第二個時將四個索引都不使用，那么Mysql就進行全表掃描了。
總結：
1.Mysql的語句優化，沒有絕對的正確，explain也只是給出個大致的方向，例如 key_len值小的，rows小的按理說，時間應該最短，效率最高。但是，實驗中時間最少的卻不是那個值最小的。
2. 優化還需根據實際數據情況，例如，假如我where選取的時間范圍變化，或者說CustName的分布有些變化，可能跟剛才的實驗，又會產生一定偏差。
3. 同樣我還實驗了，當給表加上主鍵時，整體的查詢時間會縮短些。
------------------附相關index命令--------------
刪除主鍵
ALTER TABLE WorkOrder MODIFY id int(11)； --1.先刪除auto_increment
ALTER TABLE WorkOrder DROP PRIMARY KEY; --2.再刪除主鍵 ALTER TABLE WorkOrder DROP index ix_datename;--刪除索引