MySQL索引失效之隱式轉換

GeaoZhang發表於2022-01-08

常見索引失效：

1. 條件索引欄位"不乾淨"：函式操作、運算操作

2. 隱式型別轉換：字串轉數值；其他型別轉換

3. 隱式字元編碼轉換：按字元編碼資料長度大的方向轉換，避免資料擷取

一、常見索引失效場景

root@test 10:50 > show create table t_num\G
*************************** 1. row ***************************
       Table: t_num
Create Table: CREATE TABLE `t_num` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `c1` int(11) NOT NULL,
  `c2` varchar(11) NOT NULL,
  PRIMARY KEY (`id`),
  KEY `ix_c1` (`c1`)
) ENGINE=InnoDB AUTO_INCREMENT=6 DEFAULT CHARSET=utf8mb4

root@test 10:51 > select * from t_num;
+----+----+----+
| id | c1 | c2 |
+----+----+----+
|  1 | -2 | -2 |
|  2 | -1 | -1 |
|  3 |  0 |  0 |
|  4 |  1 |  1 |
|  5 |  2 |  2 |
+----+----+----+

# 在c1欄位上加上索引
root@test 10:52 > alter table t_num add index ix_c1(c1);

# 標準使用情況下，索引有效
root@test 10:55 > explain select * from t_num where c1 = -1;
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+
| id | select_type | table | partitions | type | possible_keys | key   | key_len | ref   | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+
|  1 | SIMPLE      | t_num | NULL       | ref  | ix_c1         | ix_c1 | 4       | const |    1 |   100.00 | NULL  |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+

1、條件欄位函式操作

# 在where中c1上加上abs()絕對值函式，可以看到type=ALL，全表掃描，在Server層進行絕對值處理後進行比較
root@test 10:58 > explain select * from t_num where abs(c1) = 1;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | t_num | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    5 |   100.00 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+

如上，對索引欄位做函式操作，即where條件列上不乾淨時，可能會破壞索引值的有序性（按照c1的值有序組織索引樹），因此優化器就決定放棄走索引樹搜尋功能。

但是，條件欄位函式操作下，也並非完全的走全表掃描，優化器並非完全的放棄該欄位索引。

# 選擇查詢的資料，只有id和c1欄位，可以看到type=index，使用到了ix_c1索引
root@test 10:59 > explain select id,c1 from t_num where abs(c1) = 1;
+----+-------------+-------+------------+-------+---------------+-------+---------+------+------+----------+--------------------------+
| id | select_type | table | partitions | type  | possible_keys | key   | key_len | ref  | rows | filtered | Extra                    |
+----+-------------+-------+------------+-------+---------------+-------+---------+------+------+----------+--------------------------+
|  1 | SIMPLE      | t_num | NULL       | index | NULL          | ix_c1 | 4       | NULL |    5 |   100.00 | Using where; Using index |
+----+-------------+-------+------------+-------+---------------+-------+---------+------+------+----------+--------------------------+

如上，由於ix_c1索引樹是根節點c1和葉子節點id構造的，雖然因為c1上的函式操作導致放棄索引定位，但優化器可以選擇遍歷該索引樹，使用覆蓋索引（Using index），無需回表，將所需的id和c1資料返回Server層後進行後續的abs()和where過濾。

2、條件欄位運算操作

# where條件裡，對c1進行運算操作
root@test 11:03 > explain select * from t_num where c1 + 1 = 2;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | t_num | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    5 |   100.00 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+

如上，雖然“+1”的操作並沒有破壞c1索引的有序性，但優化器仍然沒有使用該索引快速定位。因此，等號左邊，注意優化掉索引欄位上的運算操作。

3、隱式型別轉換

# 在c2欄位上加上索引
root@test 12:30 > alter table t_num add index ix_c2(c2);

# 標準使用情況下（注：c2是varchar型別的），索引有效
root@test 12:30 > explain select * from t_num where c2 = "2";
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+
| id | select_type | table | partitions | type | possible_keys | key   | key_len | ref   | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+
|  1 | SIMPLE      | t_num | NULL       | ref  | ix_c2         | ix_c2 | 42      | const |    1 |   100.00 | NULL  |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-------+

# 去掉等號右邊值的引號，即字串和數值進行比較，索引失效
root@test 12:30 > explain select * from t_num where c2 = 2;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | t_num | NULL       | ALL  | ix_c2         | NULL | NULL    | NULL |    5 |    20.00 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+

如上，c2欄位是varchar型別，是字串和數值的比較，此時，MySQL是將字串轉換成數字，即此處的c2被CAST(c2 AS signed int)，這就相當於對條件欄位做了函式操作，優化器放棄走樹索引定位。

4、隱式字元編碼轉換

# 建立一個t_cou表，表結構基本和前面的t_num相同，唯一不同的設定是表字符集CHARSET=utf8
root@test 14:02 > show create table t_cou\G
*************************** 1. row ***************************
       Table: t_cou
Create Table: CREATE TABLE `t_cou` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `c1` int(11) NOT NULL,
  `c2` varchar(10) NOT NULL,
  PRIMARY KEY (`id`),
  KEY `ix_c1` (`c1`),
  KEY `ix_c2` (`c2`)
) ENGINE=InnoDB AUTO_INCREMENT=1 DEFAULT CHARSET=utf8

root@test 14:02 > insert into t_cou select * from t_num;

# join表，t_num和t_cou通過c2欄位進行關聯查詢
root@test 14:03 > select n.* from t_num n
    -> join t_cou c
    -> on n.c2 = c.c2
    -> where n.c1 = 1;
+----+----+----+
| id | c1 | c2 |
+----+----+----+
|  4 |  1 | 1  |
+----+----+----+

root@test 14:23 > explain select n.* from t_num n join t_cou c  on n.c2 = c.c2 where c.c1 = 1;
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-----------------------+
| id | select_type | table | partitions | type | possible_keys | key   | key_len | ref   | rows | filtered | Extra                 |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-----------------------+
|  1 | SIMPLE      | c     | NULL       | ref  | ix_c1         | ix_c1 | 4       | const |    1 |   100.00 | NULL                  |
|  1 | SIMPLE      | n     | NULL       | ref  | ix_c2         | ix_c2 | 42      | func  |    1 |   100.00 | Using index condition |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+-----------------------+
# 執行計劃分析:
# 1.操作的c表，使用了ix_c1定位到一行資料
# 2.從c表定位到的行資料，拿到c2欄位去操作n表，t_cou稱為驅動表，t_num稱為被驅動表
# 3.ref=func說明使用了函式操作，指的是n.c2=CONVERT(c.c2 USING utf8mb4)
# 4.同時Using index condition，ix_c2讀取查詢時，使用被下推的條件過濾，滿足條件的才回表

root@test 14:23 > explain select n.* from t_num n join t_cou c  on n.c2 = c.c2 where n.c1 = 1;
+----+-------------+-------+------------+-------+---------------+-------+---------+-------+------+----------+-----------------------------------------------------------------+
| id | select_type | table | partitions | type  | possible_keys | key   | key_len | ref   | rows | filtered | Extra                                                           |
+----+-------------+-------+------------+-------+---------------+-------+---------+-------+------+----------+-----------------------------------------------------------------+
|  1 | SIMPLE      | n     | NULL       | ref   | ix_c1,ix_c2   | ix_c1 | 4       | const |    1 |   100.00 | NULL                                                            |
|  1 | SIMPLE      | c     | NULL       | index | NULL          | ix_c2 | 32      | NULL  |    5 |   100.00 | Using where; Using index; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+-------+---------------+-------+---------+-------+------+----------+-----------------------------------------------------------------+
# 執行計劃分析:
# 1.操作的n表，使用了ix_c1定位到一行資料
# 2.從n表定位到的行資料，拿到c2欄位去操作c表，t_num稱為驅動表，t_cou稱為被驅動表
# 3.同樣的n.c2=c.c2，會將c.c2的字符集進行轉換，即被驅動表的索引欄位上加函式操作，索引失效
# 4.BNL，表join時，驅動表資料讀入join buffer，被驅動表連線欄位無索引則全表掃，每取一行和join buffer資料對比判斷，作為結果集返回

如上，分別對t_num、 t_cou作為驅動表和被驅動表的執行計劃分析，總結：

utf8mb4和utf8兩種不同字符集（編碼）型別的字串在做比較時，MySQL會先把 utf8 字串轉成 utf8mb4 字符集，再做比較。為什麼？字符集 utf8mb4 是 utf8 的超集，再做隱式自動型別轉換時，為了避免資料在轉換過程中由於截斷導致資料錯誤，會“按資料長度增加的方向”進行轉換。
表連線過程中，被驅動表的索引欄位上加函式操作，會導致對被驅動表做全表掃描。

優化手法：

修改統一join欄位的字符集
對驅動表下手，將連線欄位的字符集轉換成被驅動表連線欄位的字符集

root@test 18:09 > explain select n.* from t_num n join t_cou c  on convert(n.c2 using utf8) = c.c2 where n.c1 = 1;
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+--------------------------+
| id | select_type | table | partitions | type | possible_keys | key   | key_len | ref   | rows | filtered | Extra                    |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+--------------------------+
|  1 | SIMPLE      | n     | NULL       | ref  | ix_c1         | ix_c1 | 4       | const |    1 |   100.00 | NULL                     |
|  1 | SIMPLE      | c     | NULL       | ref  | ix_c2         | ix_c2 | 32      | func  |    1 |   100.00 | Using where; Using index |
+----+-------------+-------+------------+------+---------------+-------+---------+-------+------+----------+--------------------------+

二、型別轉換

1、字串轉整型

# 字元開頭的一律為0
root@test 18:44 > select convert("abc", unsigned integer);
+----------------------------------+
| convert("abc", unsigned integer) |
+----------------------------------+
|                                0 |
+----------------------------------+
# 'abc' = 0是成立的，因此查詢時等號右邊使用對應的型別很重要，0匹配出欄位字元開頭資料，'0'只匹配0
root@test 18:44 > select 'abc' = 0;
+-----------+
| 'abc' = 0 |
+-----------+
|         1 |
+-----------+

# 數字開頭的，直接擷取到第一個不是字元的位置
root@test 18:45 > select convert("123abc", unsigned integer);
+-------------------------------------+
| convert("123abc", unsigned integer) |
+-------------------------------------+
|                                 123 |
+-------------------------------------+

2、時間型別轉換

root@test 19:11 > show create table time_demo\G
*************************** 1. row ***************************
       Table: time_demo
Create Table: CREATE TABLE `time_demo` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `c1` datetime DEFAULT NULL,
  `c2` date DEFAULT NULL,
  PRIMARY KEY (`id`),
  KEY `ix_c1` (`c1`)
) ENGINE=InnoDB AUTO_INCREMENT=5 DEFAULT CHARSET=utf8mb4

root@test 19:15 > select count(*) from time_demo;
+----------+
| count(*) |
+----------+
|       11 |
+----------+

root@test 19:16 > select * from time_demo limit 4;
+----+---------------------+------------+
| id | c1                  | c2         |
+----+---------------------+------------+
|  1 | 2022-01-08 00:01:01 | 2022-01-08 |
|  2 | 2022-01-06 23:01:01 | 2022-01-06 |
|  3 | 2022-01-06 00:00:00 | 2022-01-06 |
|  4 | 2022-01-08 00:00:00 | 2022-01-08 |
+----+---------------------+------------+

# 1.date轉datetime：末尾追加 00:00:00
root@test 19:11 > select * from time_demo where c1 between "2022-01-06" and "2022-01-08";
+----+---------------------+------------+
| id | c1                  | c2         |
+----+---------------------+------------+
|  2 | 2022-01-06 23:01:01 | 2022-01-06 |
|  3 | 2022-01-06 00:00:00 | 2022-01-06 |
|  4 | 2022-01-08 00:00:00 | 2022-01-08 |
+----+---------------------+------------+
# 結果分析：c1是datetime型別，進行比較時，between and中的date型別會轉換成datetime
# 即 where c1 between "2022-01-06 00:00:00" and "2022-01-08 00:00:00";
# 同 where c1 >= "2022-01-06 00:00:00" and c1 <= "2022-01-08 00:00:00";
root@test 19:42 > explain select * from time_demo where c1 between "2022-01-06" and "2022-01-08";
+----+-------------+-----------+------------+-------+---------------+-------+---------+------+------+----------+-----------------------+
| id | select_type | table     | partitions | type  | possible_keys | key   | key_len | ref  | rows | filtered | Extra                 |
+----+-------------+-----------+------------+-------+---------------+-------+---------+------+------+----------+-----------------------+
|  1 | SIMPLE      | time_demo | NULL       | range | ix_c1         | ix_c1 | 6       | NULL |    3 |   100.00 | Using index condition |
+----+-------------+-----------+------------+-------+---------------+-------+---------+------+------+----------+-----------------------+
# 格式化date轉datetime
root@test 19:23 > select date_format("2022-01-08","%Y-%m-%d %H:%i:%s");
+-----------------------------------------------+
| date_format("2022-01-08","%Y-%m-%d %H:%i:%s") |
+-----------------------------------------------+
| 2022-01-06 00:00:00                           |
+-----------------------------------------------+

# 2.datetime轉date：直接擷取date部分
root@test 19:47 > select date(c1) from time_demo limit 1;
+------------+
| date(c1)   |
+------------+
| 2022-01-06 |
+------------+

# 3.date轉time，沒有意義，直接變成 00:00:00

索引失效系列——隱式型別轉換
2011-04-24
索引型別
Oracle隱式型別轉換導致索引失效
2011-01-21
Oracle型別索引
淺談MySql整型索引和字串索引失效或隱式轉換問題汊叄
2022-03-01
MySql索引字串
【隱式轉換】注意隱式轉換將導致索引無法使用
2010-04-23
索引
mysql隱式轉換問題
2021-03-11
MySql
MySQL 隱式型別轉換
2017-12-18
MySql型別
索引關鍵字的隱式轉換分析
2012-01-16
索引
MySQL和Oracle中的隱式轉換
2017-09-18
MySqlOracle
談談 MySQL 隱式型別轉換
2017-12-14
MySql型別
高效的SQL（隱式轉換導致不走索引）
2012-11-28
SQL索引
scala中隱式轉換之隱式轉換呼叫類中本不存在的方法
2017-12-07
js顯式轉換和隱式轉換
2018-12-26
JS
Oracle like、不等於、隱式轉換走索引與不走索引情況
2020-01-19
Oracle索引
一次oracle sql調優的經歷（隱士轉換導致索引失效）
2015-11-10
OracleSQL索引
javascript 隱式轉換
2020-02-21
JavaScript
sql隱式轉換
2020-11-11
SQL
Oracle 隱式轉換
2014-02-27
Oracle
java隱式轉換
2024-03-05
Java
MySql ORDER BY索引是否失效
2024-11-20
MySql索引
Scala - 隱式轉換和隱式引數
2016-08-10
Scala隱式轉換與隱式引數
2016-02-24
oracle執行計劃------未走索引，隱式轉換的坑
2021-03-21
Oracle索引
Scala Essentials: 隱式轉換
2019-02-16
mysql索引失效的情況
2023-09-25
MySql索引
探索MySQL的InnoDB索引失效
2024-10-06
MySql索引
從Java的型別轉換看MySQL和Oracle中的隱式轉換(二)
2015-09-24
Java型別MySqlOracle
JavaScript隱式型別轉換
2019-02-16
JavaScript型別
【C++】禁止隱式轉換
2024-03-19
C++
MySQL隱式轉化整理
2016-05-06
MySql
索引ROWID轉換函式
2010-06-25
索引函式
MySQL學習筆記：索引失效
2020-09-22
MySql筆記索引
Java資料型別的顯式轉換和隱式轉換
2020-09-23
Java資料型別
評“MySQL 隱式轉換引起的執行結果錯誤”
2019-07-02
MySql
[] == ![]，走進==隱式轉換的世界
2019-05-25
JS隱式轉換--寬鬆相等（==）
2018-11-18
JS
筆記：隱式轉換規則
2018-01-10
筆記
C# 隱式型別轉換（轉載）
2017-06-27
C#型別
MySQL索引失效行鎖變表鎖
2020-10-03
MySql索引