背景

本篇文章来源于今天客户问的一个问题。

问题大概意思是：我正在从 Oracle 迁移到 MySQL，数据已经转换为单纯的 INSERT 语句。由于语句很多，每次导入的时候不知道怎么定位到错误的语句。 如果 INSERT 语句少也就罢了，我可以手工看，不过 INSERT 语句很多，我怎么定位到是哪些语句出错了，我好改正呢？总不能每次遇到的错误的时候改一下，再重新运行继续改正吧？有没有简单点的方法。

其实 MySQL 自身就有错误诊断区域，如果能好好利用，则事半功倍。

演示

下面我来简单说下怎么使用错误诊断区域。

比如说我要插入的表结构为 n3，保存错误信息的日志表为 error_log 两个表结构如下：

-- tables definition.

[ytt]>create table n3 (id int not null, id2 int generated always as ((mod(id,10))));

Query OK, 0 rows affected (0.04 sec)

[ytt]>create table error_log (sqltext text, error_no int unsigned, error_message text);

Query OK, 0 rows affected (0.04 sec)

假设插入的语句，为了演示，我这里仅仅简单写了 8 条语句。

-- statements body.

set @a1 = "INSERT INTO n3 (id) VALUES(100)";

set @a2 = "INSERT INTO n3 (id) VALUES('test')";

set @a3 = "INSERT INTO n3 (id) VALUES('test123')";

set @a4 = "INSERT INTO n3 (id) VALUES('123test')";

set @a5 = "INSERT INTO n3 (id) VALUES(200)";

set @a6 = "INSERT INTO n3 (id) VALUES(500)";

set @a7 = "INSERT INTO n3 (id) VALUES(null)";

set @a8 = "INSERT INTO n3 (id) VALUES(10000000000000)";

MySQL 的错误代码很多，不过总体归为三类：

• sqlwarning SQLSTATE 代码开始为 '01'

• not found SQLSTATE 代码开始为 '02'

• sqlexception SQLSTATE 代码开始非 '00','01','02' 的所有错误代码。

为了简单方便，我们写这些代码到存储过程里。以下为示例存储过程。

-- stored routines body.

drop procedure if exists sp_insert_simple;

delimiter ||

create procedure sp_insert_simple()

l1:begin

DECLARE i,j TINYINT DEFAULT 1; -- loop counter.

DECLARE v_errcount,v_errno INT DEFAULT 0; -- error count and error number.

DECLARE v_msg TEXT; -- error details.

declare v_sql json; -- store statements list.

declare v_sql_keys varchar(100); -- array index.

declare v_sql_length int unsigned; -- array length.

-- Handler declare.

DECLARE CONTINUE HANDLER FOR SQLEXCEPTION,SQLWARNING,NOT FOUND -- exception in mysql routines.

l2:BEGIN

get stacked diagnostics v_errcount = number;

set j = 1;

WHILE j <= v_errcount

do

GET stacked DIAGNOSTICS CONDITION j v_errno = MYSQL_ERRNO, v_msg = MESSAGE_TEXT;

-- record error messages into table.

INSERT INTO error_log(sqltext,error_no,error_message) VALUES (@sqltext, v_errno,v_msg);

SET j = j + 1;

END WHILE;

end;

-- sample statements array.

set v_sql = '{

"a1": "INSERT INTO n3 (id) VALUES(100)",

"a2": "INSERT INTO n3 (id) VALUES(''test'')",

"a3": "INSERT INTO n3 (id) VALUES(''test123'')",

"a4": "INSERT INTO n3 (id) VALUES(''123test'')",

"a5": "INSERT INTO n3 (id) VALUES(200)",

"a6": "INSERT INTO n3 (id) VALUES(500)",

"a7": "INSERT INTO n3 (id) VALUES(null)",

"a8": "INSERT INTO n3 (id) VALUES(10000000000000)"

}';

set i = 1;

set v_sql_length = json_length(v_sql);

while i <=v_sql_length do

set v_sql_keys = concat('$.a',i);

set @sqltext = replace(json_extract(v_sql,v_sql_keys),'"','');

prepare s1 from @sqltext;

execute s1;

set i = i + 1;

end while;

drop prepare s1;

-- invoke procedure.

-- call sp_insert_simple;

end;

delimiter ;

我们来调用这个存储过程看下结果。

[(none)]>use ytt

Reading table information for completion of table and column names

You can turn off this feature to get a quicker startup with -A

Database changed

[ytt]>call sp_insert_simple;

Query OK, 0 rows affected (0.05 sec)

表N3的结果。

[ytt]>select * from n3;

+-----+------+

| id | id2 |

+-----+------+

| 100 | 0 |

| 200 | 0 |

| 500 | 0 |

+-----+------+

3 rows in set (0.00 sec)

错误日志记录了所有错误的语句。

[ytt]>select * from error_log;

+--------------------------------------------+----------+-------------------------------------------------------------+

| sqltext | error_no | error_message |

+--------------------------------------------+----------+-------------------------------------------------------------+

| INSERT INTO n3 (id) VALUES('test') | 1366 | Incorrect integer value: 'test' for column 'id' at row 1 |

| INSERT INTO n3 (id) VALUES('test123') | 1366 | Incorrect integer value: 'test123' for column 'id' at row 1 |

| INSERT INTO n3 (id) VALUES('123test') | 1265 | Data truncated for column 'id' at row 1 |

| INSERT INTO n3 (id) VALUES(null) | 1048 | Column 'id' cannot be null |

| INSERT INTO n3 (id) VALUES(10000000000000) | 1264 | Out of range value for column 'id' at row 1 |

+--------------------------------------------+----------+-------------------------------------------------------------+

5 rows in set (0.00 sec)

其实这个问题如果用 Python 或 PHP 等外部语言来说，将会更简单，思路差不多。