不管是程序BUG，还是业务变更，重复数据这个老生常谈的问题，总是会出现。以下是我在MariaDB或是MySQL下处理的一些经验。在SQL Server中，使用窗口函数是很容易实现的。不过听说MySQL 8.0和MariaDB 10.2以上均支持窗口函数了。等有机会再来测试使用窗口函数来删除重复记录。
背景
表t_record中的数据fromUserId, toUserId两个字段组合作为唯一的标识，删除重复记录，只留下最大id（或最新时间）的记录。id为自增无重复的主键。
表t_record的id作为自增的主键。
表t_record大概有6万多的数据。以下测试均在资源很差的主机上，t_record没有在使用的情况下的结果。
方法1
查询重复的记录
SELECT fromUserId, toUserId, count(*)
FROM t_record as tr
GROUP BY fromUserId, toUserId
HAVING count(*) > 1;把重复记录的两个字段放到临时表_tmp1中
CREATE TABLE _tmp1
SELECT fromUserId, toUserId
FROM t_record as tr
GROUP BY fromUserId, toUserId
HAVING count(*) > 1;把应该删除的id查询出来，放到临时表_tmp2中
CREATE TABLE _tmp2
SELECT id
FROM t_record as a
WHERE (a.fromUserId, a.toUserId) in (
SELECT fromUserId, toUserId from _tmp1
)
and a.id not in (
SELECT MAX(id)
FROM t_record as tr
GROUP BY fromUserId, toUserId
HAVING count(*) > 1
);删除原表的记录
DELETE from t_record
where id in (
SELECT id from _tmp2
);通过以前3个步骤，没有删除数据大概需要23秒左右。
方法2
如果表中没有主键，也没有可以标识唯一记录的字段。只能是把原表的数据分表后，插入到另一张临时表，删除原表数据，把临时表的数据导回来。
这种方法也适用合于有主键或有唯一标识的表，但操作过程中会影响在线的业务，需要中断业务。否则可能会造成数据丢失或数据不一致。
数据量大的表，导两次数据，过程会很慢，同时也需要注意硬盘空间是否足够。
方法3
测试mysql不支持以下这种delete语法来删除数据。改为select id 存到临时表，查询非常慢。
DELETE a
FROM table_nam a
WHERE EXISTS (SELECT 1 FROM table_nam b
WHERE b.userid = a.userid AND b.CreateDate > a.CreateDate);方法4
在mariadb 10.1.19下测试，60多秒。
这个方法简单，只需要一条语句，速度还行。
DELETE
from t_record
where id not in (
select maxid from
(select max(id) as maxid from t_record
group by fromUserId,toUserId
) b
);现对方法4的进行改造，再测试下。11.5秒。快好几倍了。
CREATE OR REPLACE TABLE _tmp3
SELECT id
FROM t_record
WHERE id NOT IN (
SELECT maxid FROM
(SELECT max(id) AS maxid FROM t_record
GROUP BY fromUserId,toUserId
) b
);
DELETE FROM t_record
WHERE id IN (
SELECT id from _tmp3
);对临时表创建主键，再测试下。1.2秒！WOW！！！
CREATE OR REPLACE TABLE _tmp3 (id INT NOT NULL PRIMARY KEY);
INSERT INTO _tmp3 (id)
SELECT id
FROM t_record
WHERE id NOT IN (
SELECT maxid FROM
(SELECT max(id) AS maxid FROM t_record
GROUP BY fromUserId,toUserId
) b
);
DELETE FROM t_record
WHERE id IN (
SELECT id from _tmp3
);还以不能再快呢?改IN子句为JOIN，再测试下。1秒！
CREATE OR REPLACE TABLE _tmp3 (id INT NOT NULL PRIMARY KEY);
INSERT INTO _tmp3 (id)
SELECT id
FROM t_record
WHERE id NOT IN (
SELECT maxid FROM
(SELECT max(id) AS maxid FROM t_record
GROUP BY fromUserId,toUserId
) b
);
DELETE a FROM t_record as a INNER JOIN _tmp3 as b on b.id = a.id;难道删除的那个语句的执行计划是不同的吗?由于mysql只支持select的执行查询，所以把删除的语句修改为查询语句。
root@localhost [db1]EXPLAIN SELECT id FROM t_record WHERE id IN ( SELECT id from _tmp3 );
+------+-------------+-------------------+--------+---------------+---------+---------+--------------------+------+--------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+-------------+-------------------+--------+---------------+---------+---------+--------------------+------+--------------------------+
| 1 | PRIMARY | _tmp3 | index | PRIMARY | PRIMARY | 4 | NULL | 452 | Using index |
| 1 | PRIMARY | t_record | eq_ref | PRIMARY | PRIMARY | 8 | testdb._tmp3.id | 1 | Using where; Using index |
+------+-------------+-------------------+--------+---------------+---------+---------+--------------------+------+--------------------------+
2 rows in set (0.00 sec)root@localhost [db1]EXPLAIN SELECT a.id FROM t_record as a INNER JOIN _tmp3 as b on b.id = a.id;
+------+-------------+-------+--------+---------------+---------+---------+----------------+------+--------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+-------------+-------+--------+---------------+---------+---------+----------------+------+--------------------------+
| 1 | SIMPLE | b | index | PRIMARY | PRIMARY | 4 | NULL | 452 | Using index |
| 1 | SIMPLE | a | eq_ref | PRIMARY | PRIMARY | 8 | testdb.b.id | 1 | Using where; Using index |
+------+-------------+-------+--------+---------------+---------+---------+----------------+------+--------------------------+
2 rows in set (0.01 sec)对比执行计划，是一样的。但测试多次，用JOIN方式速度还是快些。为什么呢?