本文已参与[新人创作礼]活动，一起开启掘金创作之路。
假设有一个业务场景，需要查询用户登录记录信息，其中表结构如下：
CREATE TABLE `tb` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`uid` int(11) NOT NULL,
`ip` varchar(16) NOT NULL,
`login_time` datetime,
PRIMARY KEY (`id`),
KEY (`uid`)
);
再来点测试数据：
INSERT INTO tb SELECT null, 1001, '19168.1', '2017-01-21 16:30:47';
INSERT INTO tb SELECT null, 1003, '19168.153', '2017-01-21 19:30:51';
INSERT INTO tb SELECT null, 1001, '19168.61', '2017-01-21 16:50:41';
INSERT INTO tb SELECT null, 1002, '19168.31', '2017-01-21 18:30:21';
INSERT INTO tb SELECT null, 1002, '19168.66', '2017-01-21 19:12:32';
INSERT INTO tb SELECT null, 1001, '19168.81', '2017-01-21 19:53:09';
INSERT INTO tb SELECT null, 1001, '19168.231', '2017-01-21 19:55:34';
表数据情况：
+----+------+---------------+---------------------+
| id | uid | ip | login_time |
+----+------+---------------+---------------------+
| 1 | 1001 | 19168.1 | 2017-01-21 16:30:47 |
| 2 | 1003 | 19168.153 | 2017-01-21 19:30:51 |
| 3 | 1001 | 19168.61 | 2017-01-21 16:50:41 |
| 4 | 1002 | 19168.31 | 2017-01-21 18:30:21 |
| 5 | 1002 | 19168.66 | 2017-01-21 19:12:32 |
| 6 | 1001 | 19168.81 | 2017-01-21 19:53:09 |
| 7 | 1001 | 19168.231 | 2017-01-21 19:55:34 |
+----+------+---------------+---------------------+
如果只需要针对用户查出其最后登录的时间，可以简单写出：
SELECT uid, max(login_time)
FROM tb
GROUP BY uid;
+------+---------------------+
| uid | max(login_time) |
+------+---------------------+
| 1001 | 2017-01-21 19:55:34 |
| 1002 | 2017-01-21 19:12:32 |
| 1003 | 2017-01-21 19:30:51 |
+------+---------------------+
若还需要查询用户最后登录时的其他信息，就不能用这种sql写了：
-- 错误写法
SELECT uid, ip, max(login_time)
FROM tb
GROUP BY uid;
-- 错误写法 这样的语句是非SQL标准的，虽然能够在MySQL数据库中执行成功，但返回的却是未知的
（如果sql_mode开启了only_full_group_by，则不会执行成功。）
可能ip字段会取uid分组前的第一个row的值，显然不是所需信息
写法1
写一个子查询：
SELECT a.uid, a.ip, a.login_time
FROM tb a
WHERE a.login_time in (
SELECT max(login_time)
FROM tb
GROUP BY uid); 写法2
再或者换一个写法：
SELECT a.uid, a.ip, a.login_time
FROM tb a
WHERE a.login_time = (
SELECT max(login_time)
FROM tb
WHERE a.uid = uid);
顺便测了一下
在6以前的版本中，写法②这条sql在大数据量的情况下，执行计划不理想，目测性能不佳。
在6及以后的版本中，写法②这条sql会快很多，执行计划也有了改变
+----+--------------------+-------+------+---------------+------+---------+------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+--------------------+-------+------+---------------+------+---------+------+------+-------------+
| 1 | PRIMARY | a | ALL | NULL | NULL | NULL | NULL | 7 | Using where |
| 2 | DEPENDENT SUBQUERY | tb | ALL | uid | NULL | NULL | NULL | 7 | Using where |
+----+--------------------+-------+------+---------------+------+---------+------+------+-------------+
+----+--------------------+-------+------+---------------+------+---------+------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+--------------------+-------+------+---------------+------+---------+------------+------+-------------+
| 1 | PRIMARY | a | ALL | NULL | NULL | NULL | NULL | 7 | Using where |
| 2 | DEPENDENT SUBQUERY | tb | ref | uid | uid | 4 | test.a.uid | 1 | NULL |
+----+--------------------+-------+------+---------------+------+---------+------------+------+-------------+
写法3
直接改成join性能会更加好：
SELECT a.uid, a.ip, a.login_time
FROM (SELECT uid, max(login_time) login_time
FROM tb
GROUP BY uid
) b JOIN tb a ON a.uid = b.uid AND a.login_time = b.login_time;
当然，结果都相同：
+------+---------------+---------------------+
| uid | ip | login_time |
+------+---------------+---------------------+
| 1003 | 19168.153 | 2017-01-21 19:30:51 |
| 1002 | 19168.66 | 2017-01-21 19:12:32 |
| 1001 | 19168.231 | 2017-01-21 19:55:34 |
+------+---------------+---------------------+
注：如果要分组取最小值直接改对应函数和符号就行了。
