前期数据准备
通过程序往数据库插入50w数据
-
数据表:
CREATE TABLE `users` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`time_date` datetime NOT NULL,
`time_timestamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`time_long` bigint(20) NOT NULL,
PRIMARY KEY (`id`),
KEY `time_long` (`time_long`),
KEY `time_timestamp` (`time_timestamp`),
KEY `time_date` (`time_date`)
) ENGINE=InnoDB AUTO_INCREMENT=500003 DEFAULT CHARSET=latin1其中time_long、time_timestamp、time_date为同一时间的不同存储格式
-
实体类users
@Builder
@Data
public class Users {
/**
* 自增唯一id
* */
private Long id;
/**
* date类型的时间
* */
private Date timeDate;
/**
* timestamp类型的时间
* */
private Timestamp timeTimestamp;
/**
* long类型的时间
* */
private long timeLong;
} -
dao层接口
@Mapper
public interface UsersMapper {
@Insert("insert into users(time_date, time_timestamp, time_long) value(#{timeDate}, #{timeTimestamp}, #{timeLong})")
@Options(useGeneratedKeys = true,keyProperty = "id",keyColumn = "id")
int saveUsers(Users users);
} -
测试类往数据库插入数据
public class UsersMapperTest extends BaseTest {
@Resource
private UsersMapper usersMapper;
@Test
public void test() {
for (int i = 0; i < 500000; i++) {
long time = System.currentTimeMillis();
usersMapper.saveUsers(Users.builder().timeDate(new Date(time)).timeLong(time).timeTimestamp(new Timestamp(time)).build());
}
}
}
sql查询速率测试
-
通过datetime类型查询:
select count(*) from users where time_date >="2018-10-21 23:32:44" and time_date <="2018-10-21 23:41:22"
耗时:0.171
-
通过timestamp类型查询
select count(*) from users where time_timestamp >= "2018-10-21 23:32:44" and time_timestamp <="2018-10-21 23:41:22"
耗时:0.351
-
通过bigint类型查询
select count(*) from users where time_long >=1540135964091 and time_long <=1540136482372
耗时:0.130s
-
结论 在InnoDB存储引擎下,通过时间范围查找,性能bigint > datetime > timestamp
sql分组速率测试
使用bigint 进行分组会每条数据进行一个分组,如果将bigint做一个转化在去分组就没有比较的意义了,转化也是需要时间的
-
通过datetime类型分组:
select time_date, count(*) from users group by time_date
耗时:0.176s
-
通过timestamp类型分组:
select time_timestamp, count(*) from users group by time_timestamp
耗时:0.173s
-
结论 在InnoDB存储引擎下,通过时间分组,性能timestamp > datetime,但是相差不大
sql排序速率测试
-
通过datetime类型排序:
select * from users order by time_date
耗时:1.038s
-
通过timestamp类型排序
select * from users order by time_timestamp
耗时:0.933s
-
通过bigint类型排序
select * from users order by time_long
耗时:0.775s
-
结论 在InnoDB存储引擎下,通过时间排序,性能bigint > timestamp > datetime
小结
如果需要对时间字段进行操作(如通过时间范围查找或者排序等),推荐使用bigint,如果时间字段不需要进行任何操作,推荐使用timestamp,使用4个字节保存比较节省空间,但是只能记录到2038年记录的时间有限