基本操作

数据库操作

基本操作

1.进入数据库

mysql -u root -p   
mysql -h localhost -u root -p database_name

2.列出数据库

show databases;

3.选择数据库:

use databases_name;

4.列出数据表:

show tables;

5.显示表格列的属性

show columns from table_name;   
describe table_name;

6.导出整个数据库:

mysqldump -u user_name -p database_name > /tmp/file_name

例如: mysqldump -u root -p test_db > d:/test_db.sql

7.导出一个表:

mysqldump -u user_name -p database_name table_name > /tmp/file_name

例如: mysqldump -u root -p test_db table1 > d:/table1.sql

8.导出一个数据库结构:

mysqldump -u user_name -p -d --add-drop-table database_name > file_name

例如: mysqldump -u root -p -d --add-drop-table test_db > test_db.sql

9.导入数据库:

source file_name;

mysql -u user_name -p database_name < file_name

例如:

source /tmp/bbs.sql;
mysql -u root -p bbs < "/tmp/bbs.sql"

10.将文本文件导入数据表中(excel与之相同)

load data infile "tables.txt" into table table_name;

例如:

load data infile "/tmp/bbs.txt" into table bbs;

load data infile "/tmp/bbs.xls" into table bbs;

11.将数据表导出为文本文件(excel与之相同)

select * into outfile "path_file_name" from table_name;

例如:

select * into outfile "/tmp/bbs.txt" from bbs;

select * into outfile "/tmp/bbs.xls" from bbs where id=1;

12.创建数据库时先判断数据库是否存在:

create database if not exists database_name;

例如: create database if not exists bbs;

13.创建数据库:

create database database_name;

例如: create database bbs;

14.删除数据库:

drop database database_name;

例如: drop database bbs;

15.创建数据表:

create table <table_name> ( <column 1 name> <col. 1 type> <col. 1 details>,<column 2 name> <col. 2 type> <col. 2 details>, ...);

例如: create table (id int not null auto_increment primary key,name char(16) not null default "jack",date_year date not null);

16.删除数据表中数据:

delete from table_name;

例如:

delete from bbs;

delete from bbs where id=2;

17.删除数据库中的数据表:

drop table table_name;

例如:

drop table test_db;
rm -f database_name/table_name.*  #(linux 环境下)

例如:

rm -rf bbs/accp.*

18.向数据库中添加数据:

insert into table_name set column_name1=value1,column_name2=value2;

例如: insert into bbs set name="jack",date_year="1993-10-01";

insert into table_name values (column1,column2,...);

例如: insert into bbs ("2","jack","1993-10-02")

insert into table_name (column_name1,column_name2,...) values (value1,value2);

例如: insert into bbs (name,data_year) values ("jack","1993-10-01");

19.查询数据表中的数据:

select * from table_name;

例如: select * from bbs where id=1;

20.修改数据表中的数据:

update table_name set col_name=new_value where id=1;

例如: update bbs set name="tom" where name="jack";

21.增加一个字段:

alter table table_name add column field_name datatype not null default "1";

例如: alter table bbs add column tel char(16) not null;

22.增加多个字段:(column可省略不写)

alter table table_name add column filed_name1 datatype,add column filed_name2 datatype;

例如: alter table bbs add column tel char(16) not null,add column address text;

23.删除一个字段:

alter table table_name drop field_name;

例如: alter table bbs drop tel;

24.修改字段的数据类型:

alter table table_name modify id int unsigned; //修改列id的类型为int unsigned   
alter table table_name change id sid int unsigned; //修改列id的名字为sid,而且把属性修改为int unsigned

25.修改一个字段的默认值:

alter table table_name modify column_name datatype not null default "";

例如: alter table test_db modify name char(16) default not null "yourname";

26.对表重新命名:

alter table table_name rename as new_table_name;

例如: alter table bbs rename as bbs_table;

rename table old_table_name to new_table_name;

例如: rename table test_db to accp;

27.从已经有的表中复制表的结构:

create table table2 select * from table1 where 1<>1;

例如: create table test_db select * from accp where 1<>1;

28.查询时间:

select now();

29.查询当前用户:

select user();

30.查询数据库版本:

select version();

31.创建索引:

alter table table1 add index ind_id(id);
create index ind_id on table1(id);
create unique index ind_id on table1(id);//建立唯一性索引

32.删除索引:

drop index idx_id on table1;
alter table table1 drop index ind_id;

33.联合字符或者多个列(将id与":"和列name和"="连接)

select concat(id,':',name,'=') from table;

34.limit(选出10到20条)

select * from bbs order by id limit 9,10;

35.增加一个管理员账号:

grant all on *.* to user@localhost identified by "password";

36.创建表是先判断表是否存在

create table if not exists students(……);

37.复制表:

create table table2 select * from table1;

例如: create table test_db select * from accp;

38.授于用户远程访问mysql的权限

grant all privileges on *.* to "root"@"%" identified by "password" with grant option;

或者是修改mysql数据库中的user表中的host字段

use mysql;   
select user,host from user;   
update user set host="%" where user="user_name";

39.查看当前状态

show status;

40.查看当前连接的用户

show processlist;

常用函数

ifnull

ifnull(oracle, nvl) #如果 oracle 字段为 null , 设置为 nvl

select ifnull(b,0) from c;

sum(...) over( ) 对所有行求和

sum(...) over( order by ... )

sum(...) over( partition by... order by ... )

数据占的百分: percent_rank() over(partition by init_date,interval_type order by drawdown)

时间比较

  • 转成时间戳

    • unix_timestamp(time1) 将字符型的时间,转成unix时间戳

  • 格式化

    • Date(time1) 将 datetime 转成 date

    • MONTH(time1)

    • Year(time1)

    • Month(time1)

  • 哪一天

    • DAYOFWEEK(time1) , WEEKDAY(time1) 0 = 星期一

    • DAYOFMONTH(time1)

    • DAYOFYEAR(time1)

    • DAYNAME(time1)MONTHNAME(time1)

  • 季度

    • QUARTER(time1)

  • 日期比较

    • > < =SELECT NOW() > '2019-09-21 15:59:59';

    • between ... and ...SELECT NOW() BETWEEN '2019-09-21 15:59:59' AND '2019-10-21 15:59:59' AS res;

    • DATEDIFF(time1,time2)SELECT DATEDIFF(NOW(),'2019-09-21 15:59:59') > 0;

  • 日期计算

    • DATE_SUB(CURDATE(),INTERVAL 3 MONTH)

    • DATE_ADD(CURDATE(),INTERVAL 3 MONTH)

细节专栏

DISTINCT

  1. 作用于单列

     select distinct name from A
  2. 作用于多列

     select distinct name, id from A

    实际上是根据name和id两个字段来去重的

  3. COUNT统计

     select count(distinct name) from A;

    count不能统计多个字段, 以下语句 错误 :

     select count(distinct name, id) from A;

    若使用嵌套查询,如下:

     select count(*) from (select distinct xing, name from B) AS M;
  4. distinct必须放在开头

     select id, distinct name from A;   --会提示错误,因为distinct必须放在开头

Last updated