SQL数据库常用语句集锦
时间:2014-11-03 10:22 来源:linux.it.net.cn 作者:it
基础
1、说明:创建数据库
Create DATABASE database-name
use master
go
if exists(select *from sysdatabases where name ='Mydatabase')
drop database Mydatabase
go
create database Mydatabase
on primary
(
--主数据文件属性
NAME = 'DF_dat',
FILENAME = 'c:\Mydatabase\DF_dat.mdf',
SIZE = 5MB,
MAXSIZE = 500MB,
FILEGROWTH = 15%
)
log on
(
--日志文件属性
NAME = 'DF_log',
FILENAME = 'c:\Mydatabase\DF_dat.ldf',
SIZE = 1MB,
FILEGROWTH = 15%
)
一、 go
2、说明:删除数据库
drop database dbname
3、说明:备份sql server
--- 创建 备份数据的 device
USE master
EXEC sp_addumpdevice 'disk', 'testBack', 'c:\mssql7backup\MyNwind_1.dat'
--- 开始 备份
BACKUP DATABASE pubs TO testBack
4、说明:创建新表
create table tabname(col1 type1 [not null] [primary key],col2 type2 [not null],..)
根据已有的表创建新表:
A:create table tab_new like tab_old (使用旧表创建新表)
B:create table tab_new as select col1,col2… from tab_old definition only
5、说明:删除新表
drop table tabname
6、说明:增加一个列
Alter table tabname add column col type
注:列增加后将不能删除。DB2中列加上后数据类型也不能改变,唯一能改变的是增加varchar类型的长度。
7、说明:添加主键: Alter table tabname add primary key(col)
说明:删除主键: Alter table tabname drop primary key(col)
8、说明:创建索引:create [unique] index idxname on tabname(col….)
删除索引:drop index idxname
注:索引是不可更改的,想更改必须删除重新建。
9、说明:创建视图:create view viewname as select statement
删除视图:drop view viewname
10、说明:几个简单的基本的sql语句
选择:select * from table1 where 范围
插入:insert into table1(field1,field2) values(value1,value2)
删除:delete from table1 where 范围
更新:update table1 set field1=value1 where 范围
查找:select * from table1 where field1 like ’%value1%’ ---like的语法很精妙,查资料!
排序:select * from table1 order by field1,field2 [desc]
总数:select count as totalcount from table1
求和:select sum(field1) as sumvalue from table1
平均:select avg(field1) as avgvalue from table1
最大:select max(field1) as maxvalue from table1
最小:select min(field1) as minvalue from table1
11、说明:几个高级查询运算词
A: UNION 运算符 (集合操作并)
UNION 运算符通过组合其他两个结果表(例如 TABLE1 和 TABLE2)并消去表中任何重复行而派生出一个结果表。当 ALL 随 UNION 一起使用时(即 UNION ALL),不消除重复行。两种情况下,派生表的每一行不是来自 TABLE1 就是来自 TABLE2。
select sno from sc where cno=’1’ union select sno from sc where cno=’2’
B: EXCEPT 运算符 (集合操作差或minus)
EXCEPT 运算符通过包括所有在 TABLE1 中但不在 TABLE2 中的行并消除所有重复行而派生出一个结果表。当 ALL 随 EXCEPT 一起使用时 (EXCEPT ALL),不消除重复行。
select sno from sc where cno=’1’ except select sno from sc where cno=’2’ 或select sno from sc where cno=’1’ and sno not in (select sno from sc where cno=’2’)
C: INTERSECT 运算符(集合操作交)
INTERSECT 运算符通过只包括 TABLE1 和 TABLE2 中都有的行并消除所有重复行而派生出一个结果表。当 ALL 随 INTERSECT 一起使用时 (INTERSECT ALL),不消除重复行。
注:使用运算词的几个查询结果行必须是一致的。
select * from student where sdept=’cs’ intersect select * from student where sage<=’19’ 或 select * from student where sdept=’cs’ and sage<=’19’
12、说明:使用外连接
A、left outer join:
左外连接(左连接):结果集几包括连接表的匹配行,也包括左连接表的所有行。
SQL: select a.a, a.b, a.c, b.c, b.d, b.f from a LEFT OUT JOIN b ON a.a = b.c
B:right outer join:
右外连接(右连接):结果集既包括连接表的匹配连接行,也包括右连接表的所有行。
C:full outer join:
全外连接:不仅包括符号连接表的匹配行,还包括两个连接表中的所有记录。
二、提升
1、说明:复制表(只复制结构,源表名:a 新表名:b) (Access可用)
法一:select * into b from a where 1<>1
法二:select top 0 * into b from a
2、说明:拷贝表(拷贝数据,源表名:a 目标表名:b) (Access可用)
insert into b(a, b, c) select d,e,f from b;
3、说明:跨数据库之间表的拷贝(具体数据使用绝对路径) (Access可用)
insert into b(a, b, c) select d,e,f from b in ‘具体数据库’ where 条件
例子:..from b in '"&Server.MapPath(".")&"\data.mdb" &"' where..
4、说明:子查询(表名1:a 表名2:b)
select a,b,c from a where a IN (select d from b ) 或者: select a,b,c from a where a IN (1,2,3)
5、说明:显示文章、提交人和最后回复时间
select a.title,a.username,b.adddate from table a,(select max(adddate) adddate from table where table.title=a.title) b
6、说明:外连接查询(表名1:a 表名2:b)
select a.a, a.b, a.c, b.c, b.d, b.f from a LEFT OUT JOIN b ON a.a = b.c
7、说明:在线视图查询(表名1:a )
select * from (Select a,b,c FROM a) T where t.a > 1;
8、说明:between的用法,between限制查询数据范围时包括了边界值,not between不包括
select * from table1 where time between time1 and time2
select a,b,c, from table1 where a not between 数值1 and 数值2
9、说明:in 的使用方法
select * from table1 where a [not] in (‘值1’,’值2’,’值4’,’值6’)
10、说明:两张关联表,删除主表中已经在副表中没有的信息
delete from table1 where not exists ( select * from table2 where table1.field1=table2.field1 )
11、说明:四表联查问题:
select * from a left inner join b on a.a=b.b right inner join c on a.a=c.c inner join d on a.a=d.d where .....
12、说明:日程安排提前五分钟提醒
SQL: select * from 日程安排 where datediff('minute',f开始时间,getdate())>5
13、说明:一条sql 语句搞定数据库分页
select top 10 b.* from (select top 20 主键字段,排序字段 from 表名 order by 排序字段 desc) a,表名 b where b.主键字段 = a.主键字段 order by a.排序字段
14、说明:前10条记录
select top 10 * form table1 where 范围
15、说明:选择在每一组b值相同的数据中对应的a最大的记录的所有信息(类似这样的用法可以用于论坛每月排行榜,每月热销产品分析,按科目成绩排名,等等.)
select a,b,c from tablename ta where a=(select max(a) from tablename tb where tb.b=ta.b)
16、说明:包括所有在 TableA 中但不在 TableB和TableC 中的行并消除所有重复行而派生出一个结果表
(select a from tableA ) except (select a from tableB) except (select a from tableC)
17、说明:随机取出10条数据
select top 10 * from tablename order by newid()
18、说明:随机选择记录
select newid()
19、说明:删除重复记录
Delete from tablename where id not in (select max(id) from tablename group by col1,col2,...)
20、说明:列出数据库里所有的表名
select name from sysobjects where type='U'
21、说明:列出表里的所有的
select name from syscolumns where id=object_id('TableName')
22、说明:列示type、vender、pcs字段,以type字段排列,case可以方便地实现多重选择,类似select 中的case。
select type,sum(case vender when 'A' then pcs else 0 end),sum(case vender when 'C' then pcs else 0 end),sum(case vender when 'B' then pcs else 0 end) FROM tablename group by type
显示结果:
type vender pcs
电脑 A 1
电脑 A 1
光盘 B 2
光盘 A 2
手机 B 3
手机 C 3
23、说明:初始化表table1
TRUNCATE TABLE table1
24、说明:选择从10到15的记录
select top 5 * from (select top 15 * from table order by id asc) table_别名 order by id desc
三、技巧
1、1=1,1=2的使用,在SQL语句组合时用的较多
“where 1=1” 是表示选择全部 “where 1=2”全部不选,
如:
if @strWhere !=''
begin
set @strSQL = 'select count(*) as Total from [' + @tblName + '] where ' + @strWhere
end
else
begin
set @strSQL = 'select count(*) as Total from [' + @tblName + ']'
end
我们可以直接写成
set @strSQL = 'select count(*) as Total from [' + @tblName + '] where 1=1 安定 '+ @strWhere
2、收缩数据库
--重建索引
DBCC REINDEX
DBCC INDEXDEFRAG
--收缩数据和日志
DBCC SHRINKDB
DBCC SHRINKFILE
3、压缩数据库
dbcc shrinkdatabase(dbname)
4、转移数据库给新用户以已存在用户权限
exec sp_change_users_login 'update_one','newname','oldname'
go
5、检查备份集
RESTORE VERIFYONLY from disk='E:\dvbbs.bak'
6、修复数据库
Alter DATABASE [dvbbs] SET SINGLE_USER
GO
DBCC CHECKDB('dvbbs',repair_allow_data_loss) WITH TABLOCK
GO
Alter DATABASE [dvbbs] SET MULTI_USER
GO
7、日志清除
SET NOCOUNT ON
DECLARE @LogicalFileName sysname,
@MaxMinutes INT,
@NewSize INT
USE tablename -- 要操作的数据库名
Select @LogicalFileName = 'tablename_log', -- 日志文件名
@MaxMinutes = 10, -- Limit on time allowed to wrap log.
@NewSize = 1 -- 你想设定的日志文件的大小(M)
-- Setup / initialize
DECLARE @OriginalSize int
Select @OriginalSize = size
FROM sysfiles
Where name = @LogicalFileName
Select 'Original Size of ' + db_name() + ' LOG is ' +
CONVERT(VARCHAR(30),@OriginalSize) + ' 8K pages or ' +
CONVERT(VARCHAR(30),(@OriginalSize*8/1024)) + 'MB'
FROM sysfiles
Where name = @LogicalFileName
Create TABLE DummyTrans
(DummyColumn char (8000) not null)
DECLARE @Counter INT,
@StartTime DATETIME,
@TruncLog VARCHAR(255)
Select @StartTime = GETDATE(),
@TruncLog = 'BACKUP LOG ' + db_name() + ' WITH TRUNCATE_ONLY'
DBCC SHRINKFILE (@LogicalFileName, @NewSize)
EXEC (@TruncLog)
-- Wrap the log if necessary.
WHILE @MaxMinutes > DATEDIFF (mi, @StartTime, GETDATE()) -- time has not expired
AND @OriginalSize = (Select size FROM sysfiles Where name = @LogicalFileName)
AND (@OriginalSize * 8 /1024) > @NewSize
BEGIN -- Outer loop.
Select @Counter = 0
WHILE ((@Counter < @OriginalSize / 16) AND (@Counter < 50000))
BEGIN -- update
Insert DummyTrans VALUES ('Fill Log')
Delete DummyTrans
Select @Counter = @Counter + 1
END
EXEC (@TruncLog)
END
Select 'Final Size of ' + db_name() + ' LOG is ' +
CONVERT(VARCHAR(30),size) + ' 8K pages or ' +
CONVERT(VARCHAR(30),(size*8/1024)) + 'MB'
FROM sysfiles
Where name = @LogicalFileName
Drop TABLE DummyTrans
SET NOCOUNT OFF
8、说明:更改某个表
exec sp_changeobjectowner 'tablename','dbo'
9、存储更改全部表
Create PROCEDURE dbo.User_ChangeObjectOwnerBatch
@OldOwner as NVARCHAR(128),
@NewOwner as NVARCHAR(128)
AS
DECLARE @Name as NVARCHAR(128)
DECLARE @Owner as NVARCHAR(128)
DECLARE @OwnerName as NVARCHAR(128)
DECLARE curObject CURSOR FOR
select 'Name' = name,
'Owner' = user_name(uid)
from sysobjects
where user_name(uid)=@OldOwner
order by name
OPEN curObject
FETCH NEXT FROM curObject INTO @Name, @Owner
WHILE(@@FETCH_STATUS=0)
BEGIN
if @Owner=@OldOwner
begin
set @OwnerName = @OldOwner + '.' + rtrim(@Name)
exec sp_changeobjectowner @OwnerName, @NewOwner
end
-- select @name,@NewOwner,@OldOwner
FETCH NEXT FROM curObject INTO @Name, @Owner
END
close curObject
deallocate curObject
GO
10、SQL SERVER中直接循环写入数据
declare @i int
set @i=1
while @i<30
begin
insert into test (userid) values(@i)
set @i=@i+1
end
数据库自动编号归0
DBCC CHECKIDENT (表名, RESEED, 0)
删除所有数据
truncate table 表名
(责任编辑:IT)
基础 1、说明:创建数据库 Create DATABASE database-name use master go
if exists(select *from sysdatabases where name ='Mydatabase') drop database Mydatabase go
create database Mydatabase on primary (
) log on ( --日志文件属性 NAME = 'DF_log',
)
一、
数据库自动编号归0 DBCC CHECKIDENT (表名, RESEED, 0) 删除所有数据 truncate table 表名 (责任编辑:IT) |