来源: SqlServer性能优化 查询和索引优化(十二) – 孙丽媛 – 博客园
查询优化的过程:
查询优化:
功能:分析语句后最终生成执行计划
分析:获取操作语句参数
索引选择
Join算法选择
创建测试的表:
1
|
select * into EmployeeOp from AdventureWorks2014.HumanResources.Employee |
建立非聚集索引:
1
|
create nonclustered index nc_employee_vacationhours on employeeop(vacationhours) |
执行语句:
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>40 --table scan>10% |
执行语句:
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99 --nonclustered index |
查询结果集的数据范围影响对索引的选择。
两个查询条件:
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>40 and SickLeaveHours>60--scan |
SQLServer 的查询结果集会认为用哪个列查询的结果少,就选择哪个。在去and 的第二个结果,最终返回结果集。
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99 and SickLeaveHours>60--nonclustered index nc_employee_vacationhours |
单独选择:
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where SickLeaveHours>60--table scan |
创建非聚集索引:
1
|
create nonclustered index nc_employee_sickleavehours on EmployeeOp(SickLeaveHours) |
执行:
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where SickLeaveHours>60--table scan |
执行:
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where SickLeaveHours>88--nc_employee_sickleavehours |
执行:
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99 and SickLeaveHours>88--nonclustered index nc_employee_vacationhours |
在两列上做一个索引:
1
|
create nonclustered index nc_employee_vacationsickleavehours on EmployeeOp(VacationHours,SickLeaveHours) |
执行语句:(使用了符合索引)
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99 and SickLeaveHours>88-- nc_employee_vacationsickleavehours |
执行:(随机)
1
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99<br>--nc_employee_vacationhours nc_employee_vacationsickleavehours |
执行:
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where SickLeaveHours>88 --nc_employee_sickleavehours |
执行:
1
|
select * from EmployeeOp where SickLeaveHours>88 --nc_employee_sickleavehours |
创建聚集索引:
1
|
create clustered index c_Employee_BusinessEntityID on EmployeeOp(BusinessEntityID) |
执行:
1
|
select * from EmployeeOp where SickLeaveHours>88 --nc_employee_sickleavehours key连 c_ID聚集索引 |
建立include索引:
1
2
|
create nonclustered index nc_employee_vacationsickleavehoursinclude on EmployeeOp(VacationHours, SickLeaveHours) include(LoginID,JobTitle) |
执行:
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>99 and SickLeaveHours>88 --nc_employee_vacationsickleavehoursinclude |
执行:(采用覆盖索引)
1
2
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>60 and SickLeaveHours>10--nc_employee_vacationsickleavehoursinclude--0.0048<br><br> select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp where VacationHours>60<br>--nc_employee_vacationsickleavehoursinclude |
执行:(指定使用的索引)
1
2
3
|
select LoginID,JobTitle,VacationHours,SickLeaveHours from EmployeeOp with(index=0) where VacationHours>60 and SickLeaveHours>10 |
索引的优化:
1
|
select * from EmployeeOp<br>--创建非聚集索引<br>create nonclustered index nc_EmployeeOp on employeeop (VacationHours,SickLeaveHours) include (LoginID,JobTitle)<br><br>create nonclustered index nc_EmployeeOp_Vacation on employeeop(VacationHours)<br>include(LoginID,JobTitle)<br><br>--创建聚集索引<br> set statistics io on <br>create clustered index c_Employee_id on employeeop(BusinessEntityID) --7,9,9<br> set statistics io off |
总结:先创建聚集索引在创非聚集索引
聚集索引键宽与窄:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
|
create table temptable(c1 int not null ,c2 int ) declare @c int set @c=0 while @c<50000 begin insert temptable values(@c,@c) set @c=@c+1 end create clustered index c_temptable_c1 on temptable(c1) set statistics io on select * from temptable where c1<=25000 --0.07 set statistics io off |
创建Guid的列:
1
2
3
4
5
6
7
8
9
10
11
12
|
create table temptable(c1 uniqueidentifier,c2 int ) declare @c int set @c=0 while @c<50000 begin insert temptable values(newid(),@c) set @c=@c+1 end create clustered index c_temptable_c1 on temptable(c1) set statistics io on select * from temptable where c1<= 'D144242D-BFA3-4A8C-8DCE-C35A880E8BBE' --0.11 set statistics io off |
索引设计建议:
1.where子句与连接条件列(where子句后面的列建立非聚集索引,有多列查询做成组合索引,并用inclued的方式把经常访问的列信息给包含到非聚集索引的页集,查询用到链接时(join):join的条件列做到非聚集索引中)
2.使用窄索引:索引列少、索引列数据类型空间少
1.减少IO数量
2.提高缓存效率
3.减少数据存储的空间
4.动态管理视图: sys.dm_db_index_physical_stats
选择性能高的列应该创建索引,如果有多列筛选,并尽量放置经常筛选的列和低密度的列到组合索引前面
int类型上创建索引与char 型上创建索引
1
2
3
4
5
6
7
|
create nonclustered index nc_employee_vacationsickleavehours on employeeop(vacationhours, sickleavehours) include(LoginID,JobTitle) create nonclustered index nc_employee_sickvacationleavehours on employeeop(sickleavehours,vacationhours) include(LoginID,JobTitle) select LoginID,JobTitle from EmployeeOp where VacationHours>40 and SickLeaveHours>90 -- nc_sickleavevacation |
1
|
select loginid,jobtitle from EmployeeOp where VacationHours>99 and SickLeaveHours>10--nc_vacationsickleave |
总结:会自动进行筛选与and的顺序无关。(谁的选择性度高)
非聚集索引:RID指针指向堆得行标识符或聚集索引的键值
如果有非聚集索引,一定要创建一个聚集索引
先创建聚集索引,在创建非聚集索引
保持聚集索引窄:提高非聚集索引性能,提高聚集索引性能
使用聚集索引的时机:
1.Group by列
2.Order by 列
3.没有针对某个筛选条件的非聚集索引
不合适使用聚集索引:
1.索引列值频繁跟新:频繁跟新非聚集索引降低性能
2.并发的大量的插入
如果非聚集索引需要书签查找,则建议通过聚集索引查找
建议创建覆盖索引
不适合使用非聚集索引:
1.需要获取大量的行
2.需要获取大量的字段
交叉索引:针对筛选条件分别建立非聚集索引,在查询时,获得两个子集的索引交叉,解决覆盖索引非常宽的问题
建议使用过滤索引:针对查询必然需要筛选掉的条件做成索引的过滤条件
1
|
create nonclustered index nc_employee_sickvacationleavehours on employeeop(sickleavehours,vacationhours) include (LoginID,JobTitle) where salariedFlag=1 |
恰当使用索引视图使连接与聚合实物化,平衡查询性能提升与维护视图性能开销
复合索引每列可以不按照相同排序规则
可以在计算列上创建索引,建议使用持久化的计算列
指定并行度CPU个数、制定联机索引
经常使用数据库引擎优化顾问
尽量减少书签查找
查询优化统计方面的应用:
查询优化器对索引的选择依赖于统计
统计被自动创建和更新,也可以设置异步更新统计
通过Profiler跟踪统计事件
过时统计造成查询优化器无法选择最优的执行计划
自动创建统计也会在非索引列上创建统计
跟新自动统计:
SQL完成情况:
开启跟踪:
验证事件:
创建跟踪统计的表:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
|
create table StatisticsTB(c1 int identity(1,1),c2 int ) declare @n int set @n=0 while @n<5000 begin insert StatisticsTB values(@n) set @n=@n+1 end create nonclustered index nc_StatisticsTB_t2 on StatisticsTB(c2) declare @n int set @n=5001 while @n<50000 begin insert StatisticsTB values(@n) set @n=@n+1 end select * from StatisticsTB where c2<10--index select * from StatisticsTB where c2>10--Scan |
自动统计功能出现故障:
1
2
3
4
5
6
7
8
9
|
--自动统计出现故障后 declare @n int set @n=50001 while @n<130000 begin insert StatisticsTB values(@n) set @n=@n+1 end |
本来是表扫描的就弄成索引。
1
|
select * from StatisticsTB where c2>4990--index |
查看统计信息:
1
2
3
4
|
--查看统计信息 dbcc show_statistics( 'Employeeop' ,nc_Employee_vacation)--密度:0.01 dbcc show_statistics( 'Employeeop' ,nc_Employee_vacationsickleave)--密度:0.009 |
更新统计:
1
2
3
4
|
--更新统计 use HRDB go Sp_Updatestats |
–创建统计:
1
|
create statistics s_Employee_c2 on StatisticsTB(c2) |
在非索引列上创建统计:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
|
create table t1(c1 int identity(1,1),c2 int ) insert t1 values(2) declare @count int set @count=0 while @count<1000 begin insert t1 values(1) set @count=@count+1 end create table t2(c1 int identity(1,1),c2 int ) insert t2 values(1) declare @count int set @count=0 while @count<1000 begin insert t1 values(2) set @count=@count+1 end |
关闭统计的情况:
1
2
|
select t.c1,t.c2,tt.c1,tt.c2 from t1 as t inner join t2 as tt on t.c2=tt.c2--0.045 |
删除重新创建表:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
|
drop table t1 drop table t2 create table t1(c1 int identity(1,1),c2 int ) insert t1 values(2) declare @count int set @count=0 while @count<1000 begin insert t1 values(1) set @count=@count+1 end create table t2(c1 int identity(1,1),c2 int ) insert t2 values(1) declare @count int set @count=0 while @count<1000 begin insert t1 values(2) set @count=@count+1 end select t.c1,t.c2,tt.c1,tt.c2 from t1 as t inner join t2 as tt on t.c2=tt.c2--0.045 |
统计建议:
查看索引是否有碎片:
1
2
3
|
--查看索引是否有碎片 select * from sys.dm_db_index_physical_stats(db_id( 'HRDB' ),object_id( 'EmployeeOp' ), null , null , 'Detailed' ) |
做碎片的整理:
1
|
--对页面进行重排:<br>alter index nc_Employee_Vacation on EmployeeOp Reorganize |
重建索引:
1
|
alter index nc_Employee_Vacation on employeeop rebuild with(fillfactor=40) |
填充因子的方式重建索引:
1
2
|
--指定填充因子重建索引 create nonclustered index nc_Employee_Vacation on Employeeop (VacationHours) with(fillfactor=40,drop_existing= on ) |
查询优化器Join的选择:
1.嵌套循环的join NestedLoop Join
2.合并的join Merge Join算法
1.链接表记录数都比较多,并且针对连接列进行了物理排序
2.Inner表的行有范围约束
3.Hash join算法
对Join算法的选择:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
|
create table parenttb(c1 int ,name varchar(500)) declare @c int set @c=0 while @c<10 begin insert parenttb values(@c,GETDATE()) set @c=@c+1 end go create table subtb(c1 int ,cardid uniqueidentifier) declare @c int set @c=0 while @c<250 begin insert subtb values(@c,NEWID()) set @c=@c+1 end |
执行语句:
1
|
select p.name,s.cardid from parenttb as p inner join subtb as s on p.c1=s.c1 --hash --0.29 io: |
手工指定:
1
2
3
4
|
set statistics io on select p.name,s.cardid from parenttb as p inner loop join subtb as s on p.c1=s.c1--nested loop --0.21 io:p 1 s 20 set statistics io off |
多添加一些记录:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
|
create table parenttb(c1 int ,name varchar(500)) declare @c int set @c=0 while @c<1000 begin insert parenttb values(@c,getdate()) set @c=@c+1 end go create table subtb(c1 int ,cardid uniqueidentifier) declare @c int set @c=0 while @c<25000 begin insert subtb values(@c,NEWID()) set @c=@c+1 end |
执行语句:
1
2
3
4
5
6
7
|
set statistics io on select p.name,s.cardid from parenttb as p inner join subtb as s on p.c1=s.c1--hash --0.5 io:p 7 s 140 set statistics io off set statistics io on select p.name,s.cardid from parenttb as p inner loop join subtb as s on p.c1=s.c1--loop --64 io:p 7 s 560 set statistics io off |
创建唯一的聚集索引:
1
2
3
|
--创建唯一的聚集索引 create unique clustered index c_parent_c1 on Parenttb(c1) create unique clustered index c_sub_c1 on Subtb(c1) |
执行:
1
2
3
|
set statistics io on select p.name,s.cardid from parenttb as p inner join subtb as s on p.c1=s.c1--Merge --0.16 io:p 6 s 7 set statistics io off |