Biegral Blog

一:sqlserver 执行计划介绍

　    sqlserver 执行计是在sqlser manager studio 工具中打开，是检查一条sql执行效率的工具。建议配合SET STATISTICS IO ON等语句来一起使用，执行计划是从右向左看，耗时高的一般显示在右边,我们知道，sqlserver 查询数据库的方式为:

　　1:表扫描(table scan) 查询速度最慢.

　　2:聚集索引扫描(Clustered Index Scan),按聚集索引逐行进行查询，效率比表扫描高，但速度还是慢.

　　3:索引扫描（index scan）效率比聚集索引快，根据索引滤出部分数据在进行逐行检查。

　　4;索引查找(index seek) 效率比索引扫描还要快，根据索引定位记录所在位置再取出记录.

　　5:聚集索引查找(Clustered Index Seek) 效率最快，直接根据聚集索引获取记录。

当发现某个查询比较慢时，可以首先检查哪些操作的成本比较高，再看看那些操作在查找记录时， 是不是【Table Scan】或者【Clustered Index Scan】，如果确实和这二种操作类型有关，则要考虑增加索引来解决了,sqlser 索引有两种，聚集索引和非聚集索引，聚集索引是一张表只能有一个，比如id,非聚集索引可以有多个，聚集索引是顺序排列的类似于字典查找拼音a、b、c……和字典文字内容顺序是相同的，非聚集索引与内容是非顺序排列的，类似字典偏旁查找时，同一个偏旁‘王’的汉字可能一个在第1页一个在第5页。

二：创建测试表

create table shopping_user(uId bigint primary key,uName varchar(10));
create table shopping_goods_category(cId bigint primary key,cName varchar(20));
create table shopping_goods(gId bigint primary key,gName varchar(50),gcId bigint,gPrice int);
create table shopping_order(oId bigint primary key,oUserId bigint,oAddTime datetime,oGoodsId bigint,oMoney int);
　　

　　创建测试sql

declare @index int;
set @index = 1;
while(@index<=10)
begin
    insert into shopping_user (uId,uName) values(@index,'user'+cast(@index as varchar(10)));
    set @index = @index+1;
end;

insert into shopping_goods_category (cid,cName) values(1,'水果');
insert into shopping_goods_category (cid,cName) values( 2,'电脑');
insert into shopping_goods_category (cid,cName) values (3,'手机');
insert into shopping_goods_category (cid,cName) values (4,'服装');
insert into shopping_goods_category (cid,cName) values (5,'食品');

------ 商品表sql

declare @index int;
declare @num int;
set @index = 1;
set @num = 10000;
begin
    while(@index<=100*@num)
    begin
        if @index<=10*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,1,'水果'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end;
        else if @index >10*@num and @index <=20*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,1,'水果'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end;
        else if @index >20*@num and @index <=30*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,2,'电脑'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end;
        else if @index >30*@num and @index <=40*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,2,'电脑'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end;
        else if @index >40*@num and @index <=50*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,3,'手机'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end;
        else if @index >50*@num and @index <=60*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,3,'手机'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end; 
        else if @index >60*@num and @index <=70*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,4,'服装'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end; 
        else if @index >70*@num and @index <=80*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,4,'服装'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end; 
        else if @index >80*@num and @index <=90*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,5,'食品'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end; 
        else if @index >90*@num and @index <=100*@num
            begin
                insert into shopping_goods (gId,gcId,gName,gPrice)
                values (@index,5,'食品'+cast (@index as varchar(10)),cast( floor(rand()*100) as int) );
            end; 
        set @index = @index+1;
    end; 
end;


------- 订单表sql

declare @index int;
declare @num int;
declare @timeNum int;
declare @userId int;
declare @goodsId int; 
declare @money int;
declare @addTime varchar(30);
set @index = 1;
set @num = 10000; 
set @timeNum = 0;
set @userId = 1;
set @goodsid = 1;
set @money = 100;
set @addTime = '';
begin
    while(@index<=100*@num)
    begin
        set @timeNum = cast( floor(rand()*30)+1 as int)
    set @userId = cast( floor(rand()*99)+1 as int)
    set @money = cast ( floor(rand()*5000)+@userId as int)
    set @addTime = dateadd(day,@timeNum,getdate())
    set @goodsId = cast( floor(rand()*999999)+1 as int)
        if @index<=10*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end;
        else if @index >10*@num and @index <=20*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end;
        else if @index >20*@num and @index <=30*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end;
        else if @index >30*@num and @index <=40*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end;
        else if @index >40*@num and @index <=50*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end;
        else if @index >50*@num and @index <=60*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end; 
        else if @index >60*@num and @index <=70*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end; 
        else if @index >70*@num and @index <=80*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end; 
        else if @index >80*@num and @index <=90*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end; 
        else if @index >90*@num and @index <=100*@num
            begin
                insert into shopping_order (oid,oUserId,oAddTime,oGoodsId,oMoney)
                values (@index,@userId,@addTime,@goodsId,@money );
            end; 
    
    set @index = @index+1;
    end;
    
end;

　　创建索引

create index gcid_index on shopping_goods (gcid);
create index userid_index on shopping_order(ouserid);
create index goodsid_index on shopping_order(ogoodsid);

三：执行计划分析

　　这里使用上一篇文章sql语句百万数据量优化方案中提到的，in和exists来分析,sql语句如下:

SET STATISTICS IO ON

select top 20 * from shopping_order where exists (
select top 10 gid from shopping_goods where gcid =2 and ogoodsid = gid order by gprice desc)

select top 20 * from shopping_order where goodsid in (
select top 10 gid from shopping_goods where gcid =2 order by gprice desc)
 

-- DBCC DROPCLEANBUFFERS 

从上图中发现，使用exists，开销最大的是，使用聚集索引查找，而使用in,第一次操作（从右各左看），就使用了聚集索引扫描，in的效果明显差。我们再来看聚集索引查找结果,聚集索引返回的行数是20，见下图.

然后我们来看使用in查询，聚集索引扫描，查询结果却是20w

接着我们来看使用in查询，第二个开销大的排序，从刚才查询出来的20w数据中，order by desc 返回前20条数据。

此处我们还可以使用SET STATISTICS IO ON来查询这两者的io开销: 

    扫描计数：执行的扫描次数;

    逻辑读取：从数据缓存读取的页数;

    物理读取：从磁盘读取的页数;

    预读：为进行查询而放入缓存的页数

重要：如果对于一个SQL查询有多种写法，那么这四个值中的逻辑读(logical reads)决定了哪个是最优化的。

从上图中发现，exists查询：shopping_order表扫描次数是2,逻辑读取是80，shopping_goods表，扫描次数是1，逻辑读取是6次,

　　　　　　　　　　而in  shopping_order表扫描次数是2,逻辑读取是55，shopping_goods表，扫描次数是5，逻辑读取是5247次,当然工作中的sql肯定要复杂得多，但我们可以借助这个工具来找到需要优化的sql，当然这也只是执行计划，可能实际执行的效率和这个计划有出入，但我们还是可以借鉴执行计划来找到其中的不足。