Question

第一个查询适用于索引搜索（https://www.brentozar.com/pastetheplan/?id=B1XMM_YeW）：

SELECT *
FROM dbo.MessageLast c
    WHERE c.Stat_Date >= '2017-04-25' 
                AND c.Stat_Date <= '2017-04- 26' 
                AND c.partner_id IN (111294, 100001, 111295, 111435, 111296, 118155,111498, 118397);

第二次查询错误，使用idex扫描（https://www.brentozar.com/pastetheplan/?id=H1uvEdFgZ）：

CREATE TABLE #t (partner_id Int primary key clustered);
INSERT INTO  #t(partner_id) 
values  (111294), (100001), (111295), (111435), (111296), (118155),(111498), (118397);

SELECT *
FROM dbo.MessageLast c
    where  c.Stat_Date >= '2017-04-25' 
                and c.Stat_Date <= '2017-04- 26' 
                and c.partner_id in (select partner_id from #t);

使用EXISTS进行的第三次查询也很糟糕（https://www.brentozar.com/pastetheplan/?id=ryk3IntgZ）：

 SELECT *
 FROM   dbo.MessageLast c with (forceseek)

 WHERE  c.Stat_Date >= '2017-04-25'
        and c.Stat_Date <= '2017-04-26'
        and exists(select
                       * 
                   from  #d as d where  c.partner_id = d.partner_id)

更改表变量或标准表上的临时表#t没有帮助 - 与第二个查询相同的计划。

简单的演示脚本：

create table _tBig (
  dt Date
, id1 Int
, id2 Int
, id3 Int
, id4 Int
, txt Varchar(500)
) 
create clustered index PK_tBig  on _tBig (id4,dt)



--truncate table _tBig

declare @i Int = 1
set nocount on
while @i < 10000
begin
    insert  into _tBig with (tablock)
            (dt
           , id1
           , id2
           , id3
           , id4
           , txt
            )
    select top (1000)
            DateAdd(day, Rand() * 365 + 1, GetDate())
          , Row_Number() over (order by O.[object_id])
          , Rand() * 10000000 + 1
          , Rand() * 10000000 + 1
          , Rand() * 10000000 + 1
          , NewId()
    from    sys.all_objects O

    set @i += 1

end 


create table _tId (id1 Int)



--truncate table _tId

insert  into _tId
        (id1
        )
select top (5)
        Row_Number() over (order by O.[object_id])
from    sys.all_objects O



select  *
from    _tBig as tb
where   tb.dt >= '20170517'
        and tb.dt < '20170519'
        and tb.id1 in (1, 2, 3, 4, 5)

create index IX_Big on _tBig (dt,id1)

select  *
from    _tBig as tb
inner join _tId as ti on ti.id1 = tb.id1
where   tb.dt >= '20170517'
        and tb.dt < '20170519'


create index IX_Big2 on _tBig (id1,dt)

select  *
from    _tBig as tb
inner join _tId as ti on ti.id1 = tb.id1
where   tb.dt >= '20170517'
        and tb.dt < '20170519'

--drop table dbo._tBig
--drop table dbo._tId

Answer 1

你应该尝试：

SELECT *
FROM dbo.MessageLast c
        JOIN #t t on c.partner_id = t.partner_id
where  c.Stat_Date >= '2017-04-25' 
                and c.Stat_Date <= '2017-04- 26'

Answer 2

这是因为索引未覆盖：您正在选择表中的所有字段。

当索引未覆盖查询sql server时，请执行以下两个步骤：

1）索引寻找找到mathed谓词

的PK值（并包括列数）

2）密钥查找（在聚簇索引中搜索）每行以查找未包含在索引中的列的值。

因此，在某些情况下，查询优化器会确定全表扫描比这两个步骤更快。

您可以将SELECT *替换为SELECT dt（制作索引覆盖）并获得索引搜索。以下是更多信息：What is a Covered Index?

在SQL Server SELECT查询

2 个答案: