我想要左连接的解决方案来改善存储过程的结果

Posted 2023-04-14

【中文标题】我想要左连接的解决方案来改善存储过程的结果

【英文标题】：I want solution of left join to improve stored procedure result

【发布时间】：2017-11-06 09:53:08

【问题描述】：

我想优化存储过程。请提供改进建议。

有4个left join，耗时太长。

t1 表中有超过 1000 万行。我想按月分组并过滤其他关系。

SET NOCOUNT ON;

declare @v1 datetime
declare @v2 datetime
declare @v3 int = null
declare @v4 varchar(max)
declare @v5 varchar(max)
declare @v6 varchar(max)
declare @v7 varchar(max)
declare @v8 varchar(max)
declare @v9 int

set @v1 = @iv1
set @v2 = @iv2
set @v3 = @iv3
set @v4 = @iv4
set @v5 = @iv5
set @v6 = @iv6
set @v7 = @iv7
set @v8 = @iv8

select 
    Month(a.c1) as 'Month',
    sum(a.c2) as 'PaidAmount',
    count(a.c3)*2.5 as 'TransactionFee'
from 
    t1 a
left join 
    t2 b on a.c1 = b.c1
left join 
    t3 f on b.c2 = f.c2
left join 
    t4 c on a.c3 = c.c3
left join 
    t5 g on a.c4 = g.c4
where 
    a.c4 >= @v1 
    and a.c5 <= @v2
    and a.c6 > 0
    and a.c7 is not null 
    and a.c8 is not null 
    and a.c9 is not null
    and (@v3 is null or b.c10 = @v3)
    and (@v4 is null or CHARINDEX(','+CAST(c.c7 as varchar(max))+',',@v4) >  0)
    and (@v5 is null or CHARINDEX(','+CAST(a.c8 as varchar(max))+',',@v5) > 0)
    and (@v6 is null or CHARINDEX(','+CAST(a.c9 as varchar(max))+',',@v6) > 0)
    and (@v7 is null or CHARINDEX(','+CAST(c.c11 as varchar(max))+',',@v7) > 0)
    and (@v8 is null or CHARINDEX(','+CAST(g.c12 as varchar(max))+',',@v8) > 0)
group by 
    Month(a.c1)

【问题讨论】：

“给我改进的想法和建议。”听起来像是命令，而不是问题。你认为我们如何给你关于性能的建议？我们对您的系统一无所知！

“给我们表格规格！给我们你到目前为止尝试过的东西！”

@Tyron78 在表 t1 中有超过 1000 万条记录和 20 多列。表 t1 是索赔清单。我想按月计算，对其他聚合组求和。

@NiravPatel 只是出于好奇：您是否尝试使用固定值而不是变量来执行查询？我之前遇到过巨大的性能问题，因为优化器对查询中变量的使用有点讨厌......可能值得一试（作为第一种方法）。

你看过实际的执行计划了吗？它告诉你什么？ 表结构是什么？每个表都有哪些主键？外键列是否正确索引？

【参考方案1】：

    declare @v1 datetime
    declare @v2 datetime
    declare @v3 int = null
    declare @v4 varchar(max)
    declare @v5 varchar(max)
    declare @v6 varchar(max)
    declare @v7 varchar(max)
    declare @v8 varchar(max)
    declare @v9 int
    set @v1 = @iv1
    set @v2 = @iv2
    set @v3 = @iv3
    set @v4 = @iv4
    set @v5 = @iv5
    set @v6 = @iv6
    set @v7 = @iv7
    set @v8 = @iv8

select Month(a.c1) as 'Month',
sum(a.c2) as 'PaidAmount',
count(a.c3)*2.5 as 'TransactionFee'
from 
t1 a WITH(NOLOCK)
FULL join t2 b WITH(NOLOCK) on a.c1 = b.c1 
INNER join t4 c WITH(NOLOCK) on a.c3 = c.c3
INNER join t5 g WITH(NOLOCK) on a.c4 = g.c4
INNER join t3 f WITH(NOLOCK) on b.c2 = f.c2
where a.c4 >= @v1 
and a.c5 <= @v2
and a.c6 > 0
and a.c7 is not null 
and a.c8 is not null 
and a.c9 is not null
and (@v3 is null or b.c10 = @v3)
and (@v4 is null or CHARINDEX(','+CAST(c.c7 as varchar(max))+',',@v4) > 0)
and (@v5 is null or CHARINDEX(','+CAST(a.c8 as varchar(max))+',',@v5) > 0)
and (@v6 is null or CHARINDEX(','+CAST(a.c9 as varchar(max))+',',@v6) > 0)
and (@v7 is null or CHARINDEX(','+CAST(c.c11 as varchar(max))+',',@v7) > 0)
and (@v8 is null or CHARINDEX(','+CAST(g.c12 as varchar(max))+',',@v8) > 0)
group by Month(a.c1)

【讨论】：

请问您能EXPLAIN用简单的英语WHAT您做了什么，这应该如何提高性能？请不要在没有任何解释的情况下向我们倾倒越来越大的代码.....

由于索引扫描，左连接总是性能问题，所以我发现 t1 表作为过滤条件（索引搜索）并且与其他表有关系，所以我对表 t2 使用完全连接而不是左连接（NOLOCKS ) 提示。