在 redshift 中重复连续运行后，除查询的结果发生变化

Question

我在 aws redshift 中运行下面的 postgresql 查询。每次我运行此查询时，我都会使用 except 运算符在 daily_table.product_repeat_sub_query 端获得不同的记录数的不同结果。在此期间，daily_table.product_repeat_sub_query 表或 daily_table.daily_sku_t 都不会更新。 daily_table.product_repeat_sub_query 表和 product_repeat_sub_query 查询都具有相同的记录数。 daily_table.daily_sku_t 的架构如下，daily_table.product_repeat_sub_query 中的匹配字段具有相同的数据类型。我还包括了下表中的一些示例记录。有没有人知道，当底层表没有改变时，每次运行这个查询时，except 查询的结果会如何不同？

daily_table.daily_sku_t schema:

customer_uuid string
boardname_12 string
producttype string
productsubtype string
storeid int
product_id string
dateclosed date
Size string

查询：

with product_repeat_sub_query as
(
        select
            dateclosed, t.product_id, t.storeid, t.producttype, t.productsubtype, t.size, t.boardname_12,
            case
                when ticketid = first_value(ticketid) over (partition by t.product_id, customer_uuid
            ORDER BY
                dateclosed ASC rows between unbounded preceding and unbounded following) then 0
                else grossreceipts
            end as product_repeat_gross, datediff(day,
            lag(dateclosed, 1) over (partition by t.boardname_12, customer_uuid, t.product_id
        ORDER BY
            dateclosed ASC ),
            dateclosed) as product_cycle_days
        from
            daily_table.daily_sku_t t )
            
select count(*) from
(
select dateclosed, storeid, boardname_12, producttype, productsubtype, size, product_id, product_cycle_days from daily_table.product_repeat_sub_query
except
select dateclosed, storeid, boardname_12, producttype, productsubtype, size, product_id, product_cycle_days from product_repeat_sub_query
);

-- 36843 -- 36887 -- 36188

数据：

daily_table.product_repeat_sub_query

dateclosed  storeid boardname_12    producttype productsubtype  size    product_id  product_cycle_days
2021-04-23  427     22RED                       DRUMER          1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    2
2021-04-24  427     22RED                       DRUMER          1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    6
2021-04-26  427     22RED                       DRUMER          1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    8
2021-04-26  427     22RED                       DRUMER          1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    3
2021-05-01  427     22RED                       DRUMER          1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    13
2020-06-18  61      FLAV RX         WINGER      BEVERAGE        100MT   0000265d-6b81-4d79-90cf-xxxxxxxxxxxx    5
2020-06-29  

product_repeat_subquery

dateclosed  storeid boardname_12    producttype productsubtype  size    product_id  product_cycle_days
2021-04-23  427     22RED           DRUMER                      1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    2
2021-04-24  427     22RED           DRUMER                      1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    6
2021-04-26  427     22RED           DRUMER                      1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    8
2021-04-26  427     22RED           DRUMER                      1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    3
2021-05-01  427     22RED           DRUMER                      1T  000011aa-4f03-4f0b-a621-xxxxxxxxxxxx    13
2020-06-18  61      FLAV RX         WINGER      BEVERAGE        100MT   0000265d-6b81-4d79-90cf-xxxxxxxxxxxx    5
2020-06-29  

更新：

with product_repeat_sub_query as
    (
            select customer_uuid,
                dateclosed, t.product_id, t.storeid, t.producttype, t.productsubtype, t.size, t.boardname_12,
                case
                    when ticketid = first_value(ticketid) over (partition by t.product_id, customer_uuid
                ORDER BY
                    dateclosed ASC rows between unbounded preceding and unbounded following) then 0
                    else grossreceipts
                end as product_repeat_gross, datediff(day,
                lag(dateclosed, 1) over (partition by t.boardname_12, customer_uuid, t.product_id
            ORDER BY
                dateclosed ASC,t.boardname_12, customer_uuid, t.product_id ),
                dateclosed) as product_cycle_days
            from
                daily_table.daily_sku_t t 

where (t.customer_uuid is not null) 
            and (trim(t.customer_uuid) != '')
            and (t.product_id is not null) 
            and (trim(t.product_id) != '')

)
                
    select count(*) from
    (
    select customer_uuid, dateclosed, storeid, boardname_12, producttype, productsubtype, size, product_id, product_cycle_days from daily_table.product_repeat_sub_query
    except
    select customer_uuid, dateclosed, storeid, boardname_12, producttype, productsubtype, size, product_id, product_cycle_days from product_repeat_sub_query
    );

即使在将分区中的所有字段添加到 order by 并过滤 id 字段中的空值或空白之后，我每次仍然得到不同的计数。

Answer 1

您的窗口函数没有完全限定的 order by 子句。您在分区中重复了“日期关闭”值。这意味着 Redshift 可以对滞后函数和一值函数有不同的行顺序。我预计这些“随机”排序差异会导致您的结果发生变化。