将 Postgresql 转换为 Redshift 的挑战（过滤位置）答案

【问题标题】：Challenges with Transforming Postgresql into Redshift (Filter Where)将 Postgresql 转换为 Redshift 的挑战（过滤位置）
【发布时间】：2021-07-14 11:16:26
【问题描述】：

我在从 postgresql 转换为 redshift 方言时注意到一些事情，当使用过滤器 where 时，但不能完全解释为什么这 2 个查询会产生不同的结果。希望有人能解释一下。

Postgresql：

select date_trunc('week', date)::DATE as "Week", 
date_trunc('year', date)::DATE as "Year",
country, 
region, 
count(distinct o.id) filter (where status='delivered') as "Delivered",
count(distinct o.id) filter (where status='packaged') as "Packaged"
from country as c
left join orders as o on c.order_id=o.id and date_trunc('week', o.date)::DATE=date_trunc('week', c.date)::DATE and date_trunc('year', o.date)::DATE=date_trunc('year', c.date)::DATE
where country='IT'
and product='CAT'
group by 1,2
order by 1

通过此查询，我可以查看国家 IT、产品 CAT 在该年和该周交付和包装了多少订单

转换为 Redshift（不能在 where 使用过滤器 - 用 case when 代替）

select extract(week from date) as "Week", 
extract(year from date) as "Year", 
country, 
region, 
case when status='delivered' then count(distinct o.id) as "Delivered",
case when status='packaged' then count(distinct o.id) as "Packaged"
from country as c
left join orders as o on c.order_id=o.id and extract(week from o.date)=extract(week from c.date) and extract(year from o.date)=extract(year from c.date)
where country='IT'
and product='CAT'
group by 1,2, status
order by 1

这里有一些变化：

要提取的日期截断
过滤（其中..）以区分大小写
将状态添加到分组依据（否则会出错）

但主要问题是我没有得到相同的输出，相反，我在同一周有几行，并且值正在跳过行。（所以它不仅仅是重复的行，我可以通过使用 distinct 轻松摆脱）

【问题讨论】：

标签： sql filter group-by amazon-redshift aggregate-functions

【解决方案1】：

您将case 放在错误的位置。等价的逻辑是：

select date_trunc('week', date)::DATE as Week, 
       date_trunc('year', date)::DATE as Year,
       country, region, 
       count(distinct case when status = 'delivered' then o.id end) as Delivered,
       count(distinct case when status = 'packaged' then o.id end) as Packaged
from country c left join
     orders o
     on c.order_id = o.id and
        date_trunc('week', o.date)::DATE = date_trunc('week', c.date)::DATE and
        date_trunc('year', o.date)::DATE = date_trunc('year', c.date)::DATE
where country = 'IT' and product = 'CAT'
group by 1, 2
order by 1;

【讨论】：