为什么记录从我的左联接中删除，尽管我在where子句中包含null

Question

我运行以下查询：

create table c.hello as
select a.*, b.timestamp, b.alert 
from nice a
left join bye b 
  on a.number = b.number_nb 
where (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) >= 0)
  and (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) <= 86400) 
   or b.alert_timestamp is null;

为什么我的hello表返回的记录少于我的漂亮表具有的记录？我该如何解决此问题，我希望保留表中的所有记录。我以为我在where子句中的OR语句将处理此问题，不确定为什么不这样做。我知道一个where子句将查询转换为内部联接的事实，但是我认为with子句with为null应该可以解决该问题。您能帮忙吗？

Answer 1

因为您在WHERE子句中放置了一个条件，所以正在过滤掉行。

将该条件作为JOIN谓词的一部分，如：

create table c.hello as
select a.*, b.timestamp, b.alert 
from nice a
left join bye b 
  on a.number = b.number_nb 
 and (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) >= 0)
 and (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) <= 86400) 
  or b.alert_timestamp is null;

您看到区别了吗？

Answer 2

您可以通过将相关条件移到ON子句来解决此问题：

create table c.hello as
    select a.*, b.timestamp, b.alert 
    from nice a left join
         bye b 
         on a.number = b.number_nb and
            (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) >= 0) and
            (Unix_Timestamp(a.time) - Unix_Timestamp(b.timestamp) <= 86400) ;

您的代码失败的原因是a中的某些行具有b中的匹配行，它们不满足任何时序条件。因此，不能满足两个时间限制。并且该值不是null，因为存在匹配的行。

Answer 3

拥有一些最小的可复制示例总是有帮助的。让我们考虑以下示例：

with A as (
select stack (3,
1, '2019-11-10 10:10:10.000',
2, '2019-12-10 10:10:10.000',
4, '2019-25-10 10:10:10.000'
) as (number_nb,ts)
),

B as (
select stack (2,
1, '2019-11-10 10:10:10.000',
2, '2019-25-10 10:10:10.000' --big difference in timestamp here
) as (number_nb,ts)
)

select a_number_nb, a_ts, 
       b_number_nb, b_ts,
       ts_diff
from
(
select a.number_nb as a_number_nb,a.ts as a_ts, b.number_nb as b_number_nb, b.ts as b_ts, Unix_Timestamp(a.ts) - Unix_Timestamp(b.ts) as ts_diff  
  from A a 
       left join B b on a.number_nb=b.number_nb
)s;

返回：

a_number_nb a_ts                    b_number_nb b_ts                    ts_diff
1           2019-11-10 10:10:10.000 1           2019-11-10 10:10:10.000 0
2           2019-12-10 10:10:10.000 2           2019-25-10 10:10:10.000 -34300800
4           2019-25-10 10:10:10.000 NULL        NULL                    NULL

现在，如果您添加WHERE子句

where (ts_diff between 0 and 86400) or b_number_nb is NULL

甚至是这个：

where (ts_diff between 0 and 86400) or ts_diff is NULL

结果是：

a_number_nb a_ts                    b_number_nb b_ts                    ts_diff
1           2019-11-10 10:10:10.000 1           2019-11-10 10:10:10.000 0
4           2019-25-10 10:10:10.000 NULL        NULL                    NULL

由于未满足（0到86400之间的ts_diff），第二行被滤除。

如果您实际上想要此结果：

a_number_nb a_ts                    b_number_nb b_ts
1           2019-11-10 10:10:10.000 1           2019-11-10 10:10:10.000
2           2019-12-10 10:10:10.000 NULL        NULL
4           2019-25-10 10:10:10.000 NULL        NULL 

然后删除WHERE条件，改为使用case语句：

... --Use same CTEs as in first example, skipped

select a_number_nb, a_ts, 
       case when (ts_diff between 0 and 86400) then b_number_nb end b_number_nb , --NULL if ts_diff NOT between 0 and 86400
       case when (ts_diff between 0 and 86400) then b_ts        end b_ts
       --ts_diff
from
(
select a.number_nb as a_number_nb,a.ts as a_ts, b.number_nb as b_number_nb, b.ts as b_ts, Unix_Timestamp(a.ts) - Unix_Timestamp(b.ts) as ts_diff  
  from A a 
       left join B b on a.number_nb=b.number_nb
)s