尽管我在 where 子句中包含 null，但为什么记录从我的左连接中删除

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 表返回的记录比我的 nice 表少？ 我该如何解决这个问题，我希望表 nice 中的所有记录都得到保留。 我认为我在 where 子句中的 OR 语句可以处理这个问题，但不确定为什么不能。 我知道 where 子句将查询转换为内部联接这一事实，但我认为 there OR 子句 with is null 应该可以解决该问题。 你能帮忙吗？

Answer 1

因为你在WHERE子句中放置了一个条件，那就是过滤掉行。

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;

你看得到差别吗？

Answer 2

您可以通过将相关条件移动到ON子句来解决问题：

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因为 a 中的某些行在b中a匹配的行，这些行不满足任何一个计时条件。 因此，不满足两个时间限制。 并且该值不为null ，因为存在匹配的行。

Answer 3

拥有一些最小的可重现示例总是有帮助的。 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;

退货：

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

第二行被过滤掉，因为 (ts_diff between 0 and 86400) 不满足。

如果你真的想要这个结果：

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