Question

我有一个items表，其中包含状态和创建日期

+----+-----------+------------+
| id | status    | created    |
+----+-----------+------------+
| 1  | PROCESSED | 2018-12-01 |
+----+-----------+------------+
| 2  | PROCESSED | 2018-12-01 |
+----+-----------+------------+
| 3  | ABORTED   | 2018-12-01 |
+----+-----------+------------+

有一个相应的item status表，该表在状态更改时会更新

+----+---------+-----------+------------------+
| id | item_id | status    | created          |
+----+---------+-----------+------------------+
| 1  | 1       | RECEIVED  | 2018-12-01 10:00 |
+----+---------+-----------+------------------+
| 2  | 1       | PROCESSED | 2018-12-01 12:00 |
+----+---------+-----------+------------------+
| 3  | 2       | RECEIVED  | 2018-12-01 11:00 |
+----+---------+-----------+------------------+
| 4  | 2       | PROCESSED | 2018-12-01 12:00 |
+----+---------+-----------+------------------+
| 5  | 3       | RECEIVED  | 2018-12-01 13:00 |
+----+---------+-----------+------------------+
| 6  | 3       | ABORTED   | 2018-12-01 13:30 |
+----+---------+-----------+------------------+

我想生成一个报告，该报告按天分组显示要处理的项目的平均时间，不包括已中止的项目。（要处理项目的时间是RECEIVED和PROCESSED之间的时间差）

类似的事情（持续时间以秒为单位）：

+------------+------------------+
| day        | avg_duration     |
+------------+------------------+
| 2018-12-01 | 5400             |
+------------+------------------+

从其他问题来看，我已经确定可以使用表分区解决此问题，但是还无法编写有效的查询。最好的方法是什么？

Answer 1

使用布尔聚合bool_and()过滤掉中止的项目：

select date, avg(duration)
from (
    select created::date as date, item_id, extract(epoch from max(created)- min(created)) as duration
    from item_status
    group by created::date, item_id
    having bool_and(status <> 'ABORTED')
    ) s
group by date

    date    | avg  
------------+------
 2018-12-01 | 5400
(1 row)

Answer 2

这需要2个汇总级别，一次是在项目和日期上，然后是在日期上。

select dt_created,avg(diff) as avg_diff
from (select item_id
            ,created::date as dt_created
            ,max(case when status = 'PROCESSED' then created end) - max(case when status = 'RECEIVED' then created end) as diff 
      from item_statuses 
      group by item_id,created::date
      having count(case when status = 'ABORTED' then 1 end) = 0
    ) t
group by dt_Created

计算两行之间的差异，按天平均

2 个答案: