聚合SQL中的正确/错误序列

Question

我在psql中有一个带有id列， timestamp列和has_connection列的表

 id |         ts          | has_connection 
----+---------------------+----------------
  1 | 2016-06-22 19:10:25 | t
  1 | 2016-06-22 19:10:26 | t
  1 | 2016-06-22 19:10:27 | f
  1 | 2016-06-22 19:10:28 | t
  1 | 2016-06-22 19:10:29 | f
  1 | 2016-06-22 19:10:30 | f
  1 | 2016-06-22 19:10:31 | t
  1 | 2016-06-22 19:10:32 | t

我的目标是弄清FALSE或FALSE序列开始时以及结束时的情况。

SELECT *,  CASE WHEN lag(has_connection) OVER (PARTITION BY id ORDER BY ts ASC) <>  has_connection  THEN 'change' ELSE 'no_change' END as change FROM test

 id |         ts          | has_connection |  change   
----+---------------------+----------------+-----------
  1 | 2016-06-22 19:10:25 | t              | no_change
  1 | 2016-06-22 19:10:26 | t              | no_change
  1 | 2016-06-22 19:10:27 | f              | change
  1 | 2016-06-22 19:10:28 | t              | change
  1 | 2016-06-22 19:10:29 | f              | change
  1 | 2016-06-22 19:10:30 | f              | no_change
  1 | 2016-06-22 19:10:31 | t              | change
  1 | 2016-06-22 19:10:32 | t              | no_change

上面的内容将使我了解发生翻转的位置，但是产生低于所需输出的最有效方法是什么？

 id |     connection_end  | reconnect
----+---------------------+----------------
  1 | 2016-06-22 19:10:27 | 2016-06-22 19:10:28 
  1 | 2016-06-22 19:10:29 | 2016-06-22 19:10:31

这是数据：

CREATE TABLE test  (id INT, ts TIMESTAMP, has_connection BOOL);

INSERT INTO test (id, ts, has_connection)
VALUES
   (
      1,
      '2016-06-22 19:10:25',
      TRUE
  ),(
      1,
      '2016-06-22 19:10:26',
      TRUE
  ),(
      1,
      '2016-06-22 19:10:27',
      FALSE
  ),(
      1,
      '2016-06-22 19:10:28',
      TRUE
  ),(
      1,
      '2016-06-22 19:10:29',
      FALSE
  ),(
      1,
      '2016-06-22 19:10:30',
      FALSE
  ),(
      1,
      '2016-06-22 19:10:31',
      TRUE
  ),(
      1,
      '2016-06-22 19:10:32',
      TRUE
   );

Answer 1

您想要将一个分组分配给与以下true相同的false组。 这是标准间隔和孤岛问题的变体。

您可以通过计算给定行上或之后的true值数量来计算分组。 所有false都具有相同的值。 并且下面的true （如果有）也是如此。

然后合计：

select id,
       min(ts) as connection_end, max(ts) as reconnect
from (select t.*,
             count(*) filter (where has_connection) over (order by ts desc) as grp
      from test t
     ) t
group by id, grp
having not (array_agg(has_connection order by ts))[1]
order by min(ts);

这是db <>小提琴。

对此进行过滤有些棘手。 内部子查询可正确识别组。 但是，这些组可以同时具有true和false值，因此不能简单地对has_connection过滤。 相反，一种方法是检查每个组中的第一个值是否为false 。

另一种方法是只是为了检查是否存在任何 false组中值。 这些总是先于true ，所以两者是等效的。 短语为：

having bool_or(not has_connection)

此外，这还假定最后一个值始终为true。 如果序列可以以true值结尾，那么这会更准确：

select id, min(ts) as connection_end,
       max(ts) filter (where has_connection) as reconnect
from (select t.*,
             count(*) filter (where has_connection) over (order by ts desc) as grp
      from test t
     ) t
group by id, grp
having not (array_agg(has_connection order by ts))[1]
order by min(ts)

Answer 2

使用lag（）窗口函数：

select *,
  case when has_connection = 
    coalesce(lag(has_connection, 1) over (partition by id order by ts), has_connection)
    then 'no_change' else 'change'
  end change  
from test
order by id, ts

结果：

| id  | ts                       | has_connection | change    |
| --- | ------------------------ | -------------- | --------- |
| 1   | 2016-06-22T19:10:25.000Z | true           | no_change |
| 1   | 2016-06-22T19:10:26.000Z | true           | no_change |
| 1   | 2016-06-22T19:10:27.000Z | false          | change    |
| 1   | 2016-06-22T19:10:28.000Z | true           | change    |
| 1   | 2016-06-22T19:10:29.000Z | false          | change    |
| 1   | 2016-06-22T19:10:30.000Z | false          | no_change |
| 1   | 2016-06-22T19:10:31.000Z | true           | change    |
| 1   | 2016-06-22T19:10:32.000Z | true           | no_change |

您可以将类似的查询用作CTE以获取最终结果：

with 
  changes as (
    select *,
      case when has_connection = 
        coalesce(lag(has_connection, 1) over (partition by id order by ts), has_connection)
        then false else true
      end changed  
    from test
  ), 
  rows as (
    select *, row_number() over (partition by id, has_connection order by ts) rn
    from changes
    where changed
  )
select r1.id, r1.ts connection_end, r2.ts reconnect          
from rows r1 inner join rows r2
on r2.id = r1.id and r2.rn = r1.rn
where not r1.has_connection and r2.has_connection

结果：

id | connection_end      | reconnect          
 1 | 2016-06-22 19:10:27 | 2016-06-22 19:10:28
 1 | 2016-06-22 19:10:29 | 2016-06-22 19:10:31

参见演示 。

Answer 3

使用递归CTE的另一种解决方案：

with recursive cte as (
    (select id, null::timestamp as connection_end, ts as reconnect, has_connection 
     from test order by ts limit 1)
    union all
    (select test.id, cte.reconnect, test.ts, test.has_connection
     from test join cte on (test.ts > cte.reconnect and test.has_connection <> cte.has_connection)
     order by test.ts limit 1))
select * from cte where has_connection and connection_end is not null;