Postgres：将连续行转换为离散行

Question

I'm not sure wether the title is correct. I've spent some time to find the best title suited for this case.

Let say. I have data in this format:

| id | room |     changes      |   created_at    |
|----|------|------------------|-----------------|
|  1 | A    | [free, occupied] | 2021-11-1 18:00 |
|  2 | B    | [free, occupied] | 2021-11-1 20:30 |
|  3 | B    | [occupied, free] | 2021-11-1 23:30 |
|  4 | A    | [occupied, free] | 2021-11-2 00:15 |
|  5 | B    | [free, occupied] | 2021-11-2 01:00 |

The output required will be

+----+------+----------+-----------------+
| id | room | occupied |   Period        |
+----+------+----------+-----------------+
|  1 | A    | NO       | 2021-11-1 18:00 |
|  2 | B    | NO       | 2021-11-1 18:00 |
|  3 | A    | NO       | 2021-11-1 19:00 |
|  4 | B    | NO       | 2021-11-1 19:00 |
|  5 | A    | YES      | 2021-11-1 20:00 |
|  6 | B    | NO       | 2021-11-1 20:00 |
|  7 | A    | YES      | 2021-11-1 21:00 |
|  8 | B    | NO       | 2021-11-1 21:00 |
|  9 | A    | YES      | 2021-11-1 22:00 |
| 10 | B    | NO       | 2021-11-1 22:00 |
| 11 | A    | YES      | 2021-11-1 23:00 |
| 12 | B    | YES      | 2021-11-1 23:00 |
| 13 | A    | YES      | 2021-11-2 00:00 |
| 14 | B    | YES      | 2021-11-2 00:00 |
| 15 | A    | NO       | 2021-11-2 01:00 |
| 16 | B    | YES      | 2021-11-2 01:00 |
| 17 | A    | NO       | 2021-11-2 02:00 |
| 18 | B    | YES      | 2021-11-2 02:00 |
+----+------+----------+-----------------+

The output generated using this logic:

• When during that period. the room is changed to occupied Occupied will be "YES"
• when during that period. no changes, the change is taken from the previous change
• When during that period. the room is changed from occupied to free. The next period occupied will be "NO"

Answer 1

First some comments about your question:

• some period values in your output doesn't correspond to the created_at values in your table
• the id column in your output doesn't relate to the id column in your table because half of the output rows don't correspond to a row in your table, ie their id value can't be determined
• the occupied value for the first period value in your output can be set up for the associated room (A) in your table but not for the others (B), so here an arbitrary value must be set up

Then I propose you a query which implements the following rules:

• When the room is changed to occupied , then occupied = "YES" for period = created_at
• when no change, ie the row doesn't exist in your_table for the given (room, period) of your output, the occupied value is copied from the previous rows with the same room and the most recent created_at value
• When the room is changed to free , then occupied = "NO" for period = created_at

This query may provide a result slightly different from your expectation, but it can be easily adjusted.

see the SQLfiddle

SELECT t.id -- may be NULL according to the LEFT JOIN and this is a gap with the expected result as presented in the question
     , r.room
     , CASE
         WHEN t.room = r.room AND t.changes = '[free, occupied]'
         THEN 'YES'
         WHEN t.room = r.room AND t.changes = '[occupied, free]'
         THEN 'NO'
         WHEN ( jsonb_agg(t.changes)
                         FILTER (WHERE t.room = r.room)
                         OVER (PARTITION BY r.room ORDER BY p.created_at ASC ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW)
              )->>-1 IS NOT DISTINCT FROM '[free, occupied]'
         THEN 'YES'
         ELSE 'NO'
       END AS occupied
     , p.created_at AS period
  FROM 
     ( SELECT DISTINCT ON (created_at) created_at
         FROM your_table
     ) AS p
 CROSS JOIN
     ( SELECT DISTINCT ON (room) room
         FROM your_table
     ) AS r
  LEFT JOIN your_table AS t
    ON t.room = r.room
   AND t.created_at = p.created_at
 ORDER BY p.created_at, r.room

In the result, the id column corresponds to the id value in your_table . So this id is NULL when the output row doesn't correspond to any rows in your_table .