UPDATE rows with values from the same table

问题:

I have a table like this:

+------+-------+
|ID    | value |
+------+-------+
| 1    | 150   |
| 2    |       |
| 3    |       |
| 4    |       |
| 5    | 530   |
| 6    | 950   |
| 7    | 651   |
+-------+------+

I want to copy the last 3 values and at the end my table will look like this:

+------+-------+
|ID    | value |
+------+-------+
| 1    | 150   |
| 2    | 530   |
| 3    | 950   |
| 4    | 651   |
| 5    | 530   |
| 6    | 950   |
| 7    | 651   |
+-------+------+

Is it possible?

回答1:

Use a self-join:

UPDATE mytable m
SET    value = m0.value
FROM   mytable m0
WHERE  m.id = (m0.id - 3)   -- define offset
AND    m.id BETWEEN 2 AND 4 -- define range to be affected
AND    m.value IS NULL;     -- make sure only NULL values are updated

If there are gaps in the ID space, use the windows function row_number() to get gapless ID to work with. I do that in a CTE, because I am going to reuse the table twice for a self-join:

WITH x AS (
   SELECT *, row_number() OVER (ORDER BY ID) AS rn
   FROM   mytable
   )
UPDATE mytable m
SET    value = y.value
FROM   x
JOIN   x AS y ON x.rn = (y.rn - 4567)   -- JOIN CTE x AS y with an offset
WHERE  x.id = m.id                      -- JOIN CTE x to original
AND    m.id BETWEEN 1235 AND 3455
AND    m.value IS NULL;

You need PostgreSQL 9.1 or later for data-modifying CTEs.

回答2:

For an ad-hoc update like this, there probably isn't going to be a better way than three simple update statements:

UPDATE mytable SET value = 530 WHERE id = 2;
UPDATE mytable SET value = 950 WHERE id = 3;
UPDATE mytable SET value = 651 WHERE id = 4;

The question is, is this an ad-hoc update that only applies to this exact data, or a case of a general update rule that you want to implement for all possible data in that table? If so, then we need more detail.

回答3:

The hard-coded 3 appears twice and would be replaced by however many rows you want. It assumes the last 3 records actually have values. It takes those values and applies them in sequence to the set of records with null values.

update a
  set value = x.value
  from (

        select nullRows.id, lastRows.value

          from ( select id, value
                       ,(row_number() over(order by id) - 1) % 3 + 1 AS key
                   from ( select id, value
                            from a
                            order by id desc
                            limit 3
                        ) x
                   order by 1

               ) AS lastRows

              ,( select id
                       ,(row_number() over(order by id) - 1) % 3 + 1 AS key
                   from a
                   where value is null
                   order by id

               ) AS nullRows

         where lastRows.key = nullRows.key

      ) x

where a.id = x.id