Question

I have this table :

| Column        | Type                           |
+---------------+--------------------------------+         
| id            | integer                        |
| recipient_id  | integer                        | 
| is_read       | boolean                        | 
| updated_at    | timestamp(0) without time zone | 

I have to delete items from this table with this specific rule:

• for each recipient_id, we keep the 5 last read items, and we delete the old read one.

I tried to bend my mind with RECURSIVE WITH statements but failed miserably. I’ve implemented my solution programmatically but I wanted to know if there was a decent pure SQL solution.

Answer 1

DELETE FROM tbl t
USING (
    SELECT id, row_number() OVER (PARTITION BY recipient_id
                                  ORDER BY updated_at DESC) as rn
    FROM   tbl
    WHERE  is_read
  ) x
WHERE  x.rn > 5
AND    x.id = t.id;

A JOIN is usually faster than an IN expression, especially with larger numbers of items.
And use row_number(), not rank()!