将总计行的值拆分为其他多个行，直到总和达到 REDSHIFT 中总计行的值

Question

DB-Fiddle

CREATE TABLE inbound (
    id SERIAL PRIMARY KEY,
    campaign VARCHAR,
    expected_inbound_date DATE,
    expected_inbound_quantity DECIMAL,
    received_inbound_quantity DECIMAL
);

INSERT INTO inbound
(campaign, expected_inbound_date, expected_inbound_quantity, received_inbound_quantity)
VALUES 
('C001', '2022-05-03', '500', '0'),
('C001', '2022-05-03', '800', '0'),
('C001', '2022-05-03', '400', '0'),
('C001', '2022-05-03', '200', '0'),
('C001', NULL, '0', '700'),

('C002', '2022-08-20', '3000', '0'),
('C002', '2022-08-20', '5000', '0'),
('C002', '2022-08-20', '2800', '0'),
('C002', NULL, '0', '4000');

Expected Result

campaign |  expected_inbound_date |  expected_inbound_quantity  |  split_received_inbound_quantity
---------|------------------------|-----------------------------|----------------------------------
  C001   |        2022-05-03      |             200             |          200
  C001   |        2022-05-03      |             400             |          400
  C001   |        2022-05-03      |             500             |          100
  C001   |        2022-05-03      |             800             |            0
  C001   |                        |                             |          700
---------|------------------------|-----------------------------|----------------------------------
  C002   |       2022-08-20       |           3.800             |         3.800
  C002   |       2022-08-20       |           5.000             |           200
  C002   |       2022-08-20       |           2.800             |             0
  C002   |                        |                             |         4.000

---

I want to split the received_inbound_quantity to each row of the expected_inbound_quantity until the total of the received_inbound_quantity is reached.
With reference to the answer in this question I tried to go with this solution:

SELECT
i.campaign AS campaign,
i.expected_inbound_date AS expected_inbound_date,
i.expected_inbound_quantity AS expected_inbound_quantity,
i.received_inbound_quantity AS received_inbound_quantity,

(SELECT 
   GREATEST(
     LEAST(i.expected_inbound_quantity, 
          (SELECT 
           SUM(i3.received_inbound_quantity) 
           FROM inbound i3 
           WHERE i.campaign = i3.campaign)  -
           
            (
                SELECT 
                t1.cumulated_value AS cumulated_value 
                FROM
                
                   (SELECT
                    i2.campaign, 
                    i2.expected_inbound_date, 
                    i2.expected_inbound_quantity, 
                    i2.received_inbound_quantity,
                    SUM(i2.expected_inbound_quantity) OVER (PARTITION BY i2.campaign ORDER BY i2.expected_inbound_date, i2.expected_inbound_quantity, i2.received_inbound_quantity ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING) AS cumulated_value
                    FROM inbound i2
                    GROUP BY 1,2,3,4) t1
                    
                WHERE (t1.campaign, t1.expected_inbound_date, t1.expected_inbound_quantity, t1.received_inbound_quantity) = (i.campaign, i.expected_inbound_date, i.expected_inbound_quantity, i.received_inbound_quantity)
            )
            
        ),
        0
   )
) AS split

FROM inbound i
GROUP BY 1,2,3,4
ORDER BY 1,2,3,4

---

However, in redshift I get error:

Invalid operation: This type of correlated subquery pattern is not supported yet;

How do I need to modify the query to also make it work in redshift?

Answer 1

Window functions are your friend. When you have a query that compares rows you should first look to window functions on Redshift. This simpler, cleaner, and faster than any self joining pattern.

select 
  campaign,
  expected_inbound_date,
  expected_inbound_quantity,
  received_inbound_quantity,
  case when (inbound_total - inbound_sum) >= 0 then expected_inbound_quantity
       else case when (expected_inbound_quantity + inbound_total - inbound_sum) >= 0 then expected_inbound_quantity + inbound_total - inbound_sum
                else 0 end
    end as split

from (SELECT
  campaign,
  expected_inbound_date,
  expected_inbound_quantity,
  received_inbound_quantity,
  sum(expected_inbound_quantity) over (partition by campaign order by expected_inbound_date, expected_inbound_quantity) as inbound_sum,
  max(received_inbound_quantity) over (partition by campaign) as inbound_total

  FROM inbound i
) subq
ORDER BY 1,2,3,4; 

Updated fiddle here - https://dbfiddle.uk/?rdbms=postgres_13&fiddle=2381abdf5a90a997a4f05b809c892c40

When you port this to Redshift you may want to convert the CASE statements to DECODE() functions as these are more readable IMHO.

PS. Thank you for setting up the fiddle as this greatly speeds up providing an answer.