SQL Query to Get Valid Data on Given Date

user_12726150 0
Viewed 257

I have database that tracked person salary over time like table below:

database based on

I want to query the person (based on id) salary each month to give output like the table below result query

I don't know what query to use since it needs to iterate in the salary database to check what is the valid salary for certain date. Any idea for this?

Thanks!

recursive-query datetime postgresql sql
3 Answers

Here you have all with the example data. As already shown, you need a valid-from-date ...

WITH
-- your input ...
indata(id,datevaliduntil,salary) AS (
          SELECT 1001,DATE '9999-12-31', 5000
UNION ALL SELECT 1001,DATE '2020-08-31', 4000
UNION ALL SELECT 1001,DATE '2020-04-30', 3000
)
,
-- make it almost like a slowly changing dimension
-- table - ad a valid-from-date ...
scd AS (
  SELECT
    id
  , LAG(datevaliduntil,1,DATE '1900-01-01') OVER (
      PARTITION BY id ORDER BY datevaliduntil
    ) AS datevalidfrom
  , datevaliduntil
  , salary
  FROM indata
)
,
-- the months from the example ...
months(monthend) AS (
  SELECT 
    mon::DATE - 1 AS monthend
  FROM 
  GENERATE_SERIES(
    '2020-04-01'::DATE
  , '2021-03-01'::DATE
  , INTERVAL '1 MONTH'
  ) gs(mon)
)
SELECT
  monthend
, id
, salary
FROM scd
JOIN months ON monthend >  datevalidfrom
           AND monthend <= datevaliduntil
ORDER BY 1
;
-- out   monthend  |  id  | salary 
-- out ------------+------+--------
-- out  2020-03-31 | 1001 |   3000
-- out  2020-04-30 | 1001 |   3000
-- out  2020-05-31 | 1001 |   4000
-- out  2020-06-30 | 1001 |   4000
-- out  2020-07-31 | 1001 |   4000
-- out  2020-08-31 | 1001 |   4000
-- out  2020-09-30 | 1001 |   5000
-- out  2020-10-31 | 1001 |   5000
-- out  2020-11-30 | 1001 |   5000
-- out  2020-12-31 | 1001 |   5000
-- out  2021-01-31 | 1001 |   5000
-- out  2021-02-28 | 1001 |   5000
user_67882550

This is a convenient place to use a lateral join. The following goes by the first day of the month rather than the last day -- because that is simpler to generate:

select i.id, gs.mon, s.salary
from generate_series('2019-01-01'::date, '2020-12-01'::date, interval '1 month') gs(mon) cross join
    (select distinct id from salaries) i left join lateral
    (select s.salary
     from salaries s
     where s.id = i.id and s.datevaliduntil >= gs.mon
     order by s.datevaliduntil asc
     limit 1
    ) s;

Of course, you can just subtract 1 day from each date if you want the last day.

user_11440350

I would use a lateral join, but the other way around: start from the table itself, bring the previous date with lag(), then use generate series to generate the dates in between. A little bit of additional logic is needed to adjust the end of months:

select x.date - interval '1 day' date, t.id, t.salary
from (
    select id, salary,
        datevaliduntil + interval '1 day' datevaliduntil, 
        lag(datevaliduntil, 1, datevaliduntil) 
            over(partition by id order by datevaliduntil) + interval '1 day' lag_datevaliduntil 
    from mytable t
) t
cross join lateral generate_series(
    t.lag_datevaliduntil, 
    least(t.datevaliduntil, '2021-03-01'),
    '1 month'
) x(date)

You control the overall upper bound with the literal date in the second argument to generate_series (here, you want to stop end of March 2021).

Demo on DB Fiddle:

date                |   id | salary
:------------------ | ---: | -----:
2020-04-30 00:00:00 | 1001 |   3000
2020-04-30 00:00:00 | 1001 |   4000
2020-05-31 00:00:00 | 1001 |   4000
2020-06-30 00:00:00 | 1001 |   4000
2020-07-31 00:00:00 | 1001 |   4000
2020-08-31 00:00:00 | 1001 |   4000
2020-08-31 00:00:00 | 1001 |   5000
2020-09-30 00:00:00 | 1001 |   5000
2020-10-31 00:00:00 | 1001 |   5000
2020-11-30 00:00:00 | 1001 |   5000
2020-12-31 00:00:00 | 1001 |   5000
2021-01-31 00:00:00 | 1001 |   5000
2021-02-28 00:00:00 | 1001 |   5000
user_106767160
Related