CODEWITHSUNDEEP
sqlamazon-redshift
Allen Qin
Allen Qin

Reputation: 19947

SQL: Rolling sum in the last 30 days by groups

I have a table as below:

date, custid, sales
2015-01-01, 01, 100
2015-01-10, 01, 200
2015-02-05, 01, 300
2015-03-02, 01, 400
2015-03-03, 01, 500
2015-01-01, 02, 100
2015-01-10, 02, 200
2015-02-05, 02, 300
2015-03-02, 02, 400
2015-03-03, 02, 500
...

How can I generate the rolling sum of sales of last 30 days by date and by custid.

The desired output would be:

date, custid, running_30_day_sales
2015-01-01, 01, 100
2015-01-10, 01, 300 --(100+200)
2015-02-05, 01, 500 --(200+300)
2015-03-02, 01, 700 --(300+400)
2015-03-03, 01, 1200 -- (300+400+500)
2015-01-01, 02, 100
2015-01-10, 02, 300 --(100+200)
2015-02-05, 02, 500 --(200+300)
2015-03-02, 02, 700 --(300+400)
2015-03-03, 02, 1200 -- (300+400+500)

Upvotes: 4

Views: 13269

Answers (3)

Terminator17
Terminator17

Reputation: 860

You can also use a window function to find it this way

SELECT custid, dt::date,
            SUM(sales) OVER (partition by custid ORDER BY dt
                            RANGE BETWEEN '30 days' PRECEDING AND '2 days' Following) as  sum_of_sales
            MIN(sales) OVER (partition by custid ORDER BY dt::date
                            RANGE BETWEEN '30 days' PRECEDING AND CURRENT ROW) as  minimum,
            MAX(sales) OVER (partition by custid ORDER BY dt::date
                            RANGE BETWEEN '2 days' PRECEDING AND '2 days' Following) as  maximum
      FROM atable

Upvotes: 0

Gordon Linoff
Gordon Linoff

Reputation: 1269663

Here is a trick to do this using a cumulative sum:

with t as (
      select custid, date, sales from atable
      union all
      select custid, date + interval '30 day', sales from atable
     )
select custid, date,
       sum(sum(sales)) over (partition by cust_id order by date rows between unbounded preceding and current row) as sales_30day
from t
group by custid, date;

Upvotes: 0

Vamsi Prabhala
Vamsi Prabhala

Reputation: 49260

This is one way to do it using self join. Each date is joined with all the dates whose datediff is >0 and <= 30. Thereafter, it is just a grouping operation.

select a1.custid, a1.dt, a1.sales+sum(coalesce(a2.sales,0)) total
from atable a1
left join atable a2 on a1.custid=a2.custid 
and datediff(day,a2.dt,a1.dt)<=30 and datediff(day,a2.dt,a1.dt)>0
group by a1.custid,a1.dt,a1.sales
order by 1,2

Sample Demo in Postgres

To understand it better, look at the query result of self-join using

select a1.*,a2.*
from atable a1
left join atable a2 on a1.custid=a2.custid 
and datediff(day,a1.dt,a2.dt)<=30 and datediff(day,a1.dt,a2.dt)>0

Upvotes: 8

Related Questions