Question

First, this was my first question concerning that issue : SQL request to group and sum numbers by their day of creation

Now, imagine I want to make it more complex, after previous operations and thanks to answers, I have results grouped by day, so each entries of the day are added. I have this :

1 | 200 | 2010-01-01
2 | 100 | 2010-01-01

Transformed into this :

1 | 300 | 2010-01-01

That’s already pretty good, but what if I want the amount field to increment each time ?

1 | 300 | 2010-01-01
2 | 200 | 2010-01-02

Will become :

1 | 300 | 2010-01-01
2 | 500 | 2010-01-02
x | (previous amount + this amount) | this date

The sql query I got thanks to answers to my previous question :

select sum(amount), to_char(date, 'YYYY-MM-DD') 
from mytable 
group by to_char(date, 'YYYY-MM-DD') 
order by sum(amount) ASC;

Answer 1

A running sum can be done using windowing functions:

select sum(amount) over (order by date asc), 
       to_char(date, 'YYYY-MM-DD') 
from mytable 
order by date ASC;

If you want the running sum over the aggregated result, just put it into a sub-query:

select sum(day_total) over (order by date) as running_total,
       day_total, 
       date
from (
  select sum(amount) as day_total, 
         date
  from mytable 
  group by date
) t
order by date

If the column date is actual of the data type date, then there is no reason to use to_char() on it. Btw: it’s not a terribly good idea to use reserved words (date) for column names.

The important thing here is the order by as part of the over definition.

More details in the manual: http://www.postgresql.org/docs/current/static/tutorial-window.html