Using psql, how do you get the total sum for last 3 days, on each day? - psql

Question

I have a table that contains all purchases made at each school. I’m able to get the total spent per school, per item, per day,
with the following.
SELECT
date
school_id,
item_id,
sum(price) as total_price
FROM purchases
GROUP BY school_id, item_id, date
ORDER BY school_id, date
It will return something like
date school_id item_id total_price
2016-11-18 | 1 | 1 | 0.50
2016-11-17 | 1 | 2 | 1.00
2016-11-16 | 1 | 1 | 0.50
2016-11-18 | 2 | 2 | 1.00
2016-11-17 | 2 | 2 | 1.00
2016-11-16 | 2 | 2 | 1.00
I need a table that returns the total price for the last 3 days (including the day of) on each day,
So something like
date school_id item_id total_price
2016-11-18 | 1 | 1 | 1.00
2016-11-17 | 1 | 2 | 1.00
2016-11-16 | 1 | 1 | 0.50
2016-11-18 | 2 | 2 | 3.00
2016-11-17 | 2 | 2 | 2.00
2016-11-16 | 2 | 2 | 1.00
I know I can use lag() OVER (PARTITION BY), but I may need to do this for months at the time instead of 3 days, and lag will take forever to get set up.
I’m not really sure what other method I can use. Any guidance?

Answer 1

A simple INNER JOIN would do
You join the table to itself, when the school and item match, and the date is 3 days range
Notice that this would give a moving average of the last 3 days, but it seems so from your question, since you want consecutive days, without jumps
SELECT
p1.date
p1.school_id,
p1.item_id,
SUM(p2.price) total_price_3_days
purchases p1
INNER JOIN purchases p2 ON p1.school_id = p2.school_id AND p1.item_id = p2.item_id AND p2.`date` BETWEEN DATE_SUB(p1.`date`, INTERVAL 3 DAY) AND p1.`date`
GROUP BY p1.school_id, p1.item_id, p1.date
ORDER BY p1.school_id, p1.date

Answer 2

One approach would be to just use a correlated subquery in the select clause:
SELECT
date,
school_id,
item_id,
(SELECT SUM(p2.price) FROM purchases p2
WHERE p1.school_id = p2.school_id AND
p2.date BETWEEN p1.date - INTERVAL '3 DAY' AND p1.date) AS total_price
FROM purchases p1
GROUP BY school_id, item_id, date
ORDER BY school_id, date DESC;
Demo
Another approach would be to take advantage of Postgres' window functions:
SELECT
date,
school_id,
item_id,
SUM(price) OVER (PARTITION BY school_id
ORDER BY date ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) AS total_price
FROM purchases p1
GROUP BY school_id, item_id, date
ORDER BY school_id, date DESC;
Demo
Both generate this output:
Note that my school_id=1 output does not agree with your expected output, but I think your expected data has a typo.