Rolling total with no sub-select and no vendor specific extensions - sql

Question

What I'm trying to achieve: rolling total for quantity and amount for a given day, grouped by hour.
It's easy in most cases, but if you have some additional columns (dir and product in my case) and you don't want to group/filter on them, that's a problem.
I know there are extensions in Oracle and MSSQL specifically for that, and there's SELECT OVER PARTITION in Postgres.
At the moment I'm working on an app prototype, and it's backed by MySQL, and I have no idea what it will be using in production, so I'm trying to avoid vendor lock-in.
The entrire table:
> SELECT id, dir, product, date, hour, quantity, amount FROM sales
ORDER BY date, hour;
+------+-----+---------+------------+------+----------+--------+
| id | dir | product | date | hour | quantity | amount |
+------+-----+---------+------------+------+----------+--------+
| 2230 | 65 | ABCDEDF | 2014-09-11 | 1 | 1 | 10 |
| 2231 | 64 | ABCDEDF | 2014-09-11 | 3 | 4 | 40 |
| 2232 | 64 | ABCDEDF | 2014-09-11 | 5 | 5 | 50 |
| 2235 | 64 | ZZ | 2014-09-11 | 7 | 6 | 60 |
| 2233 | 64 | ABCDEDF | 2014-09-11 | 7 | 6 | 60 |
| 2237 | 66 | ABCDEDF | 2014-09-11 | 7 | 6 | 60 |
| 2234 | 64 | ZZ | 2014-09-18 | 3 | 1 | 11 |
| 2236 | 66 | ABCDEDF | 2014-09-18 | 3 | 1 | 100 |
| 2227 | 64 | ABCDEDF | 2014-09-18 | 3 | 1 | 100 |
| 2228 | 64 | ABCDEDF | 2014-09-18 | 5 | 2 | 200 |
| 2229 | 64 | ABCDEDF | 2014-09-18 | 7 | 3 | 300 |
+------+-----+---------+------------+------+----------+--------+
For a given date:
> SELECT id, dir, product, date, hour, quantity, amount FROM sales
WHERE date = '2014-09-18'
ORDER BY hour;
+------+-----+---------+------------+------+----------+--------+
| id | dir | product | date | hour | quantity | amount |
+------+-----+---------+------------+------+----------+--------+
| 2227 | 64 | ABCDEDF | 2014-09-18 | 3 | 1 | 100 |
| 2236 | 66 | ABCDEDF | 2014-09-18 | 3 | 1 | 100 |
| 2234 | 64 | ZZ | 2014-09-18 | 3 | 1 | 11 |
| 2228 | 64 | ABCDEDF | 2014-09-18 | 5 | 2 | 200 |
| 2229 | 64 | ABCDEDF | 2014-09-18 | 7 | 3 | 300 |
+------+-----+---------+------------+------+----------+--------+
The results that I need, using sub-select:
> SELECT date, hour, SUM(quantity),
( SELECT SUM(quantity) FROM sales s2
WHERE s2.hour <= s1.hour AND s2.date = s1.date
) AS total
FROM sales s1
WHERE s1.date = '2014-09-18'
GROUP by date, hour;
+------------+------+---------------+-------+
| date | hour | sum(quantity) | total |
+------------+------+---------------+-------+
| 2014-09-18 | 3 | 3 | 3 |
| 2014-09-18 | 5 | 2 | 5 |
| 2014-09-18 | 7 | 3 | 8 |
+------------+------+---------------+-------+
My concerns for using sub-select:
once there are round million records in the table, the query may become too slow, not sure if it's subject to optimizations even though it has no HAVING statements.
if I had to filter on a product or dir, I will have to put those conditions to both main SELECT and sub-SELECT too (WHERE product = / WHERE dir =).
sub-select only counts a single sum, while I need two of them (sum(quantity) и sum(amount)) (ERROR 1241 (21000): Operand should contain 1 column(s)).
The closest result I were able to get using JOIN:
> SELECT DISTINCT(s1.hour) AS ih, s2.date, s2.hour, s2.quantity, s2.amount, s2.id
FROM sales s1
JOIN sales s2 ON s2.date = s1.date AND s2.hour <= s1.hour
WHERE s1.date = '2014-09-18'
ORDER by ih;
+----+------------+------+----------+--------+------+
| ih | date | hour | quantity | amount | id |
+----+------------+------+----------+--------+------+
| 3 | 2014-09-18 | 3 | 1 | 100 | 2236 |
| 3 | 2014-09-18 | 3 | 1 | 100 | 2227 |
| 3 | 2014-09-18 | 3 | 1 | 11 | 2234 |
| 5 | 2014-09-18 | 3 | 1 | 100 | 2236 |
| 5 | 2014-09-18 | 3 | 1 | 100 | 2227 |
| 5 | 2014-09-18 | 5 | 2 | 200 | 2228 |
| 5 | 2014-09-18 | 3 | 1 | 11 | 2234 |
| 7 | 2014-09-18 | 3 | 1 | 100 | 2236 |
| 7 | 2014-09-18 | 3 | 1 | 100 | 2227 |
| 7 | 2014-09-18 | 5 | 2 | 200 | 2228 |
| 7 | 2014-09-18 | 7 | 3 | 300 | 2229 |
| 7 | 2014-09-18 | 3 | 1 | 11 | 2234 |
+----+------------+------+----------+--------+------+
I could stop here and just use those results to group by ih (hour), calculate the sum for quantity and amount and be happy. But something eats me up telling that this is wrong.
If I remove DISTINCT most rows become to be duplicated. Replacing JOIN with its invariants doesn't help.
Once I remove s2.id from statement you get a complete mess with disappearing/collapsion meaningful rows (e.g. ids 2236/2227 got collapsed):
> SELECT DISTINCT(s1.hour) AS ih, s2.date, s2.hour, s2.quantity, s2.amount
FROM sales s1
JOIN sales s2 ON s2.date = s1.date AND s2.hour <= s1.hour
WHERE s1.date = '2014-09-18'
ORDER by ih;
+----+------------+------+----------+--------+
| ih | date | hour | quantity | amount |
+----+------------+------+----------+--------+
| 3 | 2014-09-18 | 3 | 1 | 100 |
| 3 | 2014-09-18 | 3 | 1 | 11 |
| 5 | 2014-09-18 | 3 | 1 | 100 |
| 5 | 2014-09-18 | 5 | 2 | 200 |
| 5 | 2014-09-18 | 3 | 1 | 11 |
| 7 | 2014-09-18 | 3 | 1 | 100 |
| 7 | 2014-09-18 | 5 | 2 | 200 |
| 7 | 2014-09-18 | 7 | 3 | 300 |
| 7 | 2014-09-18 | 3 | 1 | 11 |
+----+------------+------+----------+--------+
Summing doesn't help, and it adds up to the mess.
First row (hour = 3) should have SUM(s2.quantity) equal 3, but it has 9. What does SUM(s1.quantity) shows is a complete mystery to me.
> SELECT DISTINCT(s1.hour) AS hour, sum(s1.quantity), s2.date, SUM(s2.quantity)
FROM sales s1 JOIN sales s2 ON s2.date = s1.date AND s2.hour <= s1.hour
WHERE s1.date = '2014-09-18'
GROUP BY hour;
+------+------------------+------------+------------------+
| hour | sum(s1.quantity) | date | sum(s2.quantity) |
+------+------------------+------------+------------------+
| 3 | 9 | 2014-09-18 | 9 |
| 5 | 8 | 2014-09-18 | 5 |
| 7 | 15 | 2014-09-18 | 8 |
+------+------------------+------------+------------------+
Bonus points/boss level:
I also need a column that will show total_reference, the same rolling total for the same periods for a different date (e.g. 2014-09-11).

Answer 1

If you want a cumulative sum in MySQL, the most efficient way is to use variables:
SELECT date, hour,
(#q := q + #q) as cumeq, (#a := a + #a) as cumea
FROM (SELECT date, hour, SUM(quantity) as q, SUM(amount) as a
FROM sales s
WHERE s.date = '2014-09-18'
GROUP by date, hour
) dh cross join
(select #q := 0, #a := 0) vars
ORDER BY date, hour;
If you are planning on working with databases such as Oracle, SQL Server, and Postgres, then you should use a database more similar in functionality and that supports that ANSI standard window functions. The right way to do this is with window functions, but MySQL doesn't support those. Postgres, SQL Server, and Oracle all have free versions that yo can use for development purposes.
Also, with proper indexing, you shouldn't have a problem with the subquery approach, even on large tables.