SQL SUM with Repeating Sub Entries - Best Practice? - sql

Question

I hit this issue regularly but here is an example....
I have a Order and Delivery Tables. Each order can have one to many Deliveries.
I need to report totals based on the Order Table but also show deliveries line by line.
I can write the SQL and associated Access Report for this with ease ....
SELECT xxx
FROM
Order
LEFT OUTER JOIN
Delivery on Delivery.OrderNO = Order.OrderNo
until I get to the summing element. I obviously only want to sum each Order once, not the 1-many times there are deliveries for that order.
e.g. The SQL might return the following based on 2 Orders (ignore the banalness of the report, this is very much simplified)
Region OrderNo Value Delivery Date
North 1 £100 12-04-2012
North 1 £100 14-04-2012
North 2 £73 01-05-2012
North 2 £73 03-05-2012
North 2 £73 07-05-2012
South 3 £50 23-04-2012
I would want to report:
Total Sales North - £173
Delivery 12-04-2012
Delivery 14-04-2012
Delivery 01-05-2012
Delivery 03-05-2012
Delivery 07-05-2012
Total Sales South - £50
Delivery 23-04-2012
The bit I'm referring to is the calculation of the £173 and £50 which the first of which obviously shouldn't be £419!
In the past I've used things like MAX (for a given Order) but that seems like a fudge.
Surely there must be a regular answer to this seemingly common problem but I can't find one.
I don't necessarily need the code - just a helpful point in the right direction.
Many thanks,
Chris.

Answer 1

A roll up operator may not look pretty. However, it would do the regular aggregates that you see now, and it show the subtotals of the order. This is what you're looking for.
SELECT xxx
FROM
Order
LEFT OUTER JOIN
Delivery on Delivery.OrderNO = Order.OrderNo
GROUP BY xxx
WITH ROLLUP;
I'm not exactly sure how the rest of your query is set up, but it would look something like this:
Region OrderNo Value Delivery Date
North 1 £100 12-04-2012
North 1 £100 14-04-2012
North 2 £73 01-05-2012
North 2 £73 03-05-2012
North 2 £73 07-05-2012
NULL NULL f419 NULL

Answer 2

I believe what you want is called a windowing function for your aggregate operation. It looks like the following:
SELECT xxx, SUM(Value) OVER (PARTITION BY Order.Region) as OrderTotal
FROM
Order
LEFT OUTER JOIN
Delivery on Delivery.OrderNO = Order.OrderNo
Here's the MSDN article. The PARTITION BY tells the SUM to be done separately for each distinct Order.Region.
Edit: I just noticed that I missed what you said about orders being counted multiple times. One thing you could do is SUM() the values before joining, as a CTE (guessing at your schema a bit):
WITH RegionOrders AS (
SELECT Region, OrderNo, SUM(Value) OVER (PARTITION BY Region) AS RegionTotal
FROM Order
)
SELECT Region, OrderNo, Value, DeliveryDate, RegionTotal
FROM RegionOrders RO
INNER JOIN Delivery D on D.OrderNo = RO.OrderNo