SQL joining two tables with different levels of details - sql

Question

So I have two tables of sales, budget and actual.
"budget" has two columns: location and sales. For example,
location sales
24 $20000
36 $100300
40 $24700
Total $145000
"actual" has three columns: invoice_number, location, and sales. For example,
invoice location sales
10000 36 $5000
10001 40 $6000
10002 99 $7000
and so forth
Total $110000
In summary, "actual" records transactions at the invoice level, whereas "budget" is done at the location level only (no individual invoices).
I'm trying to create a summary table that lists actual and budget sales side by side, grouped by location. The total of the actual column should be $110000, and $145000 for budget. This is my attempt at it (on pgAdmin/ postgresql):
SELECT actual.location, SUM(actual.sales) AS actual_sales, SUM(budget.sales) AS budget_sales
FROM actual LEFT JOIN budget
ON actual.location = budget.location
GROUP BY actual.location;
I used LEFT JOIN because "actual" has locations that "budget" doesn't have (e.g. location 99).
I ended up with some gigantic numbers ($millions) on both the actual_sales and budget_sales columns, far exceeding the total actual ($110000) or budget sales ($145,000).
Is this because the way I wrote my query is basically asking SQL to join each invoice in "actual" to each line in "budget," therefore duplicating things many times over? If so how should I have written this?
Thanks in advance!

Answer 1

Based on your description, you seem to have duplicates in both tables. There are various ways to solve this problem. Here is one using union all and group by:
select Location,
sum(actual_sales) as actual_sales,
sum(budget_sales) as budget_sales
from ((select a.location, a.sales as actual_sales, null as budget_sales
from actual a
) union all
(select b.location, null, b.sales
from budget b
)
) ab
group by location;
This structure guarantees that each value is counted only once, regardless of the table.

Answer 2

The query looks fine to me. However, it is difficult to find out why the figures are wrong. My suggestion is that you do the sum by location separately for budget and actual into 2 temporary tables, and later put them together using LEFT JOIN.

Answer 3

Yes, you're joining the budget in once for each actual sales row. However, your Actual Sales sum shouldn't have been larger unless there were multiple budget rows for the same location. You should check for that, because it doesn't sound like there should be.
What you need to do in a case like this is sum the actual sales first in a CTE or subquery, then later join the result to the budget. That way you only have one row for each location. This does it for the actual sales. If you really do have more than one row for a location for budget as well, you might need to subquery the budget as well the same way.
Select Act.Location, Act.actual_sales, budget.sales as budget_sales
From
(
SELECT actual.location, SUM(actual.sales) AS actual_sales
FROM actual
GROUP BY actual.location
) Act
left join budget on Act.location = budget.location

Answer 4

Gordon's suggestion is good, an alternative using WITH statements is:
WITH aloc AS (
SELECT location, SUM(sales) FROM actual GROUP BY 1
), bloc AS (
SELECT location, SUM(sales) FROM budget GROUP BY 1
)
SELECT location, a.sum AS actual_sales, b.sum AS budget_sales
FROM aloc a LEFT JOIN bloc b USING (location)
This is equivalent to:
SELECT location, a.sum AS actual_sales, b.sum AS budget_sales
FROM (SELECT location, SUM(sales) FROM actual GROUP BY 1) a LEFT JOIN
(SELECT location, SUM(sales) FROM budget GROUP BY 1) b USING (location)
but I find WITH statements more readable.
The purpose of the subqueries is to get tables into a state where a row means something relevant, i.e. aloc contains a row per location, and hence cause the join to evaluate to what you want.