SQL Join or SUM is returning too many values when working with Redshift database - sql

Question

I'm working with a Redshift database and I can't understand why my join or SUM is bringing too many values. My query is below:
SELECT
date(u.created_at) AS date,
count(distinct c.user_id) AS active_users,
sum(distinct insights.spend) AS fbcosts,
count(c.transaction_amount) AS share_shake_costs,
round(((sum(distinct insights.spend) + count(c.transaction_amount)) /
count(distinct c.user_id)),2) AS cac
FROM
dbname.users AS u
LEFT JOIN
dbname.card_transaction AS c ON c.user_id = u.id
LEFT JOIN
facebookads.insights ON date(insights.date_start) = date(u.created_at)
LEFT JOIN
dbname.card_transaction AS c2 ON date(c2.timestamp) = date(u.created_at)
WHERE
c2.vendor_transaction_description ilike '%share%'
OR c2.vendor_transaction_description ilike '%shake to win%'
GROUP BY
date
ORDER BY
1 DESC;
This query returns the following data:
If we look at 2017-02-08, we can see a total of 1298 for "share_shake_costs". However, if I run the same query just on the card_transaction table I get the following results which are correct.
The query for this second table looks like this:
SELECT
date(timestamp),
sum(transaction_amount)
FROM
dbname.card_transaction AS c2
WHERE
c2.vendor_transaction_description ilike '%share%'
OR c2.vendor_transaction_description ilike '%shake to win%'
GROUP BY
1
ORDER BY
1 DESC;
I have a feeling that I have a similar issue for my "fbcosts" column. I think it has to do with my join since the SUM should be working fine.
I'm new to Redshift and SQL so perhaps there's a better way of doing this entire query. Is there anything obvious that I'm missing?

Answer 1

It seems you have a table that contains 1:n mapping and when you join over a common clause, that number is being counted n times.
Let us say one of your tables, orders contains user_id and the total bill_amount and the other table, order_details contains the detail of the sub-items placed by that user_id.
If you do a left join, by definition, orders.user_id will join n times to order_details.user_id, where
n = total number of rows in order_details table
and would perform the aggregation (sum, count etc) n times.
+------------------+ +----------------------+
| orders | | order_details |
+------------------+ +----------------------+
|amount user_id | | user_id items |
+------------------+ +----------------------+
| 1000 123 ---------> | 123 apple |
+ +----------------------+
+-------------> | 123 guava |
| +----------------------+
v-------------> | 123 mango |
+----------------------+
select sum(amount) from orders o left join order_details od
on o.user_id = od.user_id; // result: 3000
select count(amount) from orders o left join order_details od
on o.user_id = od.user_id; // result: 3
I hope the reason for large count is clear to you now.
PS: Also, always prefer to enclose OR conditions in ().
WHERE
(c2.vendor_transaction_description ilike '%share%'
OR c2.vendor_transaction_description ilike '%shake to win%')