Question

Here is the scenario we are working on here. We have a customer table and a sales table. The tables join via the transaction ID in cust table.

A customer can buy any fruit from the store.

What we need to do is find out for example: how many customer bought apples before they bought cherries.

Table structure:

Cust - Cust ID, Transaction ID, ...
Sales - Transaction ID, Fruit ID, Insert date for record, ...

What would be the most cost efficient way of achieving this considering that a customer can have different transaction ID for same fruit ID as he might have bought the fruit multiple times.

So we need to find, when is the 1st apple and 1st cherry bought by cust then check the dates on them.

Answer 1

I second Gary’s opinion of the data model, but that is irrelevant to the question.

This is one possible solution. It is likely to be pretty efficient if there are lots of potential values for FRUIT_ID and that column is indexed.

select apple.cust_id
from
    ( select c.cust_id, min(s.sale_date) as sale_date
      from cust c
           join sales s 
                on s.transaction_id = c.transaction_id 
      where s.fruit_id = 'CHERRY'
      group by c.cust_id ) cherry
    ,     
    ( select c.cust_id, min(s.sale_date) as sale_date
      from cust c
           join sales s 
                on s.transaction_id = c.transaction_id 
      where s.fruit_id = 'APPLE'
      group by c.cust_id ) apple
where cherry.cust_id = apple.cust_id
and cherry.sale_date > apple.sale_date
/

If there are fewer values of FRUIT_ID then a variation on Gary’s suggestion would probably be more efficient:

select cust_id
from
    ( select c.cust_id
             , min(case when s.fruit_id = 'CHERRY' = s.sale_date else null end) as cherry_date
             , min(case when s.fruit_id = 'APPLE' = s.sale_date else null end) as apple_date
      from cust c
           join sales s 
                on s.transaction_id = c.transaction_id 
      group by c.cust_id ) cherry
where cherry_date > apple_date
/

---

Caveat lector: I currently don’t have access to a database so these statements are untested and may have syntax errors. I’ll check them when I can.