Question

I’m having some trouble with a complex query involving the following tables. Assume time is using the built-in sqlite timestamp datatype.

I am trying to return the customers whose 2nd purchase is within 4 hours of their first purchase AND if it’s within 2 hours it must be from a different store.

I’m having trouble wrapping my head around how to refer to the specific rows to compare a first purchase with a second purchase.

purchases

purchase_id | customer_id | store_id | purchase_time
     1             1            1      2009-01-27 10:00:00.0
     2             1            2      2009-01-27 10:30:00.0
     3             2            1      2009-01-27 10:00:00.0
     4             2            1      2009-01-27 10:30:00.0
     5             3            1      2009-01-27 10:00:00.0
     6             3            2      2009-01-27 16:00:00.0
     7             4            3      2009-01-27 10:00:00.0
     8             4            3      2009-01-27 13:00:00.0

stores

store_id | misc columns...
    1
    2
    3

customers

customer_id | f_name
     1         name1
     2         name2
     3         name3
     4         name4

The correct return would be name1, name4 in this case.

Answer 1

You’re going to be joining the purchase table to itself, and then selecting on one of the two criteria.

The only real trick here is to formulate the different time criteria as:

1. Purchases that were made < 2 hours at different stores.
2. Purchases that were made between 2 and 4 hours, independent of store_id.

Both of which obviously apply for the same customer_id.

So, we’ve got:

select p1.purchase_id purchase_1, 
       p2.purchase_id purchase_2,
       c.name,
       p1.customer_id customer
from purchases p1
join purchases p2 on
p1.customer_id = p2.customer_id
join customer c on c.customer_id = p1.customer_id
where p1.purchase_time < p2.purchase_time
 and (
        (
        addtime(p1.purchase_time,'2:00:00') >= p2.purchase_time
        and p1.store_id <> p2.store_id
        )
      or
        (
        addtime(p1.purchase_time,'2:00:00') < p2.purchase_time
        and addtime(p1.purchase_time,'4:00:00') >= p2.purchase_time
        )
     )

Which joins purchases to itself by customer_id, first checks that you’re comparing earlier purchases to later purchases, and then applies the two different criteria in the criteria that are ORed.

I find the time comparisons easiest to do with the addtime() and then comparing the results. Others may prefer other ways.

SQL Fiddle here: http://sqlfiddle.com/#!2/14dda/2

Results:

PURCHASE_1  PURCHASE_2        NAME        CUSTOMER
1           2                 name1       1
7           8                 name4       4

—

EDIT: Perhaps, you’d get some efficiency by moving the p1.purchase_time < p2.purchase_time up into the join clause. This might be faster with lots of data, though the execution plans for this little amount of data are identical. You’d like the optimizer to eliminate all those cases where p1.purchase_time > p2.purchase_time before doing the more expensive comparisons. But that’s somewhat beyond the basic question of ways to write this query.