I have the following tables
- ORDER (idOrder int, idCustomer int) [PK: idOrder]
- ORDERLINE (idOrder int, idProduct int) [PK: idOrder, idProduct]
- PRODUCT (idProduct int, rating hstore) [PK: idProduct]
In the PRODUCT table, 'rating' is a key/value column where the key is an idCustomer, and the value is an integer rating.
The query to count the orders containing a product on which the customer has given a good rating looks like this:
select count(distinct o.idOrder)
from order o, orderline l, product p
where o.idorder = l.idorder and l.idproduct = p.idproduct
and (p.rating->(o.idcust::varchar)::int) > 4;
The query plan seems correct, but this query takes forever. So I tried a different query, where I explode all the records in the hstore:
select count(distinct o.idOrder)
from order o, orderline l,
(select idproduct, skeys(p.rating) idcustomer, svals(p.rating) intrating from product) as p
where o.idorder = l.idorder and l.idproduct = p.idproduct
and o.idcustomer = p.idcustomer and p.intrating > 4;
This query takes only a few seconds. How is this possible? I assumed that exploding all values of an hstore would be quite inefficient, but it seems to be the opposite. Is it possible that I am not writing the first query correctly?