select_related() and RawQuerySet
Tom Evans
I have a particular query that requires me to use a RawQuerySet - I
need to left join the table to itself in order to select the highest
'priority' row from the table for each distinct value of a foreign key
on the model, and I need to join to the related table in order to
filter the results.
Having generated the queryset, I then want to make a dictionary of {
foreign_key_object : model_object }. So I have this working, but I
cannot use select_related() with a RawQuerySet, and so this runs N+1
queries, where N is the number of distinct foreign keys.
Here is some code, which might explain it better:
connection.queries=[]
base_products_qs = Product.objects.raw(
"""
SELECT idp_product.*, idp_productclass.*
FROM idp_product
JOIN idp_productclass
ON idp_product.product_class_id = idp_productclass.id
LEFT JOIN idp_product p2
ON idp_product.product_class_id = p2.product_class_id
AND p2.class_priority < idp_product.class_priority
WHERE p2.id IS NULL and idp_productclass.product_type != 4
""")
base_products = dict([ (p.product_class, p) for p in base_products_qs ])
len(connection.queries) # 7 queries (6 product classes)
Is there any simple way around this? I can reduce it to two queries
already, but it seems wrong to select out the info I want, throw it
away, and then fetch it again.
two query version:
base_products_qs = Product.objects.raw(
"""
SELECT idp_product.*
FROM idp_product
JOIN idp_productclass
ON idp_product.product_class_id = idp_productclass.id
LEFT JOIN idp_product p2
ON idp_product.product_class_id = p2.product_class_id
AND p2.class_priority < idp_product.class_priority
WHERE p2.id IS NULL and idp_productclass.product_type != 4
""")
base_products_ = dict([ (p.product_class_id, p) for p in base_products_qs ])
prod_classes_qs = ProductClass.objects.filter(pk__in=base_products_.keys())
prod_classes = dict([ (pc.id, pc) for pc in prod_classes_qs ])
base_products = dict([ (prod_classes.get(k), v)
for k, v in base_products_.items() ])
Cheers for any pointers
Tom
akaariai
- If the 2-query version is fast enough, use it. Premature optimization and all that... :)
- You could of course manually iterate through the SQL and instantiate the models by hand.
- Beware of the left join, it seems if you have a lot of rows with different class_priorities for each product_class_id you might be in trouble. The filtering on IS NULL happens after doing the join, so before the filter you could have a lot of intermediate rows. It might be there is no problem depending of the amount of rows and SQL vendor. Hard to know without testing. There are a couple of other ways to do the query, either by subquery doing a select min(class_priority), product_class_id or if you are using PostgreSQL, distinct on would do the job for you, too.
All that being said, I would really like the ability to have select_related available for RawQuerySet. It might be judged to be a feature not needed commonly enough, so it is possible it would not get in even with a good quality patch. However, if it had a nice API, and the implementation reused the code of QuerySet select_related, I think it could have a chance to get in.
- Anssi
