Join annotations in Django without raw SQL - django

Question

I have a model that has arbitrary key/value pairs (attributes) associated with it. I'd like to have the option of sorting by those dynamic attributes. Here's what I came up with:
class Item(models.Model):
pass
class Attribute(models.Model):
item = models.ForeignKey(Item, related_name='attributes')
key = models.CharField()
value = models.CharField()
def get_sorted_items():
return Item.objects.all().annotate(
first=models.select_attribute('first'),
second=models.select_attribute('second'),
).order_by('first', 'second')
def select_attribute(attribute):
return expressions.RawSQL("""
select app_attribute.value from app_attribute
where app_attribute.item_id = app_item.id
and app_attribute.key = %s""", (attribute,))
This works, but it has a bit of raw SQL in it, so it makes my co-workers wary. Is it possible to do this without raw SQL? Can I make use of Django's ORM to simplify this?
I would expect something like this to work, but it doesn't:
def get_sorted_items():
return Item.objects.all().annotate(
first=Attribute.objects.filter(key='first').values('value'),
second=Attribute.objects.filter(key='second').values('value'),
).order_by('first', 'second')

Answer 1

Approach 1
Using Djagno 1.8+ Conditional Expressions
(see also Query Expressions)
items = Item.objects.all().annotate(
first=models.Case(models.When(attribute__key='first', then=models.F('attribute__value')), default=models.Value('')),
second=models.Case(models.When(attribute__key='second', then=models.F('attribute__value')), default=models.Value(''))
).distinct()
for item in items:
print item.first, item.second
Approach 2
Using prefetch_related with custom models.Prefetch object
keys = ['first', 'second']
items = Item.objects.all().prefetch_related(
models.Prefetch('attributes',
queryset=Attribute.objects.filter(key__in=keys),
to_attr='prefetched_attrs'),
)
This way every item from the queryset will contain a list under the .prefetched_attrs attribute.
This list will contains all filtered-item-related attributes.
Now, because you want to get the attribute.value, you can implement something like this:
class Item(models.Model):
#...
def get_attribute(self, key, default=None):
try:
return next((attr.value for attr in self.prefetched_attrs if attr.key == key), default)
except AttributeError:
raise AttributeError('You didnt prefetch any attributes')
#and the usage will be:
for item in items:
print item.get_attribute('first'), item.get_attribute('second')
Some notes about the differences in using both approaches.
you have a one idea better control over the filtering process using the approach with the custom Prefetch object. The conditional-expressions approach is one idea harder to be optimized IMHO.
with prefetch_related you get the whole attribute object, not just the value you are interested in.
Django executes prefetch_related after the queryset is being evaluated, which means a second query is being executed for each clause in the prefetch_related call. On one way this can be good, because it this keeps the main queryset untouched from the filters and thus not additional clauses like .distinct() are needed.
prefetch_related always put the returned objects into a list, its not very convenient to use when you have prefetchs returning 1 element per object. So additional model methods are required in order to use with pleasure.