Question

This code is currently executing about 50 SQL queries:

c = Category.objects.all()

categories_w_rand_books = []

for category in c:
    r = Book.objects.filter(author__category=category).order_by('?')[:5]

    categories_w_rand_books.append((category, r))

I need to cut down the number of used queries to the minimum to speed up things and do not cause server load.

Basically, I have three models: Category, Author, Book. The Author belong to the Category (not books) and I need to get a list of all categories with 5 random books under each one.

Answer 1

If you prefer single query and are using MySQL, check the excellent link provided by @Crazyshezy in his comment.
For PostgreSQL backends, a possible query is (assuming there are non-nullable FK relationships from Book to Author and from Author to Category):

SELECT * FROM (
    SELECT book_table.*, row_number() OVER (PARTITION BY category_id ORDER BY RANDOM()) AS rn 
    FROM book_table INNER JOIN author_table ON book_table.author_id = author_table.id
) AS sq 
WHERE rn <= 5 

You could then wrap it inside a RawQuerySet to get Book instances

from collections import defaultdict
qs = Book.objects.raw("""The above sql suited for your tables...""")
collection = defaultdict(list)
for obj in qs:
    collection[obj.category_id].append(obj)

categories_w_rand_books = []
for category in c:
    categories_w_rand_books.append((category, collection[category.id]))

You may not want to run this query for each request directly w/o some caching.

Furthermore, your code generates at most 50*5=250 Books, randomly, I just wonder why because it seems too many for a single page. Are items displayed as tabs or something else? Perhaps you could reduce the counts of SQLs by doing Ajax, or simplify the requirement?

Update

To use book.author w/o triggering more than another query, try prefetch_related_objects

from django.db.models.query import prefetch_related_objects
qs = list(qs) # have to evaluate at first
prefetch_related_objects(qs, ['author'])
# now instances inside qs already contain cached author instances, and
qs[0].author # will not trigger an extra query

The above code prefetches authors in batch and fills them into the qs. This just adds another query.