Django Batch Select
The idea of Django Batch Select is to provide an equivalent to Django's select_related functionality. As of such it's another handy tool for avoiding the "n+1 query problem".
select_related is handy for minimizing the number of queries that need to be made in certain situations. However it is only usual for pre-selecting ForeignKey relations.
batch_select is handy for pre-selecting ManyToManyField relations and reverse ForeignKey relations.
It works by performing a single extra SQL query after a QuerySet has been evaluated to stitch in the the extra fields asked for. This requires the addition of a custom Manager, which in turn returns a custom QuerySet with extra methods attached.
Example Usage
Assuming we have models defined as the following:
from batch_select.models import BatchManager
class Tag(models.Model):
name = models.CharField(max_length=32)
class Section(models.Model):
name = models.CharField(max_length=32)
objects = BatchManager()
class Entry(models.Model):
title = models.CharField(max_length=255)
section = models.ForeignKey(Section, blank=True, null=True)
tags = models.ManyToManyField(Tag)
objects = BatchManager()
I'll also define a helper function to show the SQL queries generated:
from django import db
def show_queries():
for query in db.connection.queries:
print query["sql"]
db.reset_queries()
Here are a few example (with generated sql queries):
>>> Entry.objects.batch_select('tags').all()
[]
>>> show_queries() # no results, so no 2nd query
SELECT "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id" FROM "batch_select_entry"
>>> Entry.objects.create()
>>> Entry.objects.create()
>>> tag1 = Tag.objects.create(name='tag1')
>>> tag2 = Tag.objects.create(name='tag2')
>>> db.reset_queries()
>>> entries = Entry.objects.batch_select('tags').all()
>>> entry = entries[0]
>>> print entry.tags_all
[]
>>> show_queries()
SELECT "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id" FROM "batch_select_entry" LIMIT 1
SELECT (`batch_select_entry_tags`.`entry_id`) AS "entry_id", "batch_select_tag"."id", "batch_select_tag"."name" FROM "batch_select_tag" INNER JOIN "batch_select_entry_tags" ON ("batch_select_tag"."id" = "batch_select_entry_tags"."tag_id") WHERE "batch_select_entry_tags".entry_id IN (1)
>>> entry.tags.add(tag1)
>>> db.reset_queries()
>>> entries = Entry.objects.batch_select('tags').all()
>>> entry = entries[0]
>>> print entry.tags_all
[<Tag: Tag object>]
>>> show_queries()
SELECT "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id" FROM "batch_select_entry" LIMIT 1
SELECT (`batch_select_entry_tags`.`entry_id`) AS "entry_id", "batch_select_tag"."id", "batch_select_tag"."name" FROM "batch_select_tag" INNER JOIN "batch_select_entry_tags" ON ("batch_select_tag"."id" = "batch_select_entry_tags"."tag_id") WHERE "batch_select_entry_tags".entry_id IN (1)
>>> entries = Entry.objects.batch_select('tags').all()
>>> for entry in entries:
.... print entry.tags_all
....
[<Tag: Tag object>]
[]
>>> show_queries()
SELECT "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id" FROM "batch_select_entry"
SELECT (`batch_select_entry_tags`.`entry_id`) AS "entry_id", "batch_select_tag"."id", "batch_select_tag"."name" FROM "batch_select_tag" INNER JOIN "batch_select_entry_tags" ON ("batch_select_tag"."id" = "batch_select_entry_tags"."tag_id") WHERE "batch_select_entry_tags".entry_id IN (1, 2)
Re-running that same last for loop without using batch_select generate three queries instead of two (n+1 queries):
>>> entries = Entry.objects.all()
>>> for entry in entries:
.... print entry.tags.all()
....
[<Tag: Tag object>]
[]
>>> show_queries()
SELECT "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id" FROM "batch_select_entry"
SELECT "batch_select_tag"."id", "batch_select_tag"."name" FROM "batch_select_tag" INNER JOIN "batch_select_entry_tags" ON ("batch_select_tag"."id" = "batch_select_entry_tags"."tag_id") WHERE "batch_select_entry_tags"."entry_id" = 1
SELECT "batch_select_tag"."id", "batch_select_tag"."name" FROM "batch_select_tag" INNER JOIN "batch_select_entry_tags" ON ("batch_select_tag"."id" = "batch_select_entry_tags"."tag_id") WHERE "batch_select_entry_tags"."entry_id" = 2
This also works with reverse foreign keys. So for example we can get this entries that belong to each section:
>>> section1 = Section.objects.create(name='section1')
>>> section2 = Section.objects.create(name='section2')
>>> Entry.objects.create(section=section1)
>>> Entry.objects.create(section=section1)
>>> Entry.objects.create(section=section2)
>>> db.reset_queries()
>>> Section.objects.batch_select('entry_set')
[<Section: Section object>, <Section: Section object>]
>>> show_queries()
SELECT "batch_select_section"."id", "batch_select_section"."name" FROM "batch_select_section" LIMIT 21
SELECT ("batch_select_entry"."section_id") AS "__section_id", "batch_select_entry"."id", "batch_select_entry"."title", "batch_select_entry"."section_id", "batch_select_entry"."location_id" FROM "batch_select_entry" WHERE "batch_select_entry"."section_id" IN (1, 2)
Each section object in that query will have an entry_set_all field containing the relevant entries.
You need to pass batch_select the "related name" of the foreign key, in this case "entry_set". NB by default the related name for a foreign key does not actually include the _set suffix, so you can use just "entry" in this case. I have made sure that the _set suffix version also works to try and keep the API simpler.
More Advanced Usage
By default the batch fields are inserted into fields named <name>_all,
on each object. So:
Entry.objects.batch_select('tags').all()
results in the Entry instances having fields called 'tags_all'
containing the Tag objects associated with that Entry.
If you want to give the field a different name just use a keyword argument - in the same way as using the Aggregation API:
Entry.objects.batch_select(selected_tags='tags').all()
Would means the Tag objects would be assigned to fields called
'selected_tags'.
If you want to perform filtering of the related objects you will need to use a Batch object. By doing this you can pass extra keyword arguments in the same way as when using the filter method of a QuerySet:
from batch_select.models import Batch
Entry.objects.batch_select(tags_containing_blue=Batch('tags', name__contains='blue'))
Would return Entry objects with fields called 'tags_containing_name' with only those Tags whose name contains 'blue'.
In addition to filtering using keyword arguments, you can also call the following methods on a Batch object, with their effects being passed on to the underlying QuerySet object:
(Note that distinct(), values() etc are not included as they would have side-effects on how the extra query is associated with the original query) So for example to achieve the same effect as the filter above you could do the following:
from batch_select.models import Batch
Entry.objects.batch_select(tags_containing_blue=Batch('tags').filter(name__contains='blue'))
Whereas the following would exclude tags containing "blue" and order by name:
from batch_select.models import Batch
batch = Batch('tags').exclude(name__contains='blue').order_by('name')
Entry.objects.batch_select(tags_not_containing_blue=batch)
Compatibility
Django batch select should work with Django 1.1-1.3 at least.