github keybase email
Bulk Update lands in Django 2.2
Oct 4, 2018
3 minutes read

My work on adding bulk_update() to Django has been merged and will be released in Django 2.2! Like my filtered aggregates feature it relies heavily on the wonderfully versatile CASE statement to achieve some pretty good speedups for certain use-cases.

The development documentation gives a pretty clear example as to why it’s useful, but I’ll expand on it below a bit.

The problem

A not-so uncommon use case in Django is iterating through some models, changing a field and saving the results back to the database. For example:

# Translate our posts into German
for post in Post.objects.all():
    post.title = translate_to_german(post.title)
    post.save()

Each call to post.save() will execute a single, individual UPDATE statement, so if you’ve got a million rows to update this can be a significant slowdown. The most performant way of achieving this is to try and push the computation down into the database. For example here we tell the database to add 10,000 views to each of our posts:

# Make our posts look really popular!
Post.objects.update(view_count=F('view_count') + 10000)

For simple numerical operations or basic string manipulations this is a great way of quickly updating a lot of rows. But in our first example we want to translate our post titles to German - no databases can do that natively (not yet at least!) so you are stuck with 1 query per model.

The solution

With .bulk_update() you can now update these rows with a greatly reduced number of queries (typically 1):

posts = list(Post.objects.all())
for post in posts:
    post.title = translate_to_german(post.title)
# Save all objects in 1 query
Post.objects.bulk_update(posts, ['title'])

Yay! Much better. With a large number of rows this is much more efficient as the database has to do a lot less work.

The implementation

The implementation turned out to be pretty straightforward. We utilize the CASE statement inside an UPDATE to customize the values we update. For example:

UPDATE posts
SET title=(CASE
            WHEN id=1 THEN 'Title 1'
            WHEN id=2 THEN 'Title 2'
           END)
WHERE id IN (1, 2)

The only tricky part of this process is with Postgres, which has some strict typing rules about expressions. If you have a CASE statement where all values are NULL it is unable to determine the type of the expression and fails. So we need to add an explicit CAST:

UPDATE posts
SET title=(CASE WHEN id=1 THEN NULL
                WHEN id=2 THEN NULL
           END)::TEXT
WHERE id IN (1, 2)

Django’s expressions framework actually supported all of this already so the method is just a nice wrapper around creating the CASE expressions and handing a few corner cases.

Future work

Postgres actually supports a nicer syntax for this which is even more performant than CASE:

update posts as t set
    t.title = c.title
from (values
    ('Title 1', 1),
    ('Title 2', 2)  
) as c(title, id) 
where c.id = t.id;

However this is going to take a bit more work as we currency do not support VALUES in this way. You can follow the ticket for adding this support here if you’re interested.


Tags: django

Back to posts