Performance

Compiled read path

TurboDRF serves list views through a compiled path that uses Django’s .values() + F() annotations instead of DRF serializers. The database returns each row as a dict, so no model instances are built and no serializer runs field by field. The gain grows with the page size.

Enabled by default for all models. Opt out with 'compiled': False in your turbodrf() config.

What it does

Instead of fetching full model objects and serializing them through DRF:

# DRF path: fetch objects → serialize field by field
books = Book.objects.select_related('author').all()
serializer = BookSerializer(books, many=True)
return Response(serializer.data)

It tells the database to return exactly the fields needed as dicts:

# Compiled path: database returns dicts directly
books = Book.objects.values('title', 'price', author_name=F('author__name'))
return Response(list(books))

A compiled row reads exactly as the serializer path renders it: values go through the DRF field for their model field (decimals, dates, times, durations), file fields render absolute URLs, binary fields base64, and permissions and nested-value scoping are the same.

The number of queries per page doesn’t depend on how many rows the page has: a COUNT, the SELECT, and one query per M2M field. @property fields are the exception to “no model instances”: they’re computed on sandbox instances (see the property sandbox), loaded in one extra query per page plus one per declared relation hop.

What stays on DRF

• Detail views (single object — overhead is negligible)
• All write operations (POST, PUT, PATCH, DELETE — validation is needed)
• Browsable API
• Models whose list fields the compiled path can’t render the way the serializer does: a reverse relation, a nested path through a reverse FK (or through an M2M below the first hop), a nested @property path, or a field type with no DRF field to render it. python manage.py turbodrf_check reports which field and why.

Fast JSON rendering

TurboDRF auto-detects the fastest available JSON library:

No configuration needed — install and it’s used automatically.

Benchmarking

How much the compiled path saves depends on your models, page size and database. Measure it with turbodrf_benchmark, which times real list requests as a given user on both paths:

python manage.py turbodrf_benchmark Book --user alice --requests 500

It prints the average and p95 time per request for each path, and the speedup. See Management Commands.