Django Anonymiser

Django app for managing / tracking Django model anonymisation.

Status

This is currently used internally only, and has not been published to PyPI - use with caution.

Background

We currently have a pattern of each model having its own anonymise method, and a management command that iterates over each model calling said method on each object. This works, but it's impossible to track - we don't know which models, and which fields on those models, are actually being anonymised, and the documentation suffers the same fate as all documentation is that is not auto-generated.

This library adopts the pattern used by the django-side-effects library of having a "registry" of anonymisers and a management command that outputs the complete listing of all anonymisers and all fields anonymised. This output can then be plugged into the django-project-checks framework and stored in the repo as a "snapshot" that is then checked in the CI pipeline, meaning it is guaranteed to be up-to-date.

The anonymisation itself doesn't change - it's just shifting the code around.

Redaction vs. Anonymisation

This library contains two flavours of anonymisation - Redaction, and Anonymisation. The two differ in how the data is overwritten:

Type	Implementation	Performance	Data
Redaction	SQL	Fast	Table level
Anonymisation	Python	Slow	Row level

Redaction

Redaction is implemented as a single SQL update statement that wipes an entire table in one go. It's very fast, but it's limited in the sense that it cannot produce realistic data. In fact it may well render your application unusable. It is recommended as the first step in data anonymisation.

Anonymisation

Anonymisation is an row-level operation that iterates over a queryset and updates each object in turn. The main advantage is that post-anonymisation you will have realistic, usable, data.

Usage

As an example - this is a hypothetical User model's anonymisation today:

# models.py
class User:

    def anonymise(self) -> None:
        self.first_name = "Fred"
        self.last_name = "Flinstone"

Using this library we remove the anonymise method and create and register a new anonymiser that splits out each field:

# anonymisers.py
@register_anonymiser
class UserAnonymiser(ModelAnonymiser):
    model = User

    def anonymise_first_name(self, obj: User) -> None:
        obj.first_name = "Fred"

    def anonymise_last_name(self, obj: User) -> None:
        obj.last_name = "Flintstone"

You should import the anonymisers module in your apps.py in order to ensure that it is registered:

# apps.py
from django.apps import AppConfig


class UsersConfig(AppConfig):
    name = "Users"

    def ready(self) -> None:
        super().ready()
        from . import anonymisers  # noqa F401

Once set up, running the display_model_anonymisation management command will output a list of all models in the project, whether they have a registered anonymiser, and then all model fields in the project and whether they are anonymised.

The snapshot for this project itself is tests/model_anonymisation.md.

The output format of the snapshot can be overridden - it's rendered using a Django template templates/display_model_anonymisation.md.