asgeirrr / pgantomizer

Anonymize data in your PostgreSQL dabatase with ease

pgantomizer

Anonymize data in your PostgreSQL database with ease. Anonymization is handy if you need to provide data to people that should not have access to the personal information of the users. Importing the data to third-party tools where you cannot guarantee what will happen to the data is also a common use case. This tool will come in handy when GDPR will take effect in EU-countries.

Anonymization Process

The rules for anonymization are written in a single YAML file.

Columns that should be left in the raw form without anonymization must be explicitly marked in the schema. This ensures that adding the new column in the DB without thinking about its sensitivity does not leak the data.

The default name of the primary key is id but a custom one can be specified form the table in the schema. If the table has no primary key, you may specify ~ or null explicitly. Primary key is NOT anonymized by default.

If you wish do bypass anonymization and truncate a specific table, you can do so by passing truncate: true.

If the table you wish to truncate is referenced by foreign keys, consider using truncate: cascade, but beware that this option will cause PostgreSQL to silently and automatically truncate any other tables that reference this table via foreign key constraints. Use with care!

You can limit the scope of the anonymization pass by providing a where clause. This is useful for retaining internal data as appropriate.

A sample YAML schema can be examined below.

customer:
    raw: [language, currency]
    pk: customer_id
    custom_rules:
        email: example_email,
        bio: x_out,
        auth_hash: clear,
        another_uniq_id: md5,
        phone_number:
            value: '[phone]'
    where: "email <> '[e-mail]'"

customer_address:
    raw: [country, customer_id]
    pk: ~
    custom_rules:
        address_line: aggregate_length

customer_transactions:
    truncate: true

Sometimes it is needed to use a different anonymization function for a particular column. It can be specified in the custom_rules directive (see example above). There is a limited set of functions you can choose from. So far:

• aggregate_length - replaces content of the column with its length (can be used on any type that supports length function)
• clear - simply nulls out the value (whatever DB constraints still apply)
• example_email - replaces the value with an @example.com based on the primary key value
• md5 - alternative to default TEXT handling, useful for creating variance aside default handling while also guaranteeing value uniqueness
• x_out - converts a string alpha-numeric characters to X's, retaining length

Additionally, you can provide a nested value with a value key to assign values directly.

Calling pgantomizer from the Command Line

pgantomizer_dump is a helper script that dumps tables specified in the YAML schema file to a compressed file using pg_dump. Just pass the path to the schema and the DB connection details. Minimal working example taking advantage of default values of some of the required parameters:

pgantomizer_dump --schema my_schema.yaml --dbname original_postgres --user alaric

To see a list of all parameters, run:

pgantomizer_dump -h

The script is able to take the DB connection details from environmental variables following the conventions of running Django in Docker. The presumed variable names are: DB_DEFAULT_NAME, DB_DEFAULT_USER, DB_DEFAULT_PASS, DB_DEFAULT_SERVICE, DB_DEFAULT_PORT.

By default, the main script, pgantomizer loads the Postgre dump into a specified instance. Then all columns except primary keys and the ones specified in the schema as raw are anonymized according to their data type. Finally, the dump file is deleted by default to reduce risk of leakage of unanonymized data. The connection details of the Postgres instance where the anonymized data should be loaded can be passed as arguments

pgantomizer --schema my_schema.yaml --dump-file ./to_anonymize.sql --dbname anonymized_postgres --user alaric --password anonymized_pass --host localhost --port 5432

or through environmental variables with following names: ANONYMIZED_DB_NAME, ANONYMIZED_DB_USER, ANONYMIZED_DB_PASS, ANONYMIZED_DB_HOST, ANONYMIZED_DB_PORT.

Note: If you wish to anonymize a source that has been previously restored using other means, you may do so by passing the --skip-restore (-s) flag to pgantomizer. In this mode pgantomizer will not try to enforce any dump file requirements and will connect directly to the target server for anonymization without any schema reconstruction.

Calling pgantomizer from Python

Use dump_db and load_anonymize_remove functions to dump anonymize the data from Python. In the following example, DB connections for the original and anonymized instance are specified via ENV variables described above.

from pgantomizer import dump_db, load_anonymize_remove

dump_db('to_anonymize.sql', 'anonymization_schema.yaml')
load_anonymize_remove('to_anonymize.sql', 'anonymization_schema.yaml')

Both functions have an optional db_args argument to pass the connection arguments explicitly in a dict. See the example below how the dict should look like.

If you are only after anonymizing an existing database, there is a function anonymize_db that will help you do that with a little extra work of parsing the YAML schema.

import yaml

from pgantomizer import anonymize_db

anonymize_db(yaml.load(open('anonymization_schema.yaml'), Loader=yaml.FullLoader), {
    'dbname': 'anonymized_postgres',
    'user': 'alaric',
    'password': 'anonymized_pass',
    'host': 'localhost',
    'port': '5432',
})

If you would like to use environmental variables instead, use function anonymize.get_db_args_from_env to construct the dict from ENV.

TODO

• expand this README
• submit package automatically to PyPI
• add --dry-run argument that will check the schema and output the operations to be performed
• remove password argument and use getpass instead for better security