Contextual Anonymization transformers shouldn't be used for primary keys

[npatki]

Environment Details

• SDV version: 1.10.0 (+ SDV Enterprise 0.10.0)
• Python version: Any
• Operating System: Any

Error Description

My metadata might list a semantic, PII type as a primary key -- for example an email.

{
  "primary_key": "user_email",
  "columns": {
    "user_email": { "sdtype": "email" },
    ...
  }
}

If I am an SDV Enterprise user and my primary keys are of premium sdtypes (email, phone_number, ...), then SDV will assign the primary key column to a contextual anonymization transformer such as DomainBasedAnonymizer or AnonymizedGeoExtractor. Such transformers are not designed to keep uniqueness in mind, meaning that it may be possible that some primary keys repeat.

Solution

As a quick solution, we should ensure that we do not assign Contextual Anonymization transformers to primary keys. Instead, we should fall back on AnonymizedFaker like we do for public SDV.

(In the future, we can think about how Contextual Anonymization transformers can support uniqueness. We'd likely need to think through a different algorithm for each one.)