You stream files containing possible PII into Cloud Storage and BigQuery and need to mask sensitive fields while preserving referential integrity for join keys. How should you use the DLP API?

Creating pseudonyms with cryptographic format-preserving tokens is the correct approach because it directly addresses both requirements: masking sensitive fields and preserving referential integrity. Format-preserving encryption (FPE) ensures that the masked data retains the original format (e.g., a 16-digit credit card number remains a 16-digit number), which is crucial for maintaining referential integrity in join keys across different datasets. The other options are less suitable: Replacing PII with cryptogenic tokens and storing non-tokenized data in a locked-down bucket doesn't fully preserve referential integrity if the token format differs significantly from the original, making joins difficult. Redacting all PII makes the data unusable for joins and analysis, as the original values are lost. Scanning every BigQuery table and masking PII found is a reactive measure and doesn't specify the masking technique, nor does it address the streaming aspect of the problem.