Restoring Dataset Provenance

This sample illustrates how a dataset producer can create a dataset and how the provenance of the dataset can be restored after loss during copying or other transformations.

You can find the implementation in restore_dataset_provenance.py.

Summary

When digital objects are copied across physical media, across cloud environments, and within a cloud environment, they typically lose provenance information such as last modified time. This can lead to severe problems in cases where the timestamps are critical provenance metadata.

For example:

• When security logs are copied, modifications and tampering may not be detectable.

• When financial datasets are copied, their revision timestamps are lost.

• Point-in-time and bitemporal data lose the revision timestamps.

This sample illustrates how such timestamps can be restored for vBase datasets after the underlying digital objects are copied to a new AWS S3 bucket.

Detailed Description

• A .env file defines the following environment variables for accessing AWS S3 service and vBase services:

  # Forwarder config
  # vBase test forwarder URL
  VBASE_FORWARDER_URL="https://test.api.vbase.com/forwarder/"
  # vBaseTest API key
  VBASE_API_KEY="YOUR_VBASE_API_KEY"
  
  # Private key for making commitments
  VBASE_COMMITMENT_SERVICE_PRIVATE_KEY="YOUR_VBASE_COMMITMENT_SERVICE_PRIVATE_KEY"
  
  # AWS Configuration
  AWS_ACCESS_KEY_ID="YOUR_AWS_ACCESS_KEY_ID"
  AWS_SECRET_ACCESS_KEY="YOUR_AWS_SECRET_ACCESS_KEY"

• Create a vBase client object using connection parameters specified in environment variables:

  vbc = VBaseClient.create_instance_from_env()

• Create an AWS S3 client object using connection parameters specified in environment variables:

  boto_client = create_s3_client_from_env()

• Create the test dataset. If this is a new dataset, this operation records that the user with the above VBASE_COMMITMENT_SERVICE_PRIVATE_KEY has created the named dataset. Dataset creation is idempotent. If this is an existing dataset, the call is ignored. Such commitments are used to validate that a given collection of user datasets is complete and mitigates Sybil attacks (https://en.wikipedia.org/wiki/Sybil_attack).

  ds = VBaseDataset(vbc, SET_NAME, VBaseIntObject)

• Add a record to the dataset. Records an object commitment for a set record. A set record commitment establishes that a dataset record with a given CID has existed at a given time for a given set.

  vbase_receipt = ds.add_record(i)

• Verify that a given set commitment exists for a given user. This will typically be called by the data consumer to verify a producer’s claims about dataset provenance.

  assert ds.verify_commitments()[0]

• Copy a folder to another folder. This could also be a copy to a different bucket, or a different storage.

  copy_s3_bucket(
      boto_client=boto_client,
      source_bucket_name=BUCKET_NAME,
      source_folder_name=FOLDER_NAME,
      destination_bucket_name=BUCKET_NAME,
      destination_folder_name=COPY_FOLDER_NAME,
  )

• Create a vBase dataset using the copied objects. These objects have lost the original timestamps.

  ds_copy = VBaseDataset(vbc, SET_NAME, VBaseIntObject)
  # Load all objects into the dataset.
  ds_copy = init_vbase_dataset_from_s3_objects(
      ds_copy, boto_client, BUCKET_NAME, COPY_FOLDER_NAME
  )

• Attempt to verify the copied objects. Since these have lost the timestamps and the original provenance information, the checks will fail:

  success, l_log = ds_copy.verify_commitments()
  assert not success
  print("Verification log:")
  for log in l_log:
      print(log)

• Fix the record timestamps using the vBase commitment information and verify the corrected provenance data:

  # Fix the timestamps.
  assert ds_copy.try_restore_timestamps_from_index()[0]
  
  print("Dataset fixed:")
  pprint.pprint(ds_copy.to_pd_object())
  
  # Verify the records again.
  assert ds_copy.verify_commitments()[0]