Preparing your design workspace

Loading the problem and your latest saved attempt…

Design a Dataset Copy API

Copy a consistent dataset snapshot across workspaces with asynchronous progress, integrity verification, and atomic publication.

Problem statement

The source asks for an API that copies a dataset. For this exercise, assume a dataset is a versioned manifest of immutable objects plus schema and descriptive metadata.

An authorized client selects an exact source snapshot and destination workspace, submits an idempotent asynchronous copy, observes progress, and can retry after an unknown outcome.

The service may reuse verified destination objects but must publish the destination dataset only after every referenced object and manifest checksum is verified. Existing destination versions remain readable during a copy. Copying credentials, source audit history, and arbitrary data transformation are outside scope.

What to build

  1. Clients can submit an idempotent copy from an exact source dataset version to an authorized destination workspace.
  2. Clients can read copy state, byte and object progress, failures, and the terminal destination dataset identity.
  3. The service can resume interrupted object transfers and safely reuse content already verified at the destination.
  4. The destination snapshot becomes visible atomically only after its complete manifest and object checksums pass verification.

At scale

  • For this exercise, assume 10,000 copy requests per day and up to 1,000 concurrently active copies.
  • For this exercise, a dataset contains up to 10 million immutable objects and 100 TB of referenced bytes.
  • For this exercise, control-plane state should be readable within two seconds, while byte transfer may run for hours.
  • For this exercise, source and destination may be in different regions, and a lost response must never create two published destination versions.