Registration systems look simple until they meet real infrastructure. A request can be repeated, delayed, partially completed, or accepted by one service while another is unavailable.
Design for the second attempt
Reliable registration work makes operations idempotent wherever retries are possible, keeps the user-facing state honest during asynchronous work, and logs enough correlation detail to explain a partial failure.
Cloud services provide excellent building blocks, but they do not remove the need for a clear domain model. The application still decides what accepted, pending, or failed means.
Reliability is often the quality of the recovery path.
Thinking of registration as a workflow with states made retries and recovery much easier to reason about.