Recalculation Is a Self-Continuing, Resumable Run

Context

Large workspaces need progress that survives request and worker budgets. Existing change delivery supplies the continuation seam.

Decision

Initial insertion reaches the runner, and a continuation increment requests another worker. Each batch locks the run and commits placement with progress. Failure settlement is separate and best effort; final delivery can leave a running row if settlement fails.

Alternatives Considered

Complete the entire traversal in the start request. Request time limits make this unsuitable for large workspaces, and one transaction would retain no intermediate progress.

Add a dedicated scheduler or state-machine execution. The current design uses captured continuation events and queue retry; a new engine would need a separate scheduling and recovery contract.

Track retry attempts inside the regional handler. The cited queue convention assigns attempts and redelivery to the queue rather than regional persistence.

Consequences

Small traversals can complete during the start request. Committed batches survive interruption, while an interrupted transaction rolls back its batch. Successfully failed runs stay historical; another run begins traversal again. There is no cancellation route, and failed terminal settlement needs recovery outside regional operations.