Cancelling an ingestion

Ingestion can be undone, but only within a specific window and only from one place.

flowchart TB
    START(["File ingested and archived"])
    Q1{"File already mapped<br/>into a dataset?"}
    CANCEL["<b>cancel</b><br/>withdraw the file"]
    REMOVED(["Archive copy removed,<br/>file marked disabled"])
    NO1(["Refused:<br/>already added to a dataset"])

    START --> Q1
    Q1 -->|"yes"| NO1
    Q1 -->|"no"| CANCEL --> REMOVED

What cancel actually does

Cancelling removes the archived copy, removes the backup copy where one is configured, and marks the file cancelled in the database. The later stages skip disabled files, so a cancelled file can no longer be verified, finalized or mapped.

Why mapping is the cutoff

Once a file belongs to a dataset, withdrawing it would leave that dataset inconsistent with what it was declared to contain. The check is explicit: ingest refuses the operation with “cannot cancel file: already added to a dataset”.

Because mapping precedes release, a dataset that has been released is necessarily past the cutoff too. But release is not what closes the window; mapping is.

Practical consequence

Map late, and only when you are confident about the contents. Everything before mapping is recoverable through cancellation; nothing after it is.

Built from f91944d