Self-service Data Import

The import pipeline starts with a parser that auto-detects CSV delimiters, handles BOM-prefixed UTF-8, accepts both .xlsx and legacy .xls, and converts Excel serial dates to ISO strings before they reach validation. Files up to documented size land on the API where they live as a temporary artefact tied to the in-progress ImportJob.

The admin UI at /import walks operators through six steps: pick the resource type (players, licenses, clubs or memberships), upload the file, review the auto-mapped column headers, run dry-run, commit, watch status. Auto-mapping is case-insensitive and ships with common aliases — email matches "e-post", first_name matches "förnamn" or "prenom" — so a federation that exports from a legacy system rarely needs to remap by hand. Anything ambiguous surfaces as a manual mapping step.

Dry-run is the safety check. It runs both schema validation (right types, required fields, format constraints) and domain validation (license number unique within tenant, club code exists, age within license-type bounds) without writing anything. Up to 100 errors come back inline with row number and column so the operator can fix the source spreadsheet before committing. When the operator hits commit, the pipeline schedules a background job that processes the file in 500-row batches; the job updates progress per batch so the operator sees a moving counter rather than wondering whether the system is alive.

Mistakes happen, so rollback is built in. Within 60 minutes of commit the operator can roll back the whole job; the service walks the created_doc_ids array and deletes every document the job created, leaving the database as if the import never happened. After the window the rollback endpoint returns 410 Gone — beyond an hour the data is considered live and rollback would risk cascading damage to records other users have started touching.

Dedup strategy is configurable per job: skip (ignore collisions), update (overwrite the existing record from the import row) or fail (abort the whole job on the first collision). Per-resource importers (players, licenses, clubs, memberships) define their own field schemas and dedup keys — license number for licenses, email for players. Templates are downloadable through GET /import/templates/{resource} as a CSV with the right column headers and an example row, and every job persists as an ImportJob with status, mapping, validation errors, created_doc_ids and timestamps so an audit can reconstruct exactly what happened.