Skip to content

fix(doc-repo): case-insensitive re-ingest/retract status guard (#831)

Jérôme Revillard requested to merge fix/831-file-labels-delete-old into main

Root cause

dataprep persists dataprep.status capitalized ('Ingested', 'Retracted') via _update_doc_status, and the /status Joi schema only accepts capitalized values. But the doc-repo ingest/retract guards compared against lowercase ('ingested' / 'retracted'), so the ingest guard never matched an already-ingested file.

Because re-ingestion was never blocked, ingest_file_with_guardrail re-inserted chunks for an already-ingested file_id without retracting the old ones first. Old chunks (with old labels) therefore remained and the duplicate insertion raised _key conflicts → "changing the file labels doesn't delete the old ones, leading to errors in ingesting".

Fix

Compare the status case-insensitively so both guards fire regardless of stored casing:

  • fileController.js:977 — ingest guard (the bug)
  • fileController.js:1060 — retract guard (same root cause, prevents recurrence)

Blocking behavior preserved. To apply new file labels to chunks: retract → re-ingest.

Verification

  • ESLint: clean
  • Prettier: clean
  • doc-repo suite: 402/402 pass, 17/17 suites
  • New unit tests cover the canonical capitalized status ('Ingested') and the case-insensitive path

Closes #831 (closed)

Merge request reports