informatics-isi-edu / deriva-mcp-core / 24969226689

90%

push

github

fix(rag): enforce per-user and per-catalog isolation in rag_search

  rag_search applied ACL filtering only to schema: results. Two gaps
  allowed cross-user or cross-catalog data to appear in scoped searches.

  data: sources (rag_index_table): source names already included user_id
  at write time, but rag_search never filtered on them at query time.
  A search scoped to hostname+catalog_id now excludes any data: result
  whose source does not match data:{hostname}:{catalog_id}:{user_id}.

  enriched: sources (rag_dataset_indexers): these are intentionally
  shared across users -- operators scope them with filters such as
  released=True. rag_search now excludes enriched: results from other
  catalogs when the search is scoped to a specific hostname+catalog_id.
  Cross-catalog bleed was the primary risk; per-user isolation for
  enriched: is deferred (per-user enrichment would be storage-expensive
  and is unnecessary when the filter already restricts to public rows).

  plugin/api.py: added is_public: bool = True to rag_dataset_indexer().
  Setting is_public=False emits a WARNING at registration time to signal
  that the enriched index will be globally shared despite ACL-sensitive
  data -- per-user enrichment is not yet implemented.

9 of 9 new or added lines in 2 files covered. (100.0%)

3164 of 3514 relevant lines covered (90.04%)

0.9 hits per line