documentdb-full-text-search

Full-Text Search — Azure DocumentDB ($search + createSearchIndexes)

Azure DocumentDB's full-text search is driven by search indexes built with the createSearchIndexes database command and queried through the $search aggregation stage. Scoring is BM25, exposed via $meta: "searchScore". The community $text operator and { field: "text" } index type are not the DocumentDB search path.

Key syntax points that differ from many blog posts and older docs:

• Index command is createSearchIndexes (not createIndexes) — each index has a name and a definition.mappings.fields block; dynamic: false is the safe default.
• Custom analyzers live inside definition.analyzers and are referenced per-field via analyzer / searchAnalyzer.
• $search targets an index by name via index: "<name>" — the engine does not auto-pick when multiple exist.
• No count field inside $search — use a downstream { $limit: N } stage.
• No compound operator yet — query one field at a time and merge in the application (see fts-multifield-index).

Rules

• fts-create-search-index — Create a search index via runCommand({ createSearchIndexes }); use definition.mappings with dynamic: false.
• fts-basic-search — $search + text operator for BM25 keyword search; target index, project searchScore, cap with $limit.
• fts-fuzzy-search — Add fuzzy: { maxEdits: 1 } to tolerate typos; keep maxEdits small.
• fts-phrase-search — phrase operator with slop for ordered-proximity matching.
• fts-custom-analyzers — Keyword tokenizer + lowerCase + asciiFolding + edgeGram for case-insensitive prefix matching on IDs, SKUs, part numbers. Index-time vs search-time analyzer pair.
• fts-path-hierarchy — pathHierarchy tokenizer for hierarchical identifiers (BN-747-ENG-2024.05, dotted, slash paths).
• fts-multifield-index — One search index mapping multiple fields; fan-out-and-merge in the app while $search compound is unavailable.
• fts-hybrid-search — Combine BM25 and vector search (RRF) on the same collection.