firestore-pipelines

Use buildPipeline(db, spec) from packages/data with a PipelineSpec containing collection, filters, search, orderBy, select, and limit. Filters map to where stages and search becomes an accent-folded regexContains, with a rowId projection added automatically for row identity.

Use pipelines for joins, grouped aggregations, server-computed fields, unnest, and regex search that classic Query lacks. Stay on classic queries when you need realtime onSnapshot updates, cursor pagination, or emulator-based testing, since pipelines are one-shot and emulator-unsupported.

Use a correlated subquery: define a variable from the outer document, then embed a nested pipeline via toScalarExpression for lookups or toArrayExpression for per-parent arrays. Correlate only through defined variables, since undefined variables are runtime errors while missing fields are silently absent.

No, pipelines never run in the emulator, including client pipelines, admin pipelines, and explain. Test client code by mocking firebase/firestore/pipelines, and test admin code with dependency-injected fetch functions stubbed in emulator suites.

Firestore Enterprise auto-creates zero indexes, so an unindexed pipeline silently full-scans and bills data scanned. Add a matching entry to firestore.indexes.json, deploy it manually, and verify with execute explainOptions analyze that indexesUsed is non-empty.

The update and delete DML stages are beta, admin-only, ignore Security Rules, cannot run in transactions, and fail on first error with partial success possible. They must be terminal stages over a where-filtered set, and a delete without a where wipes the collection.