BigQuery emulator feature reference

The current LocalCloud source packages a native GoogleSQL Analyzer and DuckDB query engine. Supervisord manages the BigQuery process inside the LocalCloud Docker container. Running LocalCloud requires Docker; the default BigQuery process does not need a Docker-socket mount. An optional custom BigQuery image uses a companion container and requires Docker access.

Connection and endpoints

The CLI may remap host ports. The Python SDK example passes the generated REST endpoint explicitly and uses anonymous credentials. A REST client configuration does not configure the Storage API client.

Current operation contract

Feature areas to test

• GoogleSQL execution: test the exact types, expressions, joins, scripting, and error behavior used by the application.
• BQML: selected local models are implemented; model types, feature shapes, and TRANSFORM variants have documented gaps.
• Vector Search and text search: check the specific index, function, and option against the feature inventory.
• Geography: spherical functions have bounded geometry and input support, including remaining snapping, covering, and union gaps.
• Security and governance: row policies, masking, and IAM-shaped APIs require path-specific tests. Nested policy tags and taxonomy operations have recorded boundaries.
• External tables and federation: validate the exact source format and connector. An emulator fixture or stored connection does not prove a live federated query.
• Storage API and Spark: test transport, TLS, schema, read/write method, connector version, and architecture separately.

To weigh LocalCloud against standalone emulators and real BigQuery, see BigQuery emulator alternatives.

Coverage matrix

The BigQuery emulator keeps a versioned coverage matrix. It includes canonical and derived records, so its total is a count of records, not a count of distinct canonical GoogleSQL capabilities or executed tests.

“Implemented” is a matrix classification tied to the emulator’s implementation and test references; test the exact feature, input shape, and client your application uses. “Implemented with limits” rows work locally within the concrete boundaries listed below; derived rows can overlap their child features.

Capabilities implemented with documented limits

The matrix records 12 development capabilities that work locally within documented limits. Each row names the capability and the boundary recorded for it.

Unsupported and unknown records

Before relying on a capability that this page does not list, test it with your application’s queries. Unsupported and unknown states must not be reduced to cloud scale alone: syntax, model variants, external sources, collation, and individual API methods can have developer-visible gaps. A successful dry run, stored configuration, or metadata operation does not prove execution support.

How to check a capability

1. Check the required capability ID and its exact operation and input boundary.
2. Confirm the dependency identity used by the LocalCloud image; dependency HEAD can differ from a packaged payload.
3. Run the application’s query or API workflow, assert the result and errors, and test restart when persistent state matters.
4. Record client versions, platform, dependency revision, assembled-image digest, failures, and skips.
5. Validate required production behavior against Google Cloud using a separate environment.

Runtime and production boundaries

• Coverage records describe the emulator source; validate the semantics your application depends on.
• Ingestion-time partitioning (PARTITION BY _PARTITIONDATE/_PARTITIONTIME) and the _PARTITIONTIME/_PARTITIONDATE pseudocolumns are not supported: the pinned analyzer rejects the pseudocolumns in DDL
• STRUCT and non-numeric ARRAY features, TRANSFORM for PCA/DNN/autoencoder/time-series/matrix-factorization/contribution models, DNN_LINEAR_COMBINED_*, and TensorFlow/AutoML model types are refused locally
• Only points and empty values are snapped; lines, polygons and multi-part values need S2Builder IntLatLngSnapFunction, which the installed S2 bindings do not expose
• Policy tags on nested RECORD fields are refused; taxonomy import/export and cross-location replication are not served
• Bigtable external tables are not implemented; Google Sheets, HTTP(S) and local-file sources are also outside the native external-table path (see the child rows)
• TRANSFORM works for linear/logistic regression, KMEANS, boosted trees, random forests and TRANSFORM_ONLY; * REPLACE, expression.*, named preprocessing arguments and WARM_START are refused
• Production Storage Read IAM is not emulated. Managed service scale and global availability are outside the pinned client profile.
• Multi-point, multi-linestring and mixed collections, and buffers on non-point values, are refused
• Linestring inputs (BigQuery merges them into one LINESTRING) and mixed point/polygon results are refused
• Google-managed reservations, fleet scale, replication, and IAM integrations require production validation.
• Survives restart with the LocalCloud volume

Validate required production IAM, billing, quotas, reservations, concurrency, managed integrations, and recovery in Google Cloud. Local functional gaps also need validation; they are not all cloud-only infrastructure limitations.