Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Streaming result delivery: the driver now decodes one Arrow batch at a time and hands it to the consumer on demand, holding only the current batch in memory instead of buffering the entire result. Memory stays flat for arbitrarily large results, and abandoning a reader early cancels the server operation.
adbc.spark.headers.<NAME> database option to set arbitrary gRPC metadata headers, at parity with the connection-string behavior. Exposed in Python as DatabaseOptions.HEADER_PREFIX.
Parameter binding for timestamp, timestamp_ntz, decimal (decimal128 and decimal256), and date64 Arrow types.

Out-of-bounds write in the C-ABI option path that could panic on the standard two-call length-probe idiom used to read a string option.
Connection-string values containing + (bearer tokens, base64 secrets, JWTs) were corrupted into spaces; they are now preserved.
Unknown adbc.spark.* options were silently ignored; they now return an error so a misspelled key (for example a token under the wrong name) is not lost.
Possible panic when calling metadata methods or creating a statement after the connection was closed; these now return a clear invalid-state error.
GetObjects now honors the table-type filter, reports temporary views as TEMPORARY, and applies the column-name filter.
GetInfo now reports the Spark server version.
gRPC Unauthenticated now maps to the unauthenticated ADBC status rather than unauthorized.
uint64 bind values above the signed 64-bit maximum are rejected instead of wrapping to a negative long.
Session release on close is bounded by a timeout so close cannot hang.
The Python DBAPI layer forwards conn_kwargs so a cloned connection keeps its configuration.

Documentation corrected to match the driver: the real adbc.spark.* option keys (previously documented as adbc.spark.connect.*), the TLS key adbc.spark.tls.enabled, the Spark Connect proto pin documented as v4.1.2, and removal of options and features that were never implemented (the adbc.connection.* and adbc.rpc.result_queue_size option keys and the reattachable-execution description).
CI actions updated to their Node 24 majors.

No automatic reconnection or retry on transient failures, and no server-side statement interrupt (client-side context cancellation only).

Initial release of the Apache Arrow ADBC driver for Apache Spark Connect.

ADBC driver for Apache Spark Connect.
Go core with a C-ABI shared library exposing the standard ADBC entrypoint (AdbcDriverInit).
Python package adbc-driver-spark with a DBAPI 2.0 (PEP 249) interface.
SQL execution that returns results as Apache Arrow.
Metadata introspection for catalogs, schemas, tables, and columns.
Prepared statements with positional parameter binding.
TLS and bearer-token authentication.
Spark Connect protos pinned to v4.1.2, wire-compatible with Spark 3.5.x, 4.0.x, and 4.1.x, each exercised on every CI run against a live Spark Connect server.
Runnable examples in Python, Go, C, R, Rust, and Ruby; the Python, C, R, Rust, and Ruby examples are validated end to end in CI against a live server.