Skip to main content

What's new in ArrowTDS

Feature-oriented release highlights for ArrowTDS — the pure-native SQL Server ADBC driver that streams Apache Arrow end-to-end. This is the reader-friendly view; the full engineering detail (every fix, every internal change) lives in CHANGELOG.md.

Get it in one line — see the install guide:

curl -fsSL https://raw.githubusercontent.com/arpe-io/adbc-drivers/main/install.sh \
  | sh -s -- arrowtds --license ./arpeio_adbc.lic

RHEL 8 support & leaner defaults — September 2026 (v0.6.0)​

The shipped Linux driver now runs on RHEL 8 / Rocky 8 and every newer distro out of the box (self-contained, static OpenSSL). The default read batch size also drops from 100,000 to 32,000 rows for a lighter memory footprint — output stays byte-for-byte unchanged.

Write geospatial data, not just read it — August 2026 (v0.5.25)​

ArrowTDS already read geometry/geography columns as GeoArrow WKB — now it writes them too. On adbc_ingest, Arrow WKB cells convert client-side to SQL Server's native spatial format and bulk-load, with full geometry coverage (points through nested collections, Z/M/ZM, empties) and automatic SRID handling. See Data types.

Hardened, reproducible release builds — August 2026 (v0.5.24)​

A supply-chain and artifact-hygiene pass over the release build: the shipped binaries no longer embed build-machine paths, toolchain identity, or internal symbols, and the pipeline fails closed if any leaks back in. No driver behaviour change — output is byte-for-byte identical.

Sturdier sessions and stronger secret hygiene — August 2026 (v0.5.23)​

A security and robustness pass on the wire-parsing, authentication, and ingest paths: secrets (Entra tokens, URI passwords) are scrubbed from memory before free, a multi-statement query no longer returns a half-read session to the pool, and malformed wire metadata is rejected rather than mis-parsed.

SQL Server 2025 VECTOR type — August 2026 (v0.5.22)​

Round-trip support for the SQL Server 2025 VECTOR(n) type — read as Arrow fixed_size_list<float32>[n] and ingested straight back into a VECTOR(n) target, with no server-side cast. See Data types.

Microsoft Entra ID authentication — August 2026​

Connect to Azure SQL Database and Microsoft Fabric with Microsoft Entra ID (Azure AD) — no passwords. Four options: bring your own token, service principal, managed identity, or the default credential chain that runs unchanged from laptop to CI to production. See Connection → Entra ID.

Azure SQL "Redirect" policy — August 2026​

ArrowTDS now follows Azure SQL and Fabric's Redirect connection policy transparently, reconnecting to the routed database node — previously only Proxy worked. (Redirect needs outbound TCP on ports 11000–11999.)

A standalone driver you install by name — August 2026 (v0.5.19)​

ArrowTDS is now an independently installable ADBC driver: one-line install of a checksum-verified binary, load by name (driver="arrowtds"), and portable sqlserver:// / mssql:// / arrowtds:// connection URIs.

Faster, more reliable connections — July 2026 (v0.5.18)​

A reworked connect path: when a host resolves to several addresses they share one timeout budget, so a single dead address (a stray link-local entry on Windows, say) can no longer stall the connection.

Faster large extracts — July 2026 (v0.5.15)​

A decode-path rework removed a scaling bottleneck under high parallelism — part of an overall ~4.6× speed-up on the reference SQL Server extract benchmark.

Native geospatial, decimal, and LOB types — 2026​

geometry / geography columns come back as interoperable GeoArrow / WKB (Z/M and curved geometries preserved), and text/ntext/image LOBs, money/smallmoney, uniqueidentifier, and xml all decode natively. See Data types.

Integrated & Kerberos authentication — May 2026 (v0.4.5)​

Trusted/integrated auth with no username or password: Windows SSPI (Negotiate → Kerberos/NTLM) and POSIX Kerberos (GSSAPI/SPNEGO). See Authentication.

The native TDS engine — foundational​

From the start, ArrowTDS speaks MS-TDS directly over TCP/TLS — no ODBC, no FreeTDS. Native bulk ingest into INSERT BULK, streaming Arrow fetch as the server produces rows, and the full ADBC 1.1.0 surface.

For the complete, dated, technical history see CHANGELOG.md. To report an issue or request a licence, contact sales@arpe.io or visit https://www.arpe.io.