mssql-python 1.9.0: Row-friendly Bulk Copy, smarter NULL parameters, and a more portable wheel
We just shipped mssql-python 1.9.0, the official Microsoft SQL driver for Python. This release focuses on day-to-day ergonomics for data loading, a long-standing correctness gap around NULL parameter binding, and a build change that makes the published wheels work on clean macOS and Linux machines. pip install --upgrade mssql-python Highlights Bulk Copy now accepts Row objects (and lists) Bulk Copy previously required tuples. In 1.9.0 you can hand it Row objects straight from a SELECT, or plain lists, and the driver converts each row to a tuple internally before passing data to the Rust backend. rows = source_cursor.execute( "SELECT id, display_name, created_at FROM users" ).fetchall() target_cursor.bulkcopy("staging.users", rows) The common "fetch from one table, bulk-insert into another" pattern just works, with no manual row reshaping and no type errors on the boundary. NULL parameters now resolve to the right SQL type When you bound None to a parameter, the driver used to fall back to SQL_VARCHAR. That was fine for character columns and quietly wrong for everything else, especially VARBINARY and all-NULL columns where the server had no other type signal to lean on. 1.9.0 adds a thread-safe per-statement cache for SQLDescribeParam results, so NULL parameters resolve to the parameter's actual declared type. The cache is invalidated when a new statement is prepared, which also cuts redundant round-trips to the server during repeated executions of the same prepared statement. simdutf is now statically linked into the extension The published wheels (especially the macOS universal2 wheel) previously dynamically linked simdutf against a path that only existed on the CI build machine. On a clean install, import mssql_python could fail with missing-symbol or dlopen errors. The build no longer calls find_package(simdutf). FetchContent is used unconditionally, simdutf is built as a static library, and its symbols are embedded directly in the extension. There is nothing for end users to do; reinstall the wheel and the import works on a clean machine. Thanks to @edgarrmondragon for the contribution. Bug fixes worth calling out executemany with large Decimal values. Batch inserts of Decimal values larger than the SQL Server MONEY range raised an SQL_C_NUMERIC type-mismatch at runtime. executemany now binds DECIMAL / NUMERIC parameters as SQL_C_CHAR and sizes the column to fit the longest string representation, so large-decimal batches (including NULLs and multi-column inserts) succeed. Exception pickle round-trips. ConnectionStringParseError and the DB-API exception subclasses now implement __reduce__, so driver exceptions survive pickle / copy.deepcopy with every attribute intact. multiprocessing, distributed task queues, and anything that ships exceptions across process boundaries no longer lose context on the way through. nextset() and PRINT output. nextset() now collects diagnostic messages whenever SQL returns SQL_SUCCESS_WITH_INFO. Previously, PRINT output from secondary result sets in multi-statement batches and stored procedures was silently dropped after the first set. executemany data-at-execution path with Row objects. The DAE fallback used by large columns such as varchar(max) only recognized primitive types, so passing Row objects in that path failed. The fallback now converts Row to a tuple before mapping types. Fetch methods now type-check under ty. Catalog and metadata result-set handling no longer monkey-patches fetchone, fetchmany, and fetchall as instance attributes. A cached _column_map is built instead, so the fetch APIs stay proper class methods and static type checkers like ty stop tripping on cursor fetch calls. Upgrading For most users, pip install --upgrade mssql-python is all you need. If you had any workarounds for NULL parameter typing (manually setting inputsizes, casting in SQL, or sending sentinel values), you can drop it. If you were converting fetched rows to tuples before handing them to bulkcopy or executemany, that's no longer required either. Thanks Thanks to everyone who filed issues, sent repros, and reviewed PRs in this cycle. The NULL parameter fix, the Row handling in Bulk Copy and executemany, and the simdutf linking change all came directly from user-reported scenarios, and the simdutf change was a community contribution from @edgarrmondragon. Full changelog and PR list: microsoft/mssql-python releases.mssql-python 1.8.0: friendlier Row access, Bulk Copy with MSI, and a refreshed ODBC driver
We just shipped mssql-python 1.8.0, the official Microsoft SQL Server driver for Python. This release focuses on day-to-day ergonomics, a long-requested authentication option for Bulk Copy, and a refresh of the bundled ODBC driver. pip install --upgrade mssql-python Highlights Row objects now support string-key indexing Row already supported integer indexing and attribute access. In 1.8.0 you can also index by column name, which is the pattern most users reach for first: cursor.execute("SELECT id, display_name FROM users WHERE id = ?", [42]) row = cursor.fetchone() row[0] # 42 - by index row.display_name # 'Ada' - by attribute row["display_name"] # 'Ada' - new in 1.8.0 If you set cursor.lowercase = True, string-key lookups become case-insensitive too, matching the casing behavior of attribute access. ActiveDirectoryMSI authentication for Bulk Copy Bulk Copy operations now support Authentication=ActiveDirectoryMSI, so workloads running with a managed identity (Azure VMs, App Service, Functions, Container Apps, AKS) can stream bulk inserts to Azure SQL without provisioning a separate credential. Bundled ODBC driver upgraded to 18.6.2.1 We've upgraded the bundled Microsoft ODBC Driver for SQL Server from 18.5.1.1 to 18.6.2.1. You pick up the upstream fixes and TLS/cert improvements just by upgrading the wheel. No separate ODBC install step. Bug fixes worth calling out Deferred connect-attribute use-after-free. Values passed to connection attributes set before connect are now stored in member buffers, so the driver no longer reads freed memory in some attribute-set paths. Connection string parsed multiple times in the auth path. The auth code path was reparsing the connection string several times per connect. It now operates on the already-parsed parameter dictionary, which is both faster and easier to reason about. Sensitive parameters (UID, PWD, Trusted_Connection, Authentication) are sanitized through a single canonical path before the ODBC handoff. executemany type annotation regression. seq_of_parameters now uses a covariant Sequence, so passing a list of tuples (or any covariant sequence type) type-checks cleanly again. Upgrading For most users, pip install --upgrade mssql-python is all you need. Nothing in this release is intentionally breaking. If you have type stubs or mypy pins that depended on the previous invariant seq_of_parameters annotation, you can drop the workaround. Thanks Thanks to everyone who filed issues, sent repros, and reviewed PRs in this cycle. The Row indexing and connection-string parsing work came directly from user-reported scenarios. Full changelog and PR list: microsoft/mssql-python releases.Announcing the release of Microsoft OLE DB Driver 19.4.2 for SQL Server
We're pleased to announce the general availability (GA) release of Microsoft OLE DB Driver 19.4.2 for SQL Server. This is a maintenance update to the 19.4 line that improves TLS handling, raises a long-standing connection redirection limit, refreshes the bundled authentication library, and fixes accessibility issues in the UDL dialog. Download x64 / Arm64 installer x86 installer The 64-bit MSI continues to install the appropriate binary (x64 or Arm64) automatically based on the OS platform. MSOLEDBSQL 19 installs side by side with MSOLEDBSQL 18. Prerequisite: the Microsoft Visual C++ Redistributable must be installed before running the driver installer. What's new in 19.4.2 Features added Feature Details SSL/TLS improvements Improved SSL/TLS handling for better security and performance. 10 redirects per connection The driver no longer caps connection redirections at 2, which removes a long-standing failure mode for workloads that hit redirection-heavy topologies (for example, certain Azure SQL routing scenarios). Bugs fixed Fix Details UDL accessibility defects Fixed accessibility issues in the Universal Data Link (UDL) dialog. Updated authentication library Microsoft SQL Driver Authentication library (mssql-auth.dll) updated to version 1.1.3. mssql-auth.dll is the ADAL replacement introduced in 19.4.1 and is installed as part of the driver setup. Supported languages 19.4.2 is available in: Chinese (Simplified), Chinese (Traditional), Czech, English (United States), French, German, Italian, Japanese, Korean, Polish, Portuguese (Brazil), Russian, Spanish, and Turkish. Direct per-language download links are available in the Release notes and the Download page. Upgrading If you're on 19.4.1 or earlier in the 19.x line, an in-place upgrade is supported. As always, we recommend validating in a non-production environment first, especially if your application depends on the redirection or TLS behaviors touched in this release. Feedback Please file issues and feature requests on the Microsoft OLE DB Driver for SQL Server feedback site or via your usual support channel. Thanks to everyone who reported the UDL accessibility and redirection-limit issues that drove this release.mssql-python 1.7.1 and the case of the missing megabytes
We just shipped mssql-python 1.7.1 to PyPI. This release is mostly polish, but the road to publishing it was bumpier than expected, and there is some housekeeping on the PyPI side worth talking about. What's in 1.7.1 A handful of correctness, performance, and platform fixes that landed since 1.6.0: PERF: UTF-16 string handling now goes through simdutf and std::u16string, cutting overhead on every string round-trip between Python and the driver. PERF: Faster execute() hot path with soft reset, prepare caching, and guarded diagnostics. FIX: Login failures now raise a proper mssql_python exception instead of a raw RuntimeError (issue #532). FIX: Released the GIL around blocking SQLSetConnectAttr calls so that in-process SSH tunnels (paramiko + sshtunnel) no longer hang connect(). FIX: executemany no longer raises a RuntimeError when decimal signs change between rows (issue #557). FIX: Consistent retrieval of CP1252-encoded data in VARCHAR columns across Windows and Linux (issue #468). FIX: manylinux_2_28 build targets so RHEL 8 / glibc 2.28 users get wheels out of the box. FIX: macOS Python 3.10 now gets a proper universal2 wheel. DOC: Setup instructions for Azure Linux. Full set of changes is in the 1.7.1 PR. The PyPI space story This release also forced us to confront something we've been watching for a while: we ran out of room on PyPI. mssql-python ships native code with bundled ODBC driver binaries for every supported OS / architecture / Python version combination. Each wheel is sizeable on its own, and the number of combinations grows quickly: Windows x64 / ARM64 / x86, macOS universal2, manylinux x86_64 / aarch64, musllinux x86_64 / aarch64, across multiple Python versions. Multiply that by every release we keep on PyPI and the project size adds up fast. PyPI projects have a per-project storage quota. When we tried to release 1.7.0 we hit it. What we did short-term To get 1.7.1 published we cleaned house on old artifacts that nobody should be pulling in fresh installs anymore: Removed all 1.0.0-alpha packages from PyPI. Removed all 1.x.x-beta packages from PyPI. Removed the 1.7.0 packages that made it to PyPI before we hit the space limit. Pre-release alphas and betas were intentionally there for early adopters during the ramp-up of the driver. None of them are recommended for production use, and the GA versions (1.0.0 and up) supersede everything those wheels offered. If you were pinned to an alpha or beta version, please move to the latest GA release (pip install -U mssql-python). If you have a specific reason that you can't, please let us know in an issue. We want to hear about it. What we're doing longer-term Cleaning out old pre-releases buys us breathing room, not a solution. The wheel matrix is going to keep growing as we add platforms, and we don't want to be back in the same spot in three releases. We're attacking the problem from two directions in parallel: Split up the deployment. We're working out how to break the published artifacts into smaller pieces so that an install only pulls down what it actually needs. Concretely we're evaluating splitting the bundled native dependencies (ODBC driver binaries, etc.) out of the main wheels so the per-platform wheel size drops substantially. Exact shape is still being designed and we'll have more to share when we've nailed it down. The goal is that a typical pip install mssql-python stays a one-liner with no extra steps, while the on-disk and on-PyPI footprint is a fraction of what it is today. Asked PyPI for more space. In parallel we've put in a request to PyPI for a storage quota increase. This is a normal process for projects that legitimately need more room, and it gives us a safety net while the longer-term splitting work lands. We are looking at the space increase as the immediate buffer, and the split is the durable fix. Upgrading Nothing user-visible has changed about how you install or use the driver: pip install -U mssql-python If you previously pinned to an alpha or beta version, you'll need to move to the latest GA. If you hit any issues during the upgrade, please file a GitHub issue — we read all of them. Thanks Thanks to everyone who filed bug reports, repro scripts, and PRs that fed into this release. The fixes we shipped came from real user issues, and the SSH-tunnel and CP1252 ones in particular were caught by community reproductions that made the root cause obvious. Keep them coming.Microsoft ODBC Driver 17.11.1 for SQL Server Released
We are pleased to announce the general availability of Microsoft ODBC Driver 17.11.1 for SQL Server, released on April 30, 2026. This servicing update delivers important bug fixes and expands Linux platform support. Key Highlights Stability and correctness fixes for parameter array processing, including accurate updates to SQL_ATTR_PARAMS_PROCESSED_PTR and improved row counting when SQL_PARAM_IGNORE is used in parameter arrays. Fixed a connection error that could occur when processing Data Classification metadata in ODBC asynchronous mode. Updated RPM packaging rules to allow installation of multiple driver versions side by side. Corrected XA recovery to ensure proper computation of transaction IDs and recovery of missing transactions. Debian package installation now honors license acceptance for successful completion. New Platform Support Platform Versions macOS 14, 15, 26 Debian 13 Red Hat Enterprise Linux 10 Oracle Linux 9, 10 SUSE Linux Enterprise Server 16 Ubuntu 24.04, 25.10 Alpine Linux 3.21, 3.22, 3.23 Download The driver is available for download from the Microsoft ODBC Driver for SQL Server documentation page. Linux Installation Install or update using your distribution's package manager: Debian/Ubuntu: sudo apt-get update sudo apt-get install msodbcsql17 Red Hat/Oracle Linux: sudo yum install msodbcsql17 SUSE: sudo zypper install msodbcsql17 Alpine: sudo apk add msodbcsql17 Feedback We welcome your feedback. Please report issues on the SQL Server feedback site or open an issue on the ODBC Driver GitHub repository.go-mssqldb v1.10.0: Better Reliability, Developer Experience, and Standards Compliance
We're excited to announce the release of go-mssqldb v1.10.0, the official Microsoft Go driver for SQL Server and Azure SQL Database. This release brings significant reliability improvements, better standards compliance, and a smoother developer onboarding experience. Highlights Detect Server-Aborted Transactions (XACT_ABORT) One of the most impactful fixes in this release addresses a subtle but dangerous bug: when SQL Server aborts a transaction due to XACT_ABORT ON, the driver now correctly detects this state. Previously, a silently aborted transaction could lead to subsequent statements being auto-committed outside the intended transaction boundary, potentially causing data integrity issues. The driver now returns a clear error when you attempt to use a connection whose transaction was server-aborted. (#370) Implement driver.DriverContext Interface The driver now implements Go's driver.DriverContext interface, enabling sql.OpenDB-style usage and better integration with connection pool configuration via sql.DB.SetConnMaxLifetime, SetMaxOpenConns, etc. This brings the driver in line with modern Go database/sql conventions. (#365) Surface Errors from Rows.Close() Previously, server errors that occurred during the token drain phase of Rows.Close() were silently swallowed. Now these errors are properly surfaced to callers, making it much easier to diagnose issues like permission errors or constraint violations that manifest during result set cleanup. (#361) Nullable Civil Types for Date/Time Parameters New nullable types (civil.NullDate, civil.NullTime, civil.NullDateTime) allow you to pass nullable date/time parameters without resorting to *time.Time or raw interface{} values. These integrate cleanly with the database/sql scanner and valuer interfaces. (#325) DevContainer for Instant Development Getting started with go-mssqldb development is now as simple as opening the repo in VS Code or GitHub Codespaces. The new devcontainer configuration includes a SQL Server instance, pre-configured environment variables, and all required tooling. (#317) All Changes Features Add devcontainer for VS Code and GitHub Codespaces (#317) Add FailoverPartnerSPN connection string parameter (#327) Add NewConnectorWithProcessQueryText for mssql driver compatibility (#341) Add nullable civil types for date/time parameters (#325) Bug Fixes Allow named pipe protocol support for ARM64 Windows (#232) Detect server-aborted transactions to prevent silent auto-commit with XACT_ABORT (#370) Expose TrustServerCertificate in msdsn.Config and URL round-trip (#312) Handle COLINFO and TABNAME TDS tokens returned by tables with triggers (#343) Implement driver.DriverContext interface (#365) Make readCancelConfirmation respect context cancellation (#359) Return interface{} scanType for sql_variant instead of nil (#362) Sanitize credentials from connection string parsing errors (#319) Surface server errors from Rows.Close() during token drain (#361) Upgrade go get github.com/microsoft/go-mssqldb@v1.10.0 This release requires Go 1.21+ and is fully compatible with SQL Server 2012 through 2022 and Azure SQL Database. Contributing We welcome contributions! The new devcontainer makes it easier than ever to get started. Open the repo in VS Code, reopen in the container, and you'll have a full development environment with SQL Server ready to go. GitHub Repository Documentation Report IssuesAnnouncing Microsoft.Data.SqlClient 6.1.5
We are pleased to announce the release of Microsoft.Data.SqlClient 6.1.5, the latest servicing update to the 6.1 line. This update focuses on connection performance, error propagation, and vector type metadata correctness. Install or update from NuGet: dotnet add package Microsoft.Data.SqlClient --version 6.1.5 Full release notes: 6.1.5 Release Notes What's in this release Faster connection opens for non-integrated authentication on native SNI A regression caused SPN (Service Principal Name) generation to run for non-integrated authentication modes, such as SQL authentication, on the native SNI path. That triggered unnecessary DNS lookups and could significantly slow down connection opens. This fix restores the expected behavior for affected .NET applications on Windows. (#3523, #3946) ExecuteScalar now propagates post-row server errors ExecuteScalar could previously hide errors when SQL Server returned row data followed by an error token. In those cases, errors such as conversion failures during WHERE clause evaluation were consumed during SqlDataReader.Close() instead of being thrown to the caller, which could leave transactions unexpectedly zombied. This fix ensures the error is surfaced correctly to application code. (#3736, #3947) Correct metadata type for vector float32 columns SqlDataReader.GetFieldType() and GetProviderSpecificFieldType() now return the correct type, SqlVector<float>, for vector float32 columns. Previously these APIs returned metadata that did not match the type-resolution behavior used by GetValue(). (#4104, #4151) Getting started If you are new to Microsoft.Data.SqlClient, check out the introduction documentation. For users of System.Data.SqlClient, please move migrate to Microsoft.Data.SqlClient now. See the porting cheat sheet. If you encounter any issues, please report them on the GitHub repository.Announcing Microsoft Drivers 5.13.1 for PHP for SQL Server
Announcing Microsoft Drivers 5.13.1 for PHP for SQL Server We have released Microsoft Drivers 5.13.1 for PHP for SQL Server (sqlsrv and pdo_sqlsrv). This patch release addresses several important bug fixes, including a security fix for access token handling in pooled connections and multiple stability improvements. Bug Fixes Access token identity leaking across pooled connections When using access token authentication with connection pooling, connections with different tokens could share the same pool entry, causing identity cross-contamination and use-after-free. This release properly incorporates the access token into the connection pool key, ensuring connections are only reused when the token matches. (#1592, fixes #1396) Prepared statement silently failing on insert A prepared INSERT statement could silently fail when triggers or SET NOCOUNT OFF produce extra result sets, causing an implicit transaction rollback with MARS enabled. The driver now correctly handles this scenario. (#1590) Fatal error re-executing prepared statements with varying result sets Re-executing a prepared statement that returns multiple result sets with different column layouts could cause a fatal error. Metadata entries are now properly freed, and the internal vector is cleared between executions. (#1596) sqlsrv_errors() returning null after failed connection When a connection attempt failed and ODBC provided no diagnostic records, sqlsrv_errors() would return null instead of surfacing the error. Connection failures now consistently report the underlying error. (#1595) Stream becoming invalid when statement goes out of scope A binary stream could become invalid when the originating statement went out of scope, leading to undefined behavior or crashes. The driver now properly invalidates streams when their parent statement is destroyed. (#1598, fixes #1443) Installation PECL (Linux/macOS) sudo pecl install sqlsrv sudo pecl install pdo_sqlsrv Windows Download the prebuilt binaries from the GitHub Releases page and follow the loading instructions. Prerequisites ODBC Driver: Microsoft ODBC Driver 17 or 18 for SQL Server PHP: 8.3, 8.4, or 8.5 For detailed platform-specific installation steps, see the Linux and macOS installation guide. Upgrading from 5.13.0 This is a drop-in hotfix release. No API changes, no configuration changes. Update via PECL or replace the DLLs on Windows. If you are using access token authentication with connection pooling, we strongly recommend upgrading to this release. Resources GitHub Repository Official Documentation PECL sqlsrv PECL pdo_sqlsrv Release Notes (CHANGELOG) Report Issues Feedback We welcome your feedback and contributions. Please file issues, feature requests and pull requests on our GitHub Issues page.Announcing Microsoft.Data.SqlClient 7.0.1
We are pleased to announce the release of Microsoft.Data.SqlClient 7.0.1, the first servicing update to the 7.0 line. This patch addresses several compatibility fixes reported by the community after the 7.0.0 GA release. Install or update from NuGet: dotnet add package Microsoft.Data.SqlClient --version 7.0.1 Full release notes: 7.0.1 Release Notes What's in this release SqlBulkCopy fixes for SQL Server 2016 and Azure Synapse Two separate issues affected SqlBulkCopy on older or specialized SQL Server engines: SQL Server 2016 compatibility -- SqlBulkCopy operations failed with Invalid column name 'graph_type' because the column metadata query referenced a column introduced in SQL Server 2017. The query now uses dynamic SQL so the graph_type reference is not compiled on versions that lack the column. (#3714) Azure Synapse dedicated SQL pools -- The column-list query used a variable-assignment pattern that Synapse does not support. The driver now detects Synapse (engine edition 6) and uses STRING_AGG instead, while preserving the variable-assignment fallback for SQL Server 2016 compatibility. (#4149) Vector column type metadata SqlDataReader.GetFieldType() and GetProviderSpecificFieldType() previously returned typeof(byte[]) for vector float32 columns instead of typeof(SqlVector<float>). These methods now follow the same type-determination logic as GetValue(), returning the correct vector type. (#4104) Missing System.Data.Common dependency on .NET Framework The inbox System.Data.Common assembly on .NET Framework predates APIs such as IDbColumnSchemaGenerator. Without the NuGet package dependency, consumers hit CS0012 compilation errors when using these types through Microsoft.Data.SqlClient. The package now explicitly depends on System.Data.Common v4.3.0 for .NET Framework targets. (#4063) User Agent TDS extension enabled unconditionally The Switch.Microsoft.Data.SqlClient.EnableUserAgent AppContext switch has been removed. The driver now always sends User Agent information during login. (#4124) Type forwards for extracted Azure authentication types Type forwards have been added from the core Microsoft.Data.SqlClient assembly to public types that moved to the Microsoft.Data.SqlClient.Extensions.Abstractions package in 7.0.0: SqlAuthenticationMethod, SqlAuthenticationParameters, SqlAuthenticationProvider, SqlAuthenticationProviderException, and SqlAuthenticationToken. This ensures binary compatibility for assemblies compiled against earlier versions of Microsoft.Data.SqlClient. (#4067) Community contributions Both of the SqlBulkCopy fixes in this release were driven by community contributor edwardneal: SQL Server 2016 fix -- Edward authored the original PR (#3719) that rewrote the column metadata query to use dynamic SQL, preventing the graph_type column reference from being compiled on servers that lack it. The fix was recreated on an internal branch (#4092) to enable CI pipeline testing against SQL Server 2016 and 2017. Azure Synapse fix -- Edward identified and fixed (#4176) the incompatibility with Azure Synapse dedicated SQL pools, where the variable-assignment concatenation pattern used to build the column list is not supported. His fix detects Synapse via SERVERPROPERTY('EngineEdition') and switches to STRING_AGG, while preserving the variable-assignment fallback for SQL Server 2016. He manually validated the fix against SQL Server 2016, SQL Server 2025, and an Azure Synapse dedicated SQL pool. We are grateful for Edward's continued contributions to SqlClient. Community involvement like this directly improves the experience for all SqlClient users. Getting started If you are new to Microsoft.Data.SqlClient, check out the documentation. For users of System.Data.SqlClient, see the porting cheat sheet. If you encounter any issues, please report them on the GitHub repository.mssql-python 1.6: Unblocking Your Threads
The last two mssql-python releases shipped big features: Bulk Copy in 1.4 for high-throughput data loading, and Apache Arrow in 1.5 for zero-copy analytics. Version 1.6 is about what happens next: you take those features into production, scale up your thread pool, and find out where the driver was quietly holding you back. This release unblocks your threads during connection setup, fixes crashes and incorrect results in common cursor patterns, and hardens security for passwords with special characters and log file paths. pip install --upgrade mssql-python Your threads can run while connections are opening If you're running mssql-python behind Flask, FastAPI, Django, or any WSGI/ASGI server with thread-based workers, this one matters. Opening a database connection is slow. There's DNS resolution, a TCP handshake, TLS negotiation, and SQL Server authentication. In previous versions, every other Python thread in your process was frozen while that happened, because the driver held the Global Interpreter Lock (GIL) during the entire operation. One thread opening a connection meant no other thread could serve requests, process data, or do anything at all. Version 1.6 releases the GIL during connect and disconnect. Your other threads keep running while the network round-trip completes. If you have a multi-threaded web server handling concurrent requests, this removes a serialization bottleneck you may not have realized you had. The connection pool was also reworked to stay safe under this change. Previously, the pool held an internal lock while calling connect, which would have created a deadlock now that connect releases the GIL. The pool now reserves a slot first, connects outside the lock, and rolls back the reservation if the connection fails. Decimal parameters work with setinputsizes If you use cursor.setinputsizes() to declare parameter types for performance-sensitive batch inserts, you may have hit a crash when specifying SQL_DECIMAL or SQL_NUMERIC. This is fixed. Decimal values now bind correctly whether you're using execute() or executemany(): cursor.setinputsizes([ (mssql_python.SQL_WVARCHAR, 100, 0), (mssql_python.SQL_INTEGER, 0, 0), (mssql_python.SQL_DECIMAL, 18, 2), ]) cursor.executemany( "INSERT INTO Products (Name, CategoryID, Price) VALUES (?, ?, ?)", [ ("Widget", 1, Decimal("19.99")), ("Gadget", 2, Decimal("29.99")), ], ) Iterating catalog results with fetchone() If you've used cursor.tables(), cursor.columns(), or other catalog methods and tried to walk the results with fetchone(), you may have gotten incorrect data. Row tracking was broken for catalog result sets. This now works the way you'd expect: cursor.tables(tableType="TABLE") while True: row = cursor.fetchone() if row is None: break print(row.table_name) This also applies to primaryKeys(), foreignKeys(), statistics(), procedures(), and getTypeInfo(). Reusing prepared statements without reset If you call cursor.execute() with reset_cursor=False to reuse a prepared statement across calls, this no longer raises an "Invalid cursor state" error. Passwords with special characters stay masked in logs If your SQL Server password contains semicolons, braces, or other ODBC-special characters (e.g., PWD={Top;Secret}), previous versions could accidentally leak part of it in sanitized log output. The password masking logic has been rewritten to correctly handle all ODBC connection string formats. If the connection string can't be parsed at all, the entire string is now redacted rather than partially exposed. The logging system also now rejects log file paths that attempt directory traversal, preventing setup_logging(log_file_path="../../somewhere/else.log") from writing outside the intended directory. Better type checker support for executemany If your type checker flagged executemany() when you passed dictionaries as parameter rows, that warning is gone. The type annotations now correctly accept Mapping types, matching the DB API 2.0 spec for named parameters. Get started pip install --upgrade mssql-python For questions or issues, file them on GitHub or email mssql-python@microsoft.com.
