litestar-org · cofin · Jul 1, 2026 · Jul 2, 2026 · Jul 2, 2026 · Jul 2, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -235,7 +235,7 @@ jobs:
   test-integration:
     needs: [ci-scope, test-unit]
     if: needs.ci-scope.outputs.ci_required == 'true'
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     strategy:
       fail-fast: true
       matrix:
@@ -288,12 +288,22 @@ jobs:
           sudo apt-get clean
           df -h /
 
+      - name: Install SQL Server ODBC driver
+        run: |
+          set -eu
+          curl -sSL -O https://packages.microsoft.com/config/ubuntu/24.04/packages-microsoft-prod.deb
+          sudo dpkg -i packages-microsoft-prod.deb
+          rm packages-microsoft-prod.deb
+          sudo apt-get update
+          sudo ACCEPT_EULA=Y apt-get install -y msodbcsql18 unixodbc-dev
+          odbcinst -q -d -n "ODBC Driver 18 for SQL Server"
+
       - name: Cache docker images
         id: docker-image-cache
         uses: actions/cache@v6
         with:
           path: /tmp/docker-images
-          key: docker-images-v1-${{ hashFiles('.github/workflows/ci.yml') }}
+          key: docker-images-v2-${{ hashFiles('.github/workflows/ci.yml') }}
 
       - name: Load cached docker images
         if: steps.docker-image-cache.outputs.cache-hit == 'true'
@@ -312,6 +322,7 @@ jobs:
             paradedb/paradedb:0.21.5-pg16
             cockroachdb/cockroach:latest
             mysql:8.4
+            mcr.microsoft.com/mssql/server:2022-latest
             gvenzl/oracle-free:23-slim-faststart
             ghcr.io/goccy/bigquery-emulator:latest
             gcr.io/cloud-spanner-emulator/emulator:latest

diff --git a/docs/extensions/adk/adapters.rst b/docs/extensions/adk/adapters.rst
@@ -18,9 +18,11 @@ Use async adapters for best performance with ADK runners:
 - **DuckDB**: ``duckdb`` (analytics; reduced-scope for ADK)
 - **ADBC**: ``adbc`` (Arrow-native portability; reduced-scope for ADK)
 - **Spanner**: ``spanner`` (Google Cloud, globally distributed)
+- **SQL Server over ODBC**: ``arrow_odbc`` with Microsoft ODBC Driver 18
 
-Sync adapters (``psycopg`` sync mode, ``sqlite``, ``mysqlconnector``, ``pymysql``)
-work but require wrapping with ``anyio`` for async ADK runners.
+Sync adapters (``psycopg`` sync mode, ``sqlite``, ``mysqlconnector``,
+``pymysql``, ``arrow_odbc``) work but require wrapping with ``anyio`` for async
+ADK runners.
 
 Each Adapter Provides
 =====================

diff --git a/docs/extensions/adk/backends.rst b/docs/extensions/adk/backends.rst
@@ -93,6 +93,11 @@ The table below classifies every backend by its ADK support level.
      - Full
      - Basic
      - Portability layer; native adapters provide optimized search.
+   * - arrow_odbc
+     - Supported
+     - Full
+     - Basic
+     - SQL Server through Microsoft ODBC Driver 18.
    * - spanner
      - Supported
      - Full
@@ -270,6 +275,20 @@ ADBC (Arrow Database Connectivity) provides a driver-agnostic interface:
 - Memory search uses the portable baseline path; choose a native adapter for
   backend-specific FTS, retention, and storage tuning.
 
+arrow-odbc
+----------
+
+``arrow_odbc`` provides SQL Server-backed ADK storage through Microsoft ODBC
+Driver 18:
+
+- Session and event storage use SQL Server tables with ``DATETIME2(6)`` and
+  ``NVARCHAR(MAX)`` JSON payload columns.
+- ``append_event_and_update_state()`` writes the session update, event row, and
+  scoped state in one committed session.
+- Memory storage uses the portable text-search path with ``LIKE`` matching.
+- Row-oriented ``execute_many()`` is not available; use native Arrow ingest for
+  bulk database writes outside the ADK store.
+
 Spanner
 -------
 

diff --git a/docs/reference/adapters/arrow_odbc.rst b/docs/reference/adapters/arrow_odbc.rst
@@ -7,6 +7,16 @@ Streams ``pyarrow.RecordBatchReader`` results from any ODBC-compliant driver,
 making it a good fit for read-heavy analytical transfer between SQL Server,
 PostgreSQL, MySQL, or other ODBC sources and the Arrow ecosystem.
 
+SQL Server coverage is exercised in CI against SQL Server 2022 through
+``pytest-databases`` and Microsoft ODBC Driver 18. The shared contract matrix
+verifies native Arrow reads, Arrow reader/batch output, and Arrow bulk ingest
+for this adapter. Row-oriented ``execute_many()`` is intentionally unsupported;
+use ``load_from_arrow()`` for bulk writes.
+
+Extension support is SQL Server-backed. The adapter exports a table-backed
+events queue store, a Litestar session store, and Google ADK session/event and
+memory stores for SQL Server connections through Microsoft ODBC Driver 18.
+
 Configuration
 =============
 
@@ -36,6 +46,25 @@ Data Dictionary
    :members:
    :show-inheritance:
 
+Extensions
+==========
+
+.. autoclass:: sqlspec.adapters.arrow_odbc.events.ArrowOdbcEventQueueStore
+   :members:
+   :show-inheritance:
+
+.. autoclass:: sqlspec.adapters.arrow_odbc.litestar.ArrowOdbcStore
+   :members:
+   :show-inheritance:
+
+.. autoclass:: sqlspec.adapters.arrow_odbc.adk.ArrowOdbcADKStore
+   :members:
+   :show-inheritance:
+
+.. autoclass:: sqlspec.adapters.arrow_odbc.adk.ArrowOdbcADKMemoryStore
+   :members:
+   :show-inheritance:
+
 Schema Discovery
 ================
 

diff --git a/docs/reference/extensions/events.rst b/docs/reference/extensions/events.rst
@@ -107,6 +107,10 @@ Listeners
 Event Queue
 ===========
 
+The durable table queue is available for SQL Server through ``arrow_odbc`` when
+configured with Microsoft ODBC Driver 18. It uses SQL Server ``DATETIME2(6)``
+timestamps and ``NVARCHAR`` payload columns.
+
 .. autoclass:: sqlspec.extensions.events.AsyncTableEventQueue
    :members:
    :show-inheritance:

diff --git a/docs/usage/drivers_and_querying.rst b/docs/usage/drivers_and_querying.rst
@@ -97,7 +97,10 @@ asyncpg and CockroachDB-asyncpg (cursors inside a stream-owned transaction),
 pymysql/aiomysql/asyncmy (``SSCursor``), mysql-connector (unbuffered cursors),
 sqlite/aiosqlite and oracledb (chunked ``fetchmany``), psqlpy (server-side
 cursor with ``array_size``), and BigQuery (page-wise result iteration).
-ADBC, DuckDB, mssql-python, arrow-odbc, and Spanner are eager-fallback only.
+ADBC, DuckDB, mssql-python, and Spanner are eager-fallback only for row
+streaming. ``arrow-odbc`` row streaming also materializes dict rows eagerly, but
+``select_to_arrow(..., return_format="reader" | "batches")`` uses the native
+``arrow_odbc`` ``RecordBatchReader`` path.
 
 Lifetime and transaction rules:
 

diff --git a/docs/usage/frameworks/litestar/session_stores.rst b/docs/usage/frameworks/litestar/session_stores.rst
@@ -32,6 +32,7 @@ SQLSpec provides stores for async adapters:
 
 - ``AsyncpgStore`` - PostgreSQL via asyncpg
 - ``AiosqliteStore`` - SQLite via aiosqlite
+- ``ArrowOdbcStore`` - SQL Server via arrow-odbc and Microsoft ODBC Driver 18
 
 Each store automatically creates its session table on first use if it doesn't exist.
 

diff --git a/sqlspec/adapters/arrow_odbc/adk/__init__.py b/sqlspec/adapters/arrow_odbc/adk/__init__.py
@@ -0,0 +1,5 @@
+"""arrow-odbc ADK extension exports."""
+
+from sqlspec.adapters.arrow_odbc.adk.store import ArrowOdbcADKConfig, ArrowOdbcADKMemoryStore, ArrowOdbcADKStore
+
+__all__ = ("ArrowOdbcADKConfig", "ArrowOdbcADKMemoryStore", "ArrowOdbcADKStore")