FastMSSQL ⚡

FastMSSQL is an async Python library for Microsoft SQL Server (MSSQL), built in Rust. Unlike standard libaries, it uses a native SQL Server client—no ODBC required—simplifying installation on Windows, macOS, and Linux. Great for data ingestion, bulk inserts, and large-scale query workloads.

Features

• High performance: optimized for very high RPS and low overhead
• Rust core: memory‑safe and reliable, tuned Tokio runtime
• No ODBC: native SQL Server client, no external drivers needed
• Azure authentication: Service Principal, Managed Identity, and access token support (BETA)
• Connection pooling: bb8‑based, smart defaults (default max_size=20, min_idle=2)
• Async first: clean async/await API with async with context managers
• Strong typing: fast conversions for common SQL Server types
• Thread‑safe: safe to use in concurrent apps
• Cross‑platform: Windows, macOS, Linux
• Batch operations: high-performance bulk inserts and batch query execution
• Apache Arrow support

Installation

From PyPI (recommended)

pip install fastmssql

Optional dependencies

Apache Arrow support (for to_arrow() method):

pip install fastmssql[arrow]

Prerequisites

• Python 3.11 to 3.14
• Microsoft SQL Server (any recent version)

Quick start

Basic async usage

import asyncio
from fastmssql import Connection

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    async with Connection(conn_str) as conn:
        # SELECT: use query() -> rows()
        result = await conn.query("SELECT @@VERSION as version")
        for row in result.rows():
            print(row['version'])

        # Pool statistics (tuple: connected, connections, idle, max_size, min_idle)
        connected, connections, idle, max_size, min_idle = await conn.pool_stats()
        print(f"Pool: connected={connected}, size={connections}/{max_size}, idle={idle}, min_idle={min_idle}")

asyncio.run(main())

Explicit Connection Management

When not utilizing Python's context manager (async with), FastMssql uses lazy connection initialization: if you call query() or execute() on a new Connection, the underlying pool is created if not already present.

For more control, you can explicitly connect and disconnect:

import asyncio
from fastmssql import Connection

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    conn = Connection(conn_str)

    # Explicitly connect
    await conn.connect()
    assert await conn.is_connected()

    # Run queries
    result = await conn.query("SELECT 42 as answer")
    print(result.rows()[0]["answer"])  # -> 42

    # Explicitly disconnect
    await conn.disconnect()
    assert not await conn.is_connected()

asyncio.run(main())

Usage

Connection options

You can connect either with a connection string or individual parameters.

1. Connection string

import asyncio
from fastmssql import Connection

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    async with Connection(connection_string=conn_str) as conn:
        rows = (await conn.query("SELECT DB_NAME() as db")).rows()
        print(rows[0]['db'])

asyncio.run(main())

1. Individual parameters

import asyncio
from fastmssql import Connection

async def main():
    async with Connection(
        server="localhost",
        database="master",
        username="myuser",
        password="mypassword"
    ) as conn:
        rows = (await conn.query("SELECT SUSER_SID() as sid")).rows()
        print(rows[0]['sid'])

asyncio.run(main())

Note: Windows authentication (Trusted Connection) is currently not supported. Use SQL authentication (username/password).

Azure Authentication (BETA)

🧪 This is a beta feature. Azure authentication functionality is experimental and may change in future versions.

FastMSSSQL supports Azure Active Directory (AAD) authentication for Azure SQL Database and Azure SQL Managed Instance. You can authenticate using Service Principals, Managed Identity, or access tokens.

Service Principal Authentication

import asyncio
from fastmssql import Connection, AzureCredential

async def main():
    # Create Azure credential using Service Principal
    azure_cred = AzureCredential.service_principal(
        client_id="your-client-id",
        client_secret="your-client-secret", 
        tenant_id="your-tenant-id"
    )
    
    async with Connection(
        server="yourserver.database.windows.net",
        database="yourdatabase",
        azure_credential=azure_cred
    ) as conn:
        result = await conn.query("SELECT GETDATE() as current_time")
        for row in result.rows():
            print(f"Connected! Current time: {row['current_time']}")

asyncio.run(main())

Managed Identity Authentication

For Azure resources (VMs, Function Apps, App Service, etc.):

import asyncio
from fastmssql import Connection, AzureCredential

async def main():
    # System-assigned managed identity
    azure_cred = AzureCredential.managed_identity()
    
    # Or user-assigned managed identity
    # azure_cred = AzureCredential.managed_identity(client_id="user-assigned-identity-client-id")
    
    async with Connection(
        server="yourserver.database.windows.net",
        database="yourdatabase",
        azure_credential=azure_cred
    ) as conn:
        result = await conn.query("SELECT USER_NAME() as user_name")
        for row in result.rows():
            print(f"Connected as: {row['user_name']}")

asyncio.run(main())

Access Token Authentication

If you already have an access token from another Azure service:

import asyncio
from fastmssql import Connection, AzureCredential

async def main():
    # Use a pre-obtained access token
    access_token = "your-access-token"
    azure_cred = AzureCredential.access_token(access_token)
    
    async with Connection(
        server="yourserver.database.windows.net",
        database="yourdatabase",
        azure_credential=azure_cred
    ) as conn:
        result = await conn.query("SELECT 1 as test")
        print("Connected with access token!")

asyncio.run(main())

Default Azure Credential

Uses the Azure credential chain (environment variables → managed identity → Azure CLI → Azure PowerShell):

import asyncio
from fastmssql import Connection, AzureCredential

async def main():
    # Use default Azure credential chain
    azure_cred = AzureCredential.default()
    
    async with Connection(
        server="yourserver.database.windows.net",
        database="yourdatabase",
        azure_credential=azure_cred
    ) as conn:
        result = await conn.query("SELECT 1 as test")
        print("Connected with default credentials!")

asyncio.run(main())

Prerequisites for Azure Authentication:

• Azure SQL Database or Azure SQL Managed Instance
• Service Principal with appropriate SQL Database permissions
• For Managed Identity: Azure resource with managed identity enabled
• For Default credential: Azure CLI installed and authenticated (az login)

See examples/azure_auth_example.py for comprehensive usage examples.

Working with data

import asyncio
from fastmssql import Connection

async def main():
    async with Connection("Server=.;Database=MyDB;User Id=sa;Password=StrongPwd;") as conn:
        # SELECT (returns rows)
        users = (await conn.query(
            "SELECT id, name, email FROM users WHERE active = 1"
        )).rows()
        for u in users:
            print(f"User {u['id']}: {u['name']} ({u['email']})")

        # INSERT / UPDATE / DELETE (returns affected row count)
        inserted = await conn.execute(
            "INSERT INTO users (name, email) VALUES (@P1, @P2)",
            ["Jane", "jane@example.com"],
        )
        print(f"Inserted {inserted} row(s)")

        updated = await conn.execute(
            "UPDATE users SET last_login = GETDATE() WHERE id = @P1",
            [123],
        )
        print(f"Updated {updated} row(s)")

asyncio.run(main())

Parameters use positional placeholders: @P1, @P2, ... Provide values as a list in the same order.

Batch operations

For high-throughput scenarios, use batch methods to reduce network round-trips:

import asyncio
from fastmssql import Connection

async def main_fetching():
    # Replace with your actual connection string
    async with Connection("Server=.;Database=MyDB;User Id=sa;Password=StrongPwd;") as conn:

        # --- 1. Prepare Data for Demonstration ---
        columns = ["name", "email", "age"]
        data_rows = [
            ["Alice Johnson", "alice@example.com", 28],
            ["Bob Smith", "bob@example.com", 32],
            ["Carol Davis", "carol@example.com", 25],
            ["David Lee", "david@example.com", 35],
            ["Eva Green", "eva@example.com", 29]
        ]
        await conn.bulk_insert("users", columns, data_rows)

        # --- 2. Execute Query and Retrieve the Result Object ---
        print("\n--- Result Object Fetching (fetchone, fetchmany, fetchall) ---")

        # The Result object is returned after the awaitable query executes.
        result = await conn.query("SELECT name, age FROM users ORDER BY age DESC")

        # fetchone(): Retrieves the next single row synchronously.
        oldest_user = result.fetchone()
        if oldest_user:
            print(f"1. fetchone: Oldest user is {oldest_user['name']} (Age: {oldest_user['age']})")

        # fetchmany(2): Retrieves the next set of rows synchronously.
        next_two_users = result.fetchmany(2)
        print(f"2. fetchmany: Retrieved {len(next_two_users)} users: {[r['name'] for r in next_two_users]}.")

        # fetchall(): Retrieves all remaining rows synchronously.
        remaining_users = result.fetchall()
        print(f"3. fetchall: Retrieved all {len(remaining_users)} remaining users: {[r['name'] for r in remaining_users]}.")

        # Exhaustion Check: Subsequent calls return None/[]
        print(f"4. Exhaustion Check (fetchone): {result.fetchone()}")
        print(f"5. Exhaustion Check (fetchmany): {result.fetchmany(1)}")

        # --- 3. Batch Commands for multiple operations ---
        print("\n--- Batch Commands (execute_batch) ---")
        commands = [
            ("UPDATE users SET last_login = GETDATE() WHERE name = @P1", ["Alice Johnson"]),
            ("INSERT INTO user_logs (action, user_name) VALUES (@P1, @P2)", ["login", "Alice Johnson"])
        ]

        affected_counts = await conn.execute_batch(commands)
        print(f"Updated {affected_counts[0]} users, inserted {affected_counts[1]} logs")

asyncio.run(main_fetching())

Apache Arrow

Convert query results to Apache Arrow tables for efficient bulk data processing and interoperability with data science tools:

import asyncio
from fastmssql import Connection

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    async with Connection(conn_str) as conn:
        # Execute query and convert to Arrow
        result = await conn.query("SELECT id, name, salary FROM employees")
        arrow_table = result.to_arrow()
        
        # Arrow Table enables:
        # - Efficient columnar storage and compute
        # - Integration with Pandas, DuckDB, Polars
        # - Parquet/ORC serialization
        df = arrow_table.to_pandas()  # Convert to pandas DataFrame
        print(df)
        
        # Write to Parquet for long-term storage
        import pyarrow.parquet as pq
        pq.write_table(arrow_table, "employees.parquet")
        
        # Or use with DuckDB for analytical queries
        import duckdb
        result = duckdb.from_arrow(arrow_table).filter("salary > 50000").execute()
        print(result.fetchall())

Requirements: Install PyArrow with pip install pyarrow

Note: Results are converted eagerly into Arrow arrays. For very large datasets, consider chunking queries or using iteration-based processing instead.

Connection pooling

Tune the pool to fit your workload. Constructor signature:

from fastmssql import PoolConfig

config = PoolConfig(
    max_size=20,              # max connections in pool
    min_idle=5,               # keep at least this many idle
    max_lifetime_secs=3600,   # recycle connections after 1h
    idle_timeout_secs=600,    # close idle connections after 10m
    connection_timeout_secs=30
)

Presets:

one   = PoolConfig.one()                     # max_size=1,  min_idle=1  (single connection)
low   = PoolConfig.low_resource()            # max_size=3,  min_idle=1  (constrained environments)
dev   = PoolConfig.development()             # max_size=5,  min_idle=1  (local development)
high  = PoolConfig.high_throughput()         # max_size=25, min_idle=8  (high-throughput workloads)
maxp  = PoolConfig.performance()             # max_size=30, min_idle=10 (maximum performance)

# ✨ RECOMMENDED: Adaptive pool sizing based on your concurrency
adapt = PoolConfig.adaptive(20)              # Dynamically sized for 20 concurrent workers
                                             # Formula: max_size = ceil(workers * 1.2) + 5

⚡ Performance Tip: Use PoolConfig.adaptive(n) where n is your expected concurrent workers/tasks. This prevents connection pool lock contention that can degrade performance with oversized pools.

Apply to a connection:

# Recommended: adaptive sizing
async with Connection(conn_str, pool_config=PoolConfig.adaptive(20)) as conn:
    rows = (await conn.query("SELECT 1 AS ok")).rows()

# Or use presets
async with Connection(conn_str, pool_config=PoolConfig.high_throughput()) as conn:
    rows = (await conn.query("SELECT 1 AS ok")).rows()

Default pool (if omitted): max_size=15, min_idle=3.

Transactions

For workloads that require SQL Server transactions with guaranteed connection isolation, use the Transaction class. Unlike Connection (which uses connection pooling), Transaction maintains a dedicated, non-pooled connection for the lifetime of the transaction. This ensures all operations within the transaction run on the same connection, preventing connection-switching issues.

Automatic transaction control (recommended)

Use the context manager for automatic BEGIN, COMMIT, and ROLLBACK:

import asyncio
from fastmssql import Transaction

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    
    async with Transaction(conn_str) as transaction:
        # Automatically calls BEGIN
        await transaction.execute(
            "INSERT INTO orders (customer_id, total) VALUES (@P1, @P2)",
            [123, 99.99]
        )
        await transaction.execute(
            "INSERT INTO order_items (order_id, product_id, qty) VALUES (@P1, @P2, @P3)",
            [1, 456, 2]
        )
        # Automatically calls COMMIT on successful exit
        # or ROLLBACK if an exception occurs

asyncio.run(main())

Manual transaction control

For more control, explicitly call begin(), commit(), and rollback():

import asyncio
from fastmssql import Transaction, SqlError

async def main():
    conn_str = "Server=localhost;Database=master;User Id=myuser;Password=mypass"
    transaction = Transaction(conn_str)
    
    try:
        await transaction.begin()
        
        result = await transaction.query("SELECT @@VERSION as version")
        print(result.rows()[0]['version'])
        
        await transaction.execute("UPDATE accounts SET balance = balance - @P1 WHERE id = @P2", [50, 1])
        await transaction.execute("UPDATE accounts SET balance = balance + @P1 WHERE id = @P2", [50, 2])
        
        await transaction.commit()
    except SqlError as e:
        await transaction.rollback()
        raise
    finally:
        await transaction.close()

asyncio.run(main())

Key differences: Transaction vs Connection

Feature	Transaction	Connection
Connection	Dedicated, non-pooled	Pooled (bb8)
Use case	SQL transactions, ACID operations	General queries, connection reuse
Isolation	Single connection per instance	Connection may vary per operation
Pooling	None (direct TcpStream)	Configurable pool settings
Lifecycle	Held until .close() or context exit	Released to pool after each operation

Choose Transaction when you need guaranteed transaction isolation; use Connection for typical queries and high-concurrency workloads with connection pooling.

SSL/TLS

For Required and LoginOnly encryption, you must specify how to validate the server certificate:

Option 1: Trust Server Certificate (development/self-signed certs):

from fastmssql import SslConfig, EncryptionLevel, Connection

ssl = SslConfig(
    encryption_level=EncryptionLevel.Required,
    trust_server_certificate=True
)

async with Connection(conn_str, ssl_config=ssl) as conn:
    ...

Option 2: Custom CA Certificate (production):

from fastmssql import SslConfig, EncryptionLevel, Connection

ssl = SslConfig(
    encryption_level=EncryptionLevel.Required,
    ca_certificate_path="/path/to/ca-cert.pem"
)

async with Connection(conn_str, ssl_config=ssl) as conn:
    ...

Note: trust_server_certificate and ca_certificate_path are mutually exclusive.

Helpers:

• SslConfig.development() – encrypt, trust all (dev only)
• SslConfig.with_ca_certificate(path) – use custom CA
• SslConfig.login_only() / SslConfig.disabled() – legacy modes
• SslConfig.disabled() – no encryption (not recommended)

Performance tips

1. Use adaptive pool sizing for optimal concurrency

Match your pool size to actual concurrency to avoid connection pool lock contention:

import asyncio
from fastmssql import Connection, PoolConfig

async def worker(conn_str, cfg):
    async with Connection(conn_str, pool_config=cfg) as conn:
        for _ in range(1000):
            result = await conn.query("SELECT 1 as v")
            # ✅ Good: Lazy iteration (minimal GIL hold per row)
            for row in result:
                process(row)

async def main():
    conn_str = "Server=.;Database=master;User Id=sa;Password=StrongPwd;"
    num_workers = 32
    
    # ✅ Adaptive sizing prevents pool contention
    cfg = PoolConfig.adaptive(num_workers)  # → max_size=43 for 32 workers
    
    await asyncio.gather(*[worker(conn_str, cfg) for _ in range(num_workers)])

asyncio.run(main())

2. Use iteration for large result sets (not .rows())

result = await conn.query("SELECT * FROM large_table")

# ✅ Good: Lazy conversion, one row at a time (minimal GIL contention)
for row in result:
    process(row)

# ❌ Bad: Eager conversion, all rows at once (GIL bottleneck)
all_rows = result.rows()  # or result.fetchall()

Lazy iteration distributes GIL acquisition across rows, dramatically improving performance with multiple Python workers.

Examples & benchmarks

• Examples: examples/comprehensive_example.py
• Benchmarks: benchmarks/

Troubleshooting

• Import/build: ensure Rust toolchain and maturin are installed if building from source
• Connection: verify connection string; Windows auth not supported
• Timeouts: increase pool size or tune connection_timeout_secs
• Parameters: use @P1, @P2, ... and pass a list of values

Contributing

Contributions are welcome. Please open an issue or PR.

License

FastMSSQL is licensed under MIT:

See the LICENSE file for details.

Third‑party attributions

Built on excellent open source projects: Tiberius, PyO3, pyo3‑asyncio, bb8, tokio, serde, pytest, maturin, and more. See licenses/NOTICE.txt for the full list. The full texts of Apache‑2.0 and MIT are in licenses/.

Acknowledgments

Thanks to the maintainers of Tiberius, bb8, PyO3, Tokio, pytest, maturin, and the broader open source community.