Rediger

Work with binary data

The mssql-python driver supports binary, varbinary, and image data types for Microsoft SQL.

Microsoft SQL stores binary data in these column types:

Type Description Max size
binary(n) Fixed-length binary data. 8,000 bytes
varbinary(n) Variable-length binary data. 8,000 bytes
varbinary(max) Large binary data. 2 GB
image Legacy large binary data (deprecated). 2 GB

The mssql-python driver returns binary data as Python bytes objects.

When to store binary in the database versus the file system:

  • Store in the database when files are small (under 1 MB), transactional consistency with other data is important, or you need to back up data and files together.
  • Store in the file system (or Azure Blob Storage) when files are large (over 1 MB), you need CDN delivery, or you need to serve files directly to clients without database round-trips.
  • For a middle ground, Microsoft SQL's FILESTREAM feature stores data in the file system with transactional consistency.

Insert binary data

Pass Python bytes objects as parameters; the driver maps them to varbinary columns.

Insert bytes directly

Create a Python bytes object and pass it as a query parameter:

import mssql_python

conn = mssql_python.connect(connection_string)
cursor = conn.cursor()

# Create a table to hold binary documents
cursor.execute("""
    CREATE TABLE #Documents (
        ID INT IDENTITY PRIMARY KEY,
        Name NVARCHAR(255),
        Content VARBINARY(MAX),
        Size INT NULL,
        ContentHash VARBINARY(32) NULL
    )
""")

# Binary data as bytes
binary_data = b'\x00\x01\x02\x03\x04\x05'

cursor.execute(
    "INSERT INTO #Documents (Name, Content) VALUES (%(name)s, %(content)s)",
    {"name": "sample.bin", "content": binary_data}
)
conn.commit()

Insert from file

Read a file in binary mode and insert its contents:

def insert_file(cursor, file_path: str, name: str):
    """Insert a file as binary data."""
    with open(file_path, "rb") as f:
        content = f.read()
    
    cursor.execute(
        "INSERT INTO #Documents (Name, Content, Size) VALUES (%(name)s, %(content)s, %(size)s)",
        {"name": name, "content": content, "size": len(content)}
    )

insert_file(cursor, "image.png", "profile_picture.png")
conn.commit()

Insert image file

Store image files with metadata by detecting MIME types from file extensions:

# Create a table to hold images
cursor.execute("""
    CREATE TABLE #Images (
        ID INT IDENTITY PRIMARY KEY,
        FileName NVARCHAR(255),
        FileSize INT NULL,
        ContentType NVARCHAR(100) NULL,
        Width INT NULL,
        Height INT NULL,
        ImageData VARBINARY(MAX),
        Description NVARCHAR(MAX) NULL
    )
""")

def insert_image(cursor, image_path: str, description: str):
    """Insert an image into the database."""
    import os
    
    with open(image_path, "rb") as f:
        image_data = f.read()
    
    cursor.execute("""
        INSERT INTO #Images (FileName, FileSize, ContentType, ImageData, Description)
        VALUES (%(filename)s, %(filesize)s, %(content_type)s, %(data)s, %(desc)s)
    """, {
        "filename": os.path.basename(image_path),
        "filesize": len(image_data),
        "content_type": get_content_type(image_path),
        "data": image_data,
        "desc": description
    })

def get_content_type(path: str) -> str:
    """Determine MIME type from file extension."""
    ext = path.lower().split(".")[-1]
    types = {
        "png": "image/png",
        "jpg": "image/jpeg",
        "jpeg": "image/jpeg",
        "gif": "image/gif",
        "pdf": "application/pdf",
    }
    return types.get(ext, "application/octet-stream")

Retrieve binary data

The driver returns varbinary and binary column values as Python bytes objects.

Fetch binary column

Query binary columns and inspect the returned bytes object:

cursor.execute(
    "SELECT LargePhotoFileName, LargePhoto FROM Production.ProductPhoto WHERE ProductPhotoID = %(id)s",
    {"id": 70}
)
row = cursor.fetchone()

# LargePhoto is bytes
print(type(row.LargePhoto))  # <class 'bytes'>
print(len(row.LargePhoto))   # Number of bytes

Save to file

Retrieve binary data and write it to disk:

def save_photo(cursor, photo_id: int, output_path: str):
    """Retrieve binary data and save to file."""
    cursor.execute(
        "SELECT LargePhoto FROM Production.ProductPhoto WHERE ProductPhotoID = %(id)s",
        {"id": photo_id}
    )
    row = cursor.fetchone()
    
    if row and row.LargePhoto:
        with open(output_path, "wb") as f:
            f.write(row.LargePhoto)
        print(f"Saved {len(row.LargePhoto)} bytes to {output_path}")
    else:
        print("Photo not found or empty")

save_photo(cursor, 70, "output.gif")

Export binary rows to files

Export multiple binary rows to local files:

import os

def export_photos(cursor, output_dir: str):
    """Export product photos to a directory."""
    os.makedirs(output_dir, exist_ok=True)
    
    cursor.execute("""
        SELECT ProductPhotoID, LargePhotoFileName, LargePhoto
        FROM Production.ProductPhoto
        WHERE ProductPhotoID > 1
    """)
    
    count = 0
    for row in cursor:
        output_path = os.path.join(output_dir, f"{row.ProductPhotoID}_{row.LargePhotoFileName}")
        with open(output_path, "wb") as f:
            f.write(row.LargePhoto)
        count += 1
    
    print(f"Exported {count} photos to {output_dir}")

export_photos(cursor, "./photos")

Insert NULL binary values

Pass None to insert a SQL NULL into a binary column. Because #Documents is a temporary table, declare the parameter types by using setinputsizes() first so the driver binds Content as varbinary:

cursor.setinputsizes([(mssql_python.SQL_WVARCHAR, 255, 0), (mssql_python.SQL_VARBINARY, 0, 0)])
cursor.execute(
    "INSERT INTO #Documents (Name, Content) VALUES (%(name)s, %(content)s)",
    {"name": "empty", "content": None}
)

Note

The driver normally infers parameter types through SQLDescribeParam, but it can't resolve type metadata for a temporary table or table variable. Inserting None into a binary or varbinary column of a temp object without setinputsizes() raises ProgrammingError: Implicit conversion from data type varchar to varbinary(max) is not allowed. Pass one entry per parameter, in order, and use a SQL type constant such as mssql_python.SQL_VARBINARY for the binary column. For a regular (permanent) table, the driver resolves the type automatically and you can pass None directly.

Large binary data

For files over a few megabytes, insert the full content in a single varbinary(max) write instead of reading the file in small chunks.

Stream large files

For large files, process in chunks:

def insert_large_file(conn, table: str, file_path: str, chunk_size: int = 8192):
    """Insert large file in chunks using updatetext-style approach."""
    # Validate table name to prevent SQL injection
    import re
    if not re.match(r'^#{0,2}[A-Za-z_][A-Za-z0-9_.]*$', table):
        raise ValueError(f"Invalid table name: {table}")

    cursor = conn.cursor()
    
    # Get file size
    import os
    file_size = os.path.getsize(file_path)
    
    # Insert initial row with empty binary
    cursor.execute(f"""
        INSERT INTO {table} (Name, Content, Size)
        VALUES (%(name)s, 0x, %(size)s);
        SELECT SCOPE_IDENTITY();
    """, {"name": os.path.basename(file_path), "size": file_size})
    
    row_id = cursor.fetchval()
    
    # For modern Microsoft SQL, better to use varbinary(max) and single insert
    # This example shows streaming approach
    with open(file_path, "rb") as f:
        content = f.read()
    
    cursor.execute(f"""
        UPDATE {table} SET Content = %(content)s WHERE ID = %(id)s
    """, {"content": content, "id": row_id})
    
    conn.commit()
    return row_id

Use bulk copy for binary data

Use the bulkcopy() method to efficiently insert multiple binary files in a single operation.

Important

bulkcopy() loads data over a separate connection to the server, so the destination table must already exist and be committed and visible to other sessions. If you create the table in the same script with autocommit off, call conn.commit() before bulkcopy(). Local temporary tables (#name) aren't supported because they're private to the session that created them. An uncommitted or unreachable destination causes bulkcopy() to fail with a Failed to retrieve destination metadata timeout.

By default, bulkcopy() maps each value in a row to a table column by ordinal position, so every column must have a value in order. When the table has an identity column or you only populate some columns, pass column_mappings to name the target columns explicitly. Otherwise the values shift onto the wrong columns and bulkcopy() fails.

def bulk_insert_files(conn, table: str, file_paths: list[str]):
    """Bulk insert multiple binary files."""
    import os

    data = []
    for path in file_paths:
        with open(path, "rb") as f:
            content = f.read()
        data.append((os.path.basename(path), content, len(content)))

    cursor = conn.cursor()
    result = cursor.bulkcopy(table, data, column_mappings=["Name", "Content", "Size"])
    conn.commit()
    return result["rows_copied"]

Binary data operations

Use Microsoft SQL's HASHBYTES function to compare binary content without fetching full values.

Compare binary data

Find binary rows that share identical content by computing and comparing content hashes in Microsoft SQL Server:

# Find rows with identical binary content by hash
cursor.execute("""
    SELECT LargePhotoFileName, HASHBYTES('SHA2_256', LargePhoto) AS PhotoHash
    FROM Production.ProductPhoto
    WHERE ProductPhotoID > 1
""")

hashes = {}
for row in cursor:
    hash_value = row.PhotoHash  # bytes
    if hash_value in hashes:
        print(f"Duplicate: {row.LargePhotoFileName} matches {hashes[hash_value]}")
    else:
        hashes[hash_value] = row.LargePhotoFileName

Compute hash in Python

Compute SHA-256 hashes in Python and store them alongside binary data:

import hashlib

def insert_with_hash(cursor, name: str, content: bytes):
    """Insert binary data with computed hash."""
    content_hash = hashlib.sha256(content).digest()
    
    cursor.execute("""
        INSERT INTO #Documents (Name, Content, ContentHash)
        VALUES (%(name)s, %(content)s, %(hash)s)
    """, {"name": name, "content": content, "hash": content_hash})

Encode and decode binary data

Convert binary data to and from base64-encoded strings:

import base64

# Store base64-encoded string
def insert_base64(cursor, name: str, base64_data: str):
    """Insert base64-encoded data as binary."""
    binary_data = base64.b64decode(base64_data)
    cursor.execute(
        "INSERT INTO #Documents (Name, Content) VALUES (%(name)s, %(content)s)",
        {"name": name, "content": binary_data}
    )

# Retrieve as base64
def get_as_base64(cursor, doc_id: int) -> str:
    """Retrieve binary data as base64 string."""
    cursor.execute("SELECT Content FROM #Documents WHERE ID = %(id)s", {"id": doc_id})
    row = cursor.fetchone()
    return base64.b64encode(row.Content).decode("utf-8")

Work with specific binary formats

These examples show how to validate and insert common file formats before storing them.

PDF documents

Verify PDF file signatures before inserting them:

def insert_pdf(cursor, pdf_path: str, title: str):
    """Insert a PDF document."""
    with open(pdf_path, "rb") as f:
        pdf_data = f.read()
    
    # Verify it's a PDF (magic bytes)
    if not pdf_data.startswith(b'%PDF'):
        raise ValueError("Not a valid PDF file")
    
    cursor.execute(
        "INSERT INTO #Documents (Name, Content) VALUES (%(title)s, %(data)s)",
        {"title": title, "data": pdf_data}
    )

Images with PIL/Pillow

Resize images using Pillow (PIL) before storing them to save space:

from PIL import Image
import io
import os

def insert_resized_image(cursor, image_path: str, max_size: tuple = (800, 600)):
    """Insert a resized image."""
    # Open and resize
    img = Image.open(image_path)
    img.thumbnail(max_size, Image.LANCZOS)
    
    # Convert to bytes
    buffer = io.BytesIO()
    img.save(buffer, format=img.format or "PNG")
    image_bytes = buffer.getvalue()
    
    cursor.execute("""
        INSERT INTO #Images (FileName, Width, Height, ImageData)
        VALUES (%(name)s, %(width)s, %(height)s, %(data)s)
    """, {
        "name": os.path.basename(image_path),
        "width": img.width,
        "height": img.height,
        "data": image_bytes
    })

def get_image_as_pil(cursor, image_id: int) -> Image.Image:
    """Retrieve image as PIL Image object."""
    cursor.execute("SELECT ImageData FROM #Images WHERE ID = %(id)s", {"id": image_id})
    row = cursor.fetchone()
    return Image.open(io.BytesIO(row.ImageData))

Compressed data

Reduce storage by compressing binary data with gzip before insertion:

import gzip

def insert_compressed(cursor, name: str, data: bytes):
    """Insert data with gzip compression."""
    compressed = gzip.compress(data)
    
    cursor.execute(
        "INSERT INTO #Documents (Name, Content, Size) VALUES (%(name)s, %(content)s, %(size)s)",
        {"name": name, "content": compressed, "size": len(data)}
    )

def get_decompressed(cursor, data_id: int) -> bytes:
    """Retrieve and decompress data."""
    cursor.execute(
        "SELECT Content FROM #Documents WHERE ID = %(id)s",
        {"id": data_id}
    )
    row = cursor.fetchone()
    return gzip.decompress(row.Content)

Best practices

Apply these guidelines to choose the right column type and storage strategy for binary data.

Use appropriate column types

Choose the column type based on your data's characteristics:

For different binary use cases, select the appropriate Microsoft SQL data type:

-- For small fixed-size binary (for example, hashes, UUIDs)
binary(32)      -- SHA-256 hash

-- For variable-size binary up to 8KB (for example, thumbnails, small icons)
varbinary(8000)

-- For large binary data (for example, documents, images)
varbinary(max)  -- Up to 2GB

Consider FILESTREAM for large files

For large files (over 1 MB), consider the Microsoft SQL FILESTREAM feature, which stores data in the file system. FILESTREAM requires server-side configuration before use:

# FILESTREAM-enabled databases store large binaries more efficiently
# Access is still through normal queries but storage is file-based
large_binary_data = b"\x25\x50\x44\x46" + b"\x00" * 100  # sample data

cursor.execute("""
    CREATE TABLE ##FileStreamDemo (Name NVARCHAR(100), Document VARBINARY(MAX))
""")
cursor.execute("""
    INSERT INTO ##FileStreamDemo (Name, Document)
    VALUES (%(name)s, %(content)s)
""", {"name": "large_doc.pdf", "content": large_binary_data})

cursor.execute("SELECT Name, DATALENGTH(Document) AS DocSize FROM ##FileStreamDemo")
row = cursor.fetchone()
print(f"{row.Name}: {row.DocSize} bytes")

Validate binary data

Validate file types by checking magic bytes (file signatures) before storing binary data:

def insert_safe_image(cursor, name: str, data: bytes):
    """Insert image with validation."""
    # Check file signatures (magic bytes)
    signatures = {
        b'\x89PNG': 'image/png',
        b'\xff\xd8\xff': 'image/jpeg',
        b'GIF87a': 'image/gif',
        b'GIF89a': 'image/gif',
    }
    
    content_type = None
    for sig, mime in signatures.items():
        if data.startswith(sig):
            content_type = mime
            break
    
    if content_type is None:
        raise ValueError("Unknown or unsupported image format")
    
    cursor.execute("""
        INSERT INTO #Images (FileName, ContentType, ImageData)
        VALUES (%(name)s, %(type)s, %(data)s)
    """, {"name": name, "type": content_type, "data": data})