处理二进制数据

mssql-python 驱动程序支持 Microsoft SQL 的 binaryvarbinaryimage 数据类型。

Microsoft SQL 在以下列类型中存储二进制数据:

类型 描述 最大大小
binary(n) 固定长度的二进制数据。 8000字节
varbinary(n) 可变长度二进制数据。 8000字节
varbinary(max) 大型二进制数据。 2 GB
image 遗留的大二进制数据(已弃用)。 2 GB

mssql-python 驱动以 Python bytes 对象的形式返回二进制数据。

何时将二进制存储在数据库中,何时存储在文件系统中:

  • 当文件较小(低于 1 MB)、需要与其他数据保持事务一致性,或者需要将数据和文件一起备份时,应将文件存储在数据库中。
  • 当文件体积大(超过1MB)、需要CDN传输,或者需要直接向客户端提供文件且无需数据库往返时,存储在文件系统(或Azure Blob 存储)中。
  • 作为折中方案,Microsoft SQL 的 FILESTREAM 功能将数据存储在文件系统中,同时保持事务一致性。

插入二进制数据

将 Python bytes 对象作为参数传递;驱动程序将其映射到 varbinary 列。

直接插入字节

创建一个 Python bytes 对象,并将其作为查询参数传递:

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()

从文件插入

以二进制模式读取文件并插入其内容:

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()

插入图像文件

通过通过文件扩展名检测MIME类型来存储包含元数据的图像文件:

# 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")

检索二进制数据

驱动程序将 varbinarybinary 列值作为 Python bytes 对象返回。

提取二进制列

查询二进制列并检查返回 bytes 的对象:

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

保存到文件

检索二进制数据并写入磁盘:

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")

导出二进制行到文件

将多行二进制数据导出到本地文件:

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")

插入空二进制值

传入 None 以在二进制列中插入 SQL NULL。 由于 #Documents 是临时表,请先使用 setinputsizes() 声明参数类型,以便驱动程序将 Content 绑定为 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}
)

注释

驱动程序通常通过 SQLDescribeParam推断参数类型,但无法解析 临时表表变量的类型元数据。 将 None 插入临时对象的 binaryvarbinary 列且未使用 setinputsizes() 时,会引发 ProgrammingError: Implicit conversion from data type varchar to varbinary(max) is not allowed。 按顺序为每个参数传递一个条目,并对二进制列使用诸如 mssql_python.SQL_VARBINARY 之类的 SQL 类型常量。 对于普通(永久)表,驱动会自动解析类型,你可以直接传递 None

大型二进制数据

对于超过几兆字节的文件,应通过一次varbinary(max)写入插入全部内容,而不是以小块方式读取文件。

流式传输大文件

对于大文件,按分块处理:

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

使用批量副本来存储二进制数据

使用该 bulkcopy() 方法在一次操作中高效插入多个二进制文件。

Important

bulkcopy() 通过与服务器的 单独连接 加载数据,因此目标表必须已存在,并且 已提交且对其他会话可见。 如果你在同一脚本中创建表并关闭了自动提交,请先调用 conn.commit(),再调用 bulkcopy()。 本地临时表(#name)不受支持,因为它们仅对创建它们的会话可见。 未提交的或无法到达的目的地会导致 bulkcopy()Failed to retrieve destination metadata 超时而失败。

默认情况下, bulkcopy() 行中的每个值按序数位置映射到表列,因此每列必须有顺序的值。 当表有身份列或只填充部分列时,明确传入 column_mappings 目标列名。 否则数值会移到错误的列,导致 bulkcopy() 失败。

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"]

二进制数据操作

使用Microsoft SQL函数HASHBYTES比较二进制内容,无需获取完整值。

比较二进制数据

通过计算和比较 Microsoft SQL Server 中的内容哈希,找到共享相同内容的二进制行:

# 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

在 Python 中计算哈希

用 Python 计算 SHA-256 哈希值,并将其与二进制数据一起存储:

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})

编码和解码二进制数据

将二进制数据转换为 Base64 编码字符串,或从 Base64 编码字符串转换为二进制数据:

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")

处理特定二进制格式

这些示例展示了如何在存储前验证和插入常见文件格式。

PDF 文档

插入前请核实PDF文件签名:

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}
    )

带有PIL/Pillow的图片

在存储图片前使用Pillow(PIL)调整大小以节省空间:

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))

压缩数据

通过插入前用 gzip 压缩二进制数据来减少存储空间:

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)

最佳做法

应用这些指南,为二进制数据选择合适的列类型和存储策略。

使用合适的柱类型

根据数据特性选择列类型:

对于不同的二进制用例,请选择合适的 Microsoft SQL 数据类型:

-- 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

对于大型文件,可考虑使用 FILESTREAM

对于大文件(超过1 MB),可以考虑Microsoft的SQL FILESTREAM功能,该功能将数据存储在文件系统中。 FILESTREAM 在使用前需要 服务器端配置

# 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")

验证二进制数据

在存储二进制数据前,通过检查魔字节(文件签名)来验证文件类型:

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})