• SQLAlchemy


    一.、介绍

    SQLAlchemy是一个基于Python实现的ORM框架。该框架建立在 DB API之上,使用关系对象映射进行数据库操作,简言之便是:将类和对象转换成SQL,然后使用数据API执行SQL并获取执行结果。

    pip3 install sqlalchemy

    • Engine,框架的引擎
    • Connection Pooling ,数据库连接池
    • Dialect,选择连接数据库的DB API种类
    • Schema/Types,架构和类型
    • SQL Exprression Language,SQL表达式语言

    SQLAlchemy本身无法操作数据库,其必须依赖pymsql等第三方插件,Dialect用于和数据API进行交流,根据配置文件的不同调用不同的数据库API,从而实现对数据库的操作,如:

    # MySQL-Python
        mysql+mysqldb://user:password@host[:port]/dbname  
    # pymysql
        mysql+pymysql://username:password@host/dbname[?options]
    # MySQL-Connector
        mysql+mysqlconnector://user:password@host[:port]/dbname  
    # cx_Oracle
        oracle+cx_oracle://user:pass@host:port/dbname[?key=value&key=value...]  
    # 更多:http://docs.sqlalchemy.org/en/latest/dialects/index.html

    二、使用

    2.1 执行原生SQL语句

    import threading
    from sqlalchemy import create_engine
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有线程最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    
    
    def task(arg):
        conn = engine.raw_connection()
        cursor = conn.cursor()
        cursor.execute(
            "select * from userinfo"
        )
        ret = cursor.fetchall()
        print(ret)
        cursor.close()
        conn.close()
    
    
    for i in range(20):
        t = threading.Thread(target=task, args=(i,))
        t.start()
    View Code

    2.2 ORM

    创建数据库表

    from sqlalchemy.ext.declarative import declarative_base
    from sqlalchemy import Column
    from sqlalchemy import Integer, String, Date, UniqueConstraint, Index
    from sqlalchemy import create_engine
    
    Base = declarative_base()
    
    
    class UserInfo(Base):
        __tablename__ = "UserInfo"
    
        id = Column(Integer, primary_key=True)
        name = Column(String(16), index=True, unique=True, nullable=False)
        age = Column(Integer)
        birthday = Column(Date)
    
        __table_args__ = (
            # UniqueConstraint("id", "name", name="unic_id_name"),  # 联合唯一索引
            # Index("idx_age_birthday", "age", "birthday"),  # 联合索引
        )
    
    def init_db():
        """根据类创建数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,  # 超过连接池大小外最多创建的连接
            pool_size=5,     # 连接池大小
            pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
            pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
        )
        Base.metadata.create_all(engine)
    
    
    def drop_db():
        """根据类删除数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,
            pool_size=5,
            pool_timeout=20,
            pool_recycle=-1
        )
        Base.metadata.drop_all(engine)
    
    
    if __name__ == "__main__":
        drop_db()
        init_db()
    创建单表

    操作数据库表(简单示例)

    import datetime
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    from models import UserInfo
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    obj = UserInfo(name="pd", age=18, birthday=datetime.date.today())
    session.add(obj)
    
    # 提交事务
    session.commit()
    # 关闭session
    session.close()

    PS:要想查看SQL语句,去掉all(),打印查询语句即可。

    基本增删改查操作

    import datetime
    from sqlalchemy import create_engine
    from sqlalchemy.sql import text
    from sqlalchemy.orm import sessionmaker
    from models import UserInfo
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    #################### 增 ####################
    """
    # 新增单条数据
    obj = UserInfo(name="pd", age=18, birthday=datetime.date.today())
    session.add(obj)
    # 新增多条数据
    session.add_all([
        UserInfo(name="盲僧", age=10, birthday=datetime.date.today()),
        UserInfo(name="妖姬", age=20, birthday=datetime.date.today())
    ])
    """
    #################### 删 ####################
    """
    session.query(UserInfo).filter(UserInfo.id==1).delete()
    """
    #################### 改 ####################
    """
    session.query(UserInfo).filter(UserInfo.name=="pd").update({UserInfo.name: "xx"})
    session.query(UserInfo).filter(UserInfo.name=="pd").update({"name": "xx"})
    session.query(UserInfo).filter(UserInfo.name=="pd").update({"age": UserInfo.age+10})
    session.query(UserInfo).filter(UserInfo.name=="pd").update({"name": UserInfo.name+"真帅"}, synchronize_session=False)  # 连接字符串
    """
    #################### 查 ####################
    """
    ret1 = session.query(UserInfo).all()
    ret2 = session.query(UserInfo.name.label("n"), UserInfo.age).all()  # label相当于起别名
    for row in ret2:
        print(row.n, row.age)  # [('pd', 10), ('盲僧', 10), ('妖姬', 20)]
    ret3 = session.query(UserInfo).filter(UserInfo.age==10).all()
    ret4 = session.query(UserInfo).filter_by(age=10).all()
    ret5 = session.query(UserInfo).filter_by(age=10).first()
    ret6 = session.query(UserInfo).from_statement(text("select * from userinfo where name=:name")).params(name="pd").all()
    for row in ret3:
        print(row.name)
    """
    # 提交事务(查询操作不需要)
    # session.commit()
    # 关闭session
    session.close()
    View Code

    常用操作(条件、与、或、排序、分组......)

    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    from models import UserInfo
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    #################### 指定查询哪列 ####################
    ret1 = session.query(UserInfo.id.label("nid"), UserInfo.name).all()
    for row in ret1:
        print(row.nid, row.name)
    #################### 条件 ####################
    ret1 = session.query(UserInfo).filter_by(name="盲僧").all()
    ret2 = session.query(UserInfo).filter(UserInfo.id > 1, UserInfo.name == "盲僧").all()
    ret3 = session.query(UserInfo).filter(UserInfo.id.between(1, 3), UserInfo.name == "盲僧").all()
    ret4 = session.query(UserInfo).filter(UserInfo.id.in_([1, 3])).all()   # 查询id在什么范围
    ret5 = session.query(UserInfo).filter(~UserInfo.id.in_([1, 3])).all()  # 查询id不在什么范围
    #################### 子查询 ####################
    ret1 = session.query(UserInfo).filter(UserInfo.id.in_(session.query(UserInfo.id).filter_by(name=""))).all()
    #################### 与、或 ####################
    from sqlalchemy import and_, or_
    ret1 = session.query(UserInfo).filter(and_(UserInfo.id > 3, UserInfo.name == "提莫")).all()
    ret2 = session.query(UserInfo).filter(or_(UserInfo.id < 3, UserInfo.name == "提莫")).all()
    for row in ret2:
        print(row.name)
    ret3 = session.query(UserInfo).filter(
        or_(
            UserInfo.id < 3,
            and_(UserInfo.id > 3, UserInfo.name == "提莫")
        )
    ).all()
    for row in ret3:
        print(row.name)
    #################### 通配符 ####################
    ret1 = session.query(UserInfo).filter(UserInfo.name.like("盲%")).first()
    ret2 = session.query(UserInfo).filter(UserInfo.name.like("盲%")).all()
    ret3 = session.query(UserInfo).filter(~UserInfo.name.like("盲%")).all()
    ################### 限制(切片) ####################
    ret1 = session.query(UserInfo)[1:2]  # 顾头不顾尾
    #################### 排序 ####################
    # desc()倒序  asc()正序
    ret1 = session.query(UserInfo).order_by(UserInfo.id.desc()).all()
    ret2 = session.query(UserInfo).order_by(UserInfo.id.desc(), UserInfo.name.asc()).all()
    #################### 分组 ####################
    from sqlalchemy.sql import func
    ret1 = session.query(UserInfo).group_by(UserInfo.age).all()
    ret2 = session.query(
        func.max(UserInfo.id),
        func.min(UserInfo.id)).group_by(UserInfo.age).all()
    for item in ret2:
        print(item)
    ret3 = session.query(
        UserInfo.depart_id,
        func.count(UserInfo.id)
    ).group_by(UserInfo.depart_id).having(func.count(UserInfo.id) >= 2).all()
    for item in ret3:
        print(item)
    #################### 连表 ####################
    session.query(Book).join(Author).all()
    session.query(tb1).join(tb2).join(tb3).all()
    
    session.query(Book, Author).join(Author).all()
    session.query(Book, Author).join(Author, Author.id == Book.author_id).all()
    session.query(Book.id, Book.title, Author.name).join(Author).all()
    session.query(Book.id, Book.title, Author.name).join(Author, Author.id == Book.author_id).all()
    session.query(Book, Author).join(Author, isouter=True).all()
    #################### 组合(两张表的数据组合到一起) ####################
    # union()去重  union_all()不去重
    q1 = session.query(UserInfo.name).filter(UserInfo.id < 3)
    q2 = session.query(UserInfo.name).filter(UserInfo.id > 3)
    ret1 = q1.union(q2).all()
    q3 = session.query(UserInfo.name).filter(UserInfo.id < 3 )
    q4 = session.query(UserInfo.name).filter(UserInfo.id < 3 )
    ret2 = q3.union_all(q4).all()
    
    # 提交事务(查询操作不需要)
    # session.commit()
    # 关闭session
    session.close()
    View Code

    创建多个表并包含Fk关系以及基于relationship操作FK

    from sqlalchemy.ext.declarative import declarative_base
    from sqlalchemy import Column
    from sqlalchemy import Integer, String, ForeignKey
    from sqlalchemy import create_engine
    from sqlalchemy.orm import relationship
    
    Base = declarative_base()
    
    
    class Publisher(Base):
        __tablename__ = "publisher"
    
        id = Column(Integer, primary_key=True)
        name = Column(String(16), index=True, nullable=False)
    
    
    class Book(Base):
        __tablename__ = "book"
    
        id = Column(Integer, primary_key=True)
        title = Column(String(16), index=True, nullable=False)
        publisher_id = Column(Integer, ForeignKey("publisher.id"))
        # 与生成表结构无关,不会在数据库生成列,仅用于查询方便
        publisher = relationship("Publisher", backref="books")
    
    
    def init_db():
        """根据类创建数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,  # 超过连接池大小外最多创建的连接
            pool_size=5,     # 连接池大小
            pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
            pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
        )
        Base.metadata.create_all(engine)
    
    
    def drop_db():
        """根据类删除数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,
            pool_size=5,
            pool_timeout=20,
            pool_recycle=-1
        )
        Base.metadata.drop_all(engine)
    
    
    if __name__ == "__main__":
        drop_db()
        init_db()
    models.py
    from models import Book, Publisher
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    # 查询所有书籍
    ret = session.query(Book).all()
    for row in ret:
        print(row.id, row.title)
    
    #################### 基于join操作FK ####################
    # 查询所有书籍及其出版社
    # 方式1:
    ret1 = session.query(Book, Publisher).join(Publisher).all()
    # ret1 = session.query(Book, Publisher).join(Publisher, Publisher.id == Book.publisher_id).all()
    for row in ret1:
        print(row[0].id, row[0].title, row[1].name)
    
    # 方式2:
    ret2= session.query(Book.id, Book.title, Publisher.name).join(Publisher).all()
    # ret2 = session.query(Book.id, Book.title, Publisher.name).join(Publisher, Publisher.id == Book.publisher_id).all()
    for row in ret2:
        print(row.id, row.title, row.name)
    
    #################### 基于relationship操作FK ####################
    # 查询所有书籍及其出版社名
    ret3 = session.query(Book).all()
    for row in ret3:
        print(row.id, row.title, row.publisher.name)
    
    # 查询"苹果出版社"出版的所有书籍
    obj = session.query(Publisher).filter(Publisher.name == "苹果出版社").first()
    for row in obj.books:
        print(row.id, row.title, obj.name)
    
    #################### 基于relationship的增操作 ####################
    # 添加"橘子出版社",再给该出版社添加一本书"你在干嘛?"
    obj = Book(title="你在干嘛?", publisher=Publisher(name="橘子出版社"))
    session.add(obj)
    
    # 添加"樱桃出版社",再给该出版社添加多本书"你看不到我"、"我看到你了"
    # 可以根据relationship中的backref反向操作
    obj = Publisher(name="樱桃出版社")
    obj.books = [Book(title="你看不到我"), Book(title="我看到你了")]
    session.add(obj)
    
    session.commit()
    # 关闭session
    session.close()
    View Code

    创建多个表并包含M2M关系以及基于relationship操作M2M

    from sqlalchemy.ext.declarative import declarative_base
    from sqlalchemy import Column
    from sqlalchemy import Integer, String, ForeignKey, UniqueConstraint, Index
    from sqlalchemy import create_engine
    from sqlalchemy.orm import relationship
    
    Base = declarative_base()
    
    
    class Author(Base):
        __tablename__ = "author"
    
        id = Column(Integer, primary_key=True)
        name = Column(String(16), index=True, nullable=False)
        # 与生成表结构无关,不会在数据库生成列,仅用于查询方便
        book_list = relationship("Book", secondary="author2book", backref="author_list")
    
    
    class Book(Base):
        __tablename__ = "book"
    
        id = Column(Integer, primary_key=True)
        title = Column(String(16), index=True, nullable=False)
    
    
    class Author2Book(Base):
        __tablename__ = "author2book"
    
        id = Column(Integer, primary_key=True, autoincrement=True)
        author_id = Column(Integer, ForeignKey("author.id"))
        book_id = Column(Integer, ForeignKey("book.id"))
    
        __table_args__ = (
            UniqueConstraint("author_id", "book_id", name="uic_ar_bk"),  # 联合唯一索引
            # Index("idx_id_name", "id", "name"),  # 联合索引
        )
    
    
    def init_db():
        """根据类创建数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,  # 超过连接池大小外最多创建的连接
            pool_size=5,     # 连接池大小
            pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
            pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
        )
        Base.metadata.create_all(engine)
    
    
    def drop_db():
        """根据类删除数据库表"""
        engine = create_engine(
            "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
            max_overflow=0,
            pool_size=5,
            pool_timeout=20,
            pool_recycle=-1
        )
        Base.metadata.drop_all(engine)
    
    
    if __name__ == "__main__":
        drop_db()
        init_db()
    models.py
    from models import Book, Author, Author2Book
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    #################### 普通操作M2M ####################
    # 添加数据
    """
    session.add_all([
        Author(name="熊大"),
        Author(name="光头强"),
        Book(title="Python"),
        Book(title="Java"),
    ])
    session.commit()
    """
    """
    session.add_all([
       Author2Book(author_id=1, book_id=1),
       Author2Book(author_id=1, book_id=2),
       Author2Book(author_id=2, book_id=1),
    ])
    session.commit()
    """
    # 三张表关联
    """
    ret = session.query(Author2Book.id, Author.name, Book.title).join(Author, Author.id == Author2Book.author_id).join(Book, Book.id == Author2Book.book_id).order_by(Author2Book.id.asc()).all()
    for row in ret:
        print(row.id, row.name, row.title)
    # 1 熊大 Python
    # 2 熊大 Java
    # 3 光头强 Python
    """
    
    #################### 基于relationship操作M2M ####################
    # 查询"熊大"的所有书籍
    obj = session.query(Author).filter(Author.name == "熊大").first()
    for item in obj.book_list:
        print(item.id, item.title)
    # 查询"Python"这本书的所有作者
    obj = session.query(Book).filter(Book.title == "Python").first()
    for item in obj.author_list:
        print(item.id, item.name)
    # 添加作者"佩奇",再添加两本书"PHP"、"C",然后这两本书关联这个作者
    obj = Author(name="佩奇")
    obj.book_list = [Book(title="PHP"), Book(title="C")]
    session.add(obj)
    # 添加书籍"C++", 再添加两位作者"蜡笔小新"、"机器猫",然后这本书关联这两位作者
    obj = Book(title="C++")
    obj.author_list = [Author(name="蜡笔小新"), Author(name="机器猫")]
    session.add(obj)
    
    session.commit()
    session.close()
    View Code

    基于scoped_session实现线程安全

    import threading
    from models import Author
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    from sqlalchemy.orm import scoped_session
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = scoped_session(SessionFactory)
    
    
    def task(arg):
        ret = session.query(Author).all()
        for row in ret:
            print(row.name)
        session.remove()
    
    
    for i in range(20):
        t = threading.Thread(target=task, args=(i,))
        t.start()
    View Code

    执行原生sql(多种方式)

    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    # 每次执行数据库操作时,都需要创建一个session
    session = SessionFactory()
    
    cursor = session.execute("insert into author(name) values(:value)", params={"value": "花木兰"})
    print(cursor.lastrowid)
    session.commit()
    session.close()
    
    ################################################################################
    
    import threading
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    from sqlalchemy.orm import scoped_session
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    session = scoped_session(SessionFactory)
    
    
    def task(arg):
        cursor = session.execute("insert into author(name) value(:value)", params={"value": "佩奇"})
        print(cursor.lastrowid)
        session.commit()
        session.remove()
    
    
    for i in range(20):
        t = threading.Thread(target=task, args=(i,))
        t.start()
    
    ################################################################################
    
    import threading
    from sqlalchemy import create_engine
    from sqlalchemy.orm import sessionmaker
    
    engine = create_engine(
        "mysql+pymysql://root:""@127.0.0.1:3306/sqlalchemy_db?charset=utf8",
        max_overflow=0,  # 超过连接池大小外最多创建的连接
        pool_size=5,     # 连接池大小
        pool_timeout=20, # 池中没有连接最多等待的时间,否则报错
        pool_recycle=-1  # 多久之后对线程池中的线程进行一次连接的回收(重置)
    )
    SessionFactory = sessionmaker(bind=engine)
    
    
    def task(arg):
        session = SessionFactory()
        cursor = session.execute("insert into author(name) value(:value)", params={"value": "乔治"})
        print(cursor.lastrowid)
        session.commit()
        session.close()
    
    
    for i in range(20):
        t = threading.Thread(target=task, args=(i,))
        t.start()
    View Code
  • 相关阅读:
    Android学习系列(34)App应用之发布各广告平台版本
    Android拓展系列(6)CM9源码下载和编译
    Android拓展系列(8)Vim插件便捷管理(使用git submodule和pathogen)
    [Android应用]《幽默笑话》V1.0 Beta 版本发布!
    [Android应用]《养生视线》V2.1 正式版本粉墨登场!
    Android设计模式系列(9)SDK源码之适配器模式
    企业定制软件开发的两个核心问题
    持续部署才是王道
    组织模式 Introduction
    高性能计算摘要
  • 原文地址:https://www.cnblogs.com/believepd/p/10365974.html
Copyright © 2020-2023  润新知