使用 SQLAlchemy ORM 设置比较

Question

查看 OReilly 的一本旧书 Transact SQL Cookbook，其中一节描述了如何在原始 SQL 中执行集合比较。

CREATE TABLE branch_book_list (
   branch_name CHAR(10),
   book_ISBN CHAR (13),
   book_name CHAR(40),
   PRIMARY KEY (branch_name, book_ISBN)

表：

branch_name book_ISBN     book_name                                
----------- ------------- ---------------------------------------- 
Branch A    1-56592-401-0 Transact-SQL Programming                
Branch A    1-56592-578-5 Oracle SQL*Plus: The Definitive Guide   
Branch A    1-56592-756-7 Transact-SQL Cookbook   

Branch B    1-56592-401-0 Transact-SQL Programming                
Branch B    1-56592-756-7 Transact-SQL Cookbook                   
Branch B    1-56592-948-9 Oracle SQL*Loader: The Definitive Guide

查找不在其他集中的行 首先，我们编写一个查询来查找分行 A 持有但分行 B 不持有的所有书籍。使用 SQL Server，我们可以通过使用子查询来实现此目标，如下所示：

SELECT bbl1.*
FROM branch_book_list bbl1
WHERE branch_name = 'Branch A'
  AND NOT EXISTS (
      SELECT bbl2.*
      FROM branch_book_list bbl2
      WHERE branch_name = 'Branch B' 
        AND bbl1.book_ISBN = bbl2.book_ISBN 
        AND bbl1.book_name = bbl2.book_name)

输出：

branch_name book_ISBN     book_name                                
----------- ------------- ---------------------------------------- 
Branch A    1-56592-578-5 Oracle SQL*Plus: The Definitive Guide

现在的问题是，

使用ORM框架的SQLAlchemy中对应的查询应该是怎样的？

Answer 1

你的模型会是这样的：

class BranchBookList(Base):

    __tablename__ = 'branch_book_list'

    branch_name = Column(Unicode)
    book_ISBN = Column(Unicode)
    book_name = Column(Unicode)

    __table_args__ = (PrimaryKeyConstraint(branch_name, book_ISBN),)

进入带有模型的模块，您应该首先定义

Session

和

Base

。然后是与建议结果相同的查询：

>>> from sqlalchemy.orm.util import aliased, and_
>>> from module_with_bbl import BranchBookList, Session
>>> bbl1 = Session.query(BranchBookList).filter(BranchBookList.branch_name == 'Branch A').subquery()
>>> bbl2 = Session.query(BranchBookList).filter(BranchBookList.branch_name == 'Branch B').subquery()
>>> query = Session.query(bbl1).outerjoin(bbl2, and_(bbl1.c.book_name == bbl2.c.book_name, bbl1.c.book_ISBN == >>> bbl2.c.book_ISBN)).filter(bbl2.c.book_ISBN == None)
>>> print(query)
SELECT anon_1.branch_name AS anon_1_branch_name, anon_1."book_ISBN" AS "anon_1_book_ISBN", anon_1.book_name AS anon_1_book_name 
FROM (SELECT branch_book_list.branch_name AS branch_name, branch_book_list."book_ISBN" AS "book_ISBN", branch_book_list.book_name AS book_name 
FROM branch_book_list 
WHERE branch_book_list.branch_name = :branch_name_1) AS anon_1 LEFT OUTER JOIN (SELECT branch_book_list.branch_name AS branch_name, branch_book_list."book_ISBN" AS "book_ISBN", branch_book_list.book_name AS book_name 
FROM branch_book_list 
WHERE branch_book_list.branch_name = :branch_name_2) AS anon_2 ON anon_1.book_name = anon_2.book_name AND anon_1."book_ISBN" = anon_2."book_ISBN" 
WHERE anon_2."book_ISBN" IS NULL
>>> query.all()
[(u'Branch A', u'1-56592-578-5', u'Oracle SQL*Loader: The Definitive Guide')]

我几乎 100% 确信我的版本更好，因为建议查询中的子查询将为

branch_book_table

中的每一行执行。因此，如果表中有数千行，则有数千个额外的子查询，而我的查询只有两个查询。

对不起我的英语:)

Answer 2

假设您的 RDBMS 支持它们，您可以使用 SQL 集合操作 - 在本例中

EXCEPT

- 来查找两个集合之间的差异。

# Select the Branch A set.
q_a = sa.select(tbl.c.book_ISBN, tbl.c.book_name).where(tbl.c.branch_name == 'Branch A')

# Select the Branch B set.
q_b = sa.select(tbl.c.book_ISBN, tbl.c.book_name).where(tbl.c.branch_name == 'Branch B')

# Select the diffence between the two sets.
q = q_a.except_(q_b).subquery()
qq = sa.select(
    sa.literal('Branch A').label('branch_name'), q.c.book_ISBN, q.c.book_name
)

with engine.connect() as conn:
    rows = conn.execute(q)
    for row in rows.mappings():
        print(row)

发出的 SQL：

SELECT tbl."book_ISBN", tbl.book_name 
FROM tbl 
WHERE tbl.branch_name = 'Branch A' EXCEPT SELECT tbl."book_ISBN", tbl.book_name 
FROM tbl 
WHERE tbl.branch_name = 'Branch B'

上面的查询不会输出分支名称，这可能没问题，因为根据定义，结果将来自“分支 A”查询。如果需要分支名称：

q = q_a.except_(q_b).subquery()
q = sa.select(
    sa.literal('Branch A').label('branch_name'), q.c.book_ISBN, q.c.book_name
)

这只会输出 ISBN 和名称：这可能并不重要，因为

有关对称差异，请参阅此处的示例（PostgreSQl，但应与 MSSQL 一起使用）。