我不是100%确定PostgreSQL的语法，但是在BudgetCategories模型中添加以下内容应该可以使用CheckConstraint：

class BudgetCategories(Base):
    __tablename__ = 'budget_categories'
    # ...

    # @note: new
    __table_args__ = (
            CheckConstraint('NOT(category IS NULL AND parent_category IS NULL)'),
            )

我在SQLAlchemy模型中需要XOR行为。我提出了以下定义（使用后端：PostgreSQL）：

from sqlalchemy.schema import (
    CheckConstraint
)

class ScheduledNotebook(Base):
    __table_args__ = (
        (CheckConstraint('(uuid::text IS NULL) <> (notebook_path IS NULL)', name='uuid_xor_notebook_path')),
    )

    id = Column(Integer, primary_key=True)
    notebook_path = Column(String, nullable=True, unique=True)
    uuid = Column(UUID(as_uuid=True), primary_key=True, unique=True, nullable=True)

和随后的Alembic迁移（注意：自动生成不会检测到它-您必须手动添加它）：

def upgrade():
    op.create_check_constraint(
        'uuid_xor_notebook_path',
        table_name='scheduled_notebooks',
        schema='metadata',
        condition='(uuid::text IS NULL) <> (notebook_path IS NULL)'
    )

def downgrade():
    op.drop_constraint('uuid_xor_notebook_path')

它的作用就像一个魔咒：

-仅笔记本路径-确定

datalake=#  INSERT INTO scheduled_notebooks (schedule,enabled,owner, notebook_path) VALUES ('{"kind":"hourly"}',true,'akos', '/a/b/c/d/e.ipynb');
INSERT 0 1

-仅uuid -确定

datalake=#  INSERT INTO scheduled_notebooks (schedule,enabled,owner, uuid) VALUES ('{"kind":"hourly"}',true,'akos', '7792bd5f-5819-45bf-8902-8cf43102434d');
INSERT 0 1

-uuid和notebook_path -依需要失败

datalake=#  INSERT INTO scheduled_notebooks (schedule,enabled,owner, uuid, notebook_path) VALUES ('{"kind":"hourly"}',true,'akos', '7792bd5f-5819-45bf-8902-8cf43102434f', '/a/b/c/d');
ERROR:  new row for relation "scheduled_notebooks" violates check constraint "uuid_xor_notebook_path"
DETAIL:  Failing row contains (567, /a/b/c/d, {"kind": "hourly"}, t, akos, null, null, null, 7792bd5f-5819-45bf-8902-8cf43102434f).

-既不是uuid也不是notebook_path -按要求失败

datalake=#  INSERT INTO scheduled_notebooks (schedule,enabled,owner) VALUES ('{"kind":"hourly"}',true,'akos');
ERROR:  new row for relation "scheduled_notebooks" violates check constraint "uuid_xor_notebook_path"
DETAIL:  Failing row contains (568, null, {"kind": "hourly"}, t, akos, null, null, null, null).

我希望还不算太晚，但这应该可以了，它被检查为与PostGreSQL DB一起工作：

class BudgetCategories(Base):
    __tablename__ = 'budget_categories'
    __table_args__ = (
        CheckConstraint('coalesce(category , parent_category ) is not null'),
    )
    # ...

当使用PostgreSQL后端时，您也可以使用num_nonnulls函数来实现此目的：

class BudgetCategories(Base):
    __tablename__ = 'budget_categories'
    __table_args__ = (
        CheckConstraint('num_nonnulls(category, parent_category) = 1'),
    )

几周前我也遇到了这个问题，并在这里提供了解决方案。所以我猜翻译成sqlalchemy（还没有测试！！）：

from sqlalchemy.schema import CheckConstraint

class MyClass(Base):
    __table_args__ = (
        (CheckConstraint(
            'num_nulls(nullif(trim(col_a::text), ''), nullif(trim(col_b::text), '')) = 1', 
            name='uuid_xor_notebook_path')),
    )
    id = Column(Integer, primary_key=True)
    col_a = Column(String, nullable=True, unique=True)
    col_b = Column(String, nullable=True, unique=True)

然而出于美观的原因，我更喜欢@andilabs解决方案。为了仍然考虑空字段，我们可以使用另一个solution from here并编写：

from sqlalchemy.schema import CheckConstraint

class MyClass(Base):
    __table_args__ = (
        (CheckConstraint(
            '(col_a::text IS NULL) <> (col_b IS NULL)', 
            name='uuid_xor_notebook_path')),
    )
    id = Column(Integer, primary_key=True)
    col_a = Column(String, nullable=True, unique=True)
    col_b = Column(String, nullable=True, unique=True)

    @validates('col_a', 'col_b')
    def empty_string_to_null(self, key, value):
        if isinstance(value, str) and value == '':
            return None
        else:
            return value

不是所有片段中最短的，我承认...