使用 ORM、声明式样式和关联对象在 SQLAlchemy 中递归 select(深度有限)关系
Recursively select (with limited depth) relationships in SQLAlchemy using ORM, declarative style, and Association objects
给定:
DIRECTIONS = db.Enum('N', 'NE', 'E', 'SE', 'S', 'SW', 'W', 'NW',
name='directions')
class Exit(BaseModel):
__tablename__ = 'exits'
src = db.Column(db.Integer, db.ForeignKey('room.id'), primary_key=True)
dst = db.Column(db.Integer, db.ForeignKey('room.id'), primary_key=True)
direction = db.Column(DIRECTIONS, primary_key=True)
from_room = db.relationship('Room', foreign_keys=[dst],
backref=db.backref('exits',
lazy='dynamic'))
to_room = db.relationship('Room', foreign_keys=[src]))
使用:
SqlAlchemy 0.9.8,PostgreSQL9.3.5
如何递归查询 select 退出到某个深度 n
给定起始房间 r
?
示例(简单地图):
E
/
B C -- D
\ /
A
|
F
/ \
G H
\
I
假设关系可以用上面的地图表示,
- 什么查询会给我所有房间,从任意房间开始?
- 如何将上述查询限制为原始房间中一定数量的 'hops'?
- 例如,从
A
开始,select 所有房间,除了 E
和 I
,将深度限制为 2。
当然可以Python:
rooms = [starting_room]
for exit in starting_room.exits.all():
if exit.to_room not in rooms:
add_to_map(exit)
rooms.append(exit.to_room)
for room in rooms[1:]:
for exit in room.exits.all():
if exit.to_room not in rooms: add_to_map(exit)
rooms.append(exit.to_room)
但这很昂贵,不适合超过 2 跳。
我尝试按照 SQLAlchemy 文档中的 CTE 示例进行操作,但是很难理解如何使它像我一样为关联对象工作,尤其是因为我没有在纯 SQL.
中使用 CTE 的经验
我的尝试:
>>> starting_exits = db.session.query(Exit).filter_by(from_room=starting_room).came='starting_exits', recursive=True)
>>> start = orm.aliased(starting_exits, name='st')
>>> exit = orm.aliased(Exit, name='e')
>>> cte = starting_exits.union_all(db.session.query(exit).filter(exit.src == start.c.dst))
>>> db.session.query(cte).all()
无限期挂起,即使我将其切片 (.all()[:5]
)。我应该做什么?
我希望我没有使您的模型过于复杂,但为了测试查询(如下所示),我使用了以下模型定义:
模特:
class Room(Base):
__tablename__ = 'room'
id = Column(Integer, primary_key=True)
name = Column(String)
exits = association_proxy(
'lnk_exits', 'to_room',
# creator=lambda v: Exit(to_room=v),
creator=lambda k, v: Exit(direction=k, to_room=v),
)
entries = association_proxy(
'lnk_entries', 'from_room',
# creator=lambda v: Exit(from_room=v),
creator=lambda k, v: Exit(direction=k, from_room=v),
)
class Exit(Base):
__tablename__ = 'exits'
src = Column(Integer, ForeignKey('room.id'), primary_key=True)
dst = Column(Integer, ForeignKey('room.id'), primary_key=True)
direction = Column(DIRECTIONS, primary_key=True)
from_room = relationship(
Room, foreign_keys=[dst],
# backref='lnk_exits',
backref=backref(
"lnk_exits",
collection_class=attribute_mapped_collection("direction"),
cascade="all, delete-orphan",
)
)
to_room = relationship(
Room,
foreign_keys=[src],
# backref='lnk_entries',
backref=backref(
"lnk_entries",
collection_class=attribute_mapped_collection("direction"),
cascade="all, delete-orphan",
)
)
你确实不需要像我那样使用关系,但我喜欢我这样做的方式,因为它允许我处理房间之间的关系下面:
# Insert test data
rooms = [Room(name=name) for name in 'ABCDEFGHI']
session.add_all(rooms)
A, B, C, D, E, F, G, H, I = rooms
A.entries = {'NW': B, 'NE': C, 'S': F}
B.entries = {'SE': A}
C.entries = {'E': D, 'SW': A}
D.entries = {'W': C, 'NE': E}
E.entries = {'SW': D}
F.entries = {'N': A, 'SW': G, 'SE': H}
G.entries = {'NE': F}
H.entries = {'NW': F, 'SE': I}
if True: # add cycle, in which case we get duplicates in the results
B.entries['E'] = C
C.entries['W'] = B
session.commit()
您可以在文档的 Association Proxy
部分阅读更多相关信息。
现在 QUERY
请注意,为了使用下面的查询,您不需要上面的任何关联代理和相关内容。即使有简单的关系 A <--> B
当前查询也会挂起,因为您 CTE
将无限期地来回导航它。所以诀窍是在CTE
中添加level
信息,这样就可以在层级上限制搜索。下面的查询应该让你开始:
# parameters
start_id = session.query(Room).filter(Room.name == 'A').first().id
max_level = 2
# CTE definition
starting_exits = (session.query(Exit, literal(0).label("level"))
.filter(Exit.src == start_id)
.cte(name="starting_exits", recursive=True)
)
start = aliased(starting_exits, name="st")
exit = aliased(Exit, name="e")
joined = (session.query(exit, (start.c.level + 1).label("level"))
.filter(exit.src == start.c.dst)
# @note: below line will avoid simple cycles of 2, which does not always help, but should reduce the result-set significantly already
.filter(exit.dst != start.c.src)
.filter(start.c.level < max_level)
)
cte = start.union_all(joined)
for x in session.query(cte).order_by(cte.c.src, cte.c.dst, cte.c.level):
print(x)
我假设您只对结果查询的第二列 (dst
) 感兴趣,以便获得您可以访问的 Room
中的 id
。为了找到通往该房间的 最短 路径,第四列 (level
) 可能也很有趣。但是您可能仍然有多种方法可以到达同一个目标房间,所以请将这些方法过滤掉 post.
编辑: 使用 cte
获取房间(模型实例)的简单方法是:
# get all Rooms (no duplicates)
s = session.query(cte.c.dst.distinct().label("room_id")).subquery(name="rooms")
q = session.query(Room).join(s, Room.id == s.c.room_id)
for r in q:
print(r)
编辑 2: 要获取出口和房间(模型实例)以重建图形,只需对上面的查询进行另一个连接:
exits = (session.query(Exit, Room)
.join(s, Exit.dst == s.c.room_id)
.join(Room, Room.id == s.c.room_id))
for exit, room in exits:
print exit, room
给定:
DIRECTIONS = db.Enum('N', 'NE', 'E', 'SE', 'S', 'SW', 'W', 'NW',
name='directions')
class Exit(BaseModel):
__tablename__ = 'exits'
src = db.Column(db.Integer, db.ForeignKey('room.id'), primary_key=True)
dst = db.Column(db.Integer, db.ForeignKey('room.id'), primary_key=True)
direction = db.Column(DIRECTIONS, primary_key=True)
from_room = db.relationship('Room', foreign_keys=[dst],
backref=db.backref('exits',
lazy='dynamic'))
to_room = db.relationship('Room', foreign_keys=[src]))
使用:
SqlAlchemy 0.9.8,PostgreSQL9.3.5
如何递归查询 select 退出到某个深度 n
给定起始房间 r
?
示例(简单地图):
E
/
B C -- D
\ /
A
|
F
/ \
G H
\
I
假设关系可以用上面的地图表示,
- 什么查询会给我所有房间,从任意房间开始?
- 如何将上述查询限制为原始房间中一定数量的 'hops'?
- 例如,从
A
开始,select 所有房间,除了E
和I
,将深度限制为 2。
- 例如,从
当然可以Python:
rooms = [starting_room]
for exit in starting_room.exits.all():
if exit.to_room not in rooms:
add_to_map(exit)
rooms.append(exit.to_room)
for room in rooms[1:]:
for exit in room.exits.all():
if exit.to_room not in rooms: add_to_map(exit)
rooms.append(exit.to_room)
但这很昂贵,不适合超过 2 跳。
我尝试按照 SQLAlchemy 文档中的 CTE 示例进行操作,但是很难理解如何使它像我一样为关联对象工作,尤其是因为我没有在纯 SQL.
中使用 CTE 的经验我的尝试:
>>> starting_exits = db.session.query(Exit).filter_by(from_room=starting_room).came='starting_exits', recursive=True)
>>> start = orm.aliased(starting_exits, name='st')
>>> exit = orm.aliased(Exit, name='e')
>>> cte = starting_exits.union_all(db.session.query(exit).filter(exit.src == start.c.dst))
>>> db.session.query(cte).all()
无限期挂起,即使我将其切片 (.all()[:5]
)。我应该做什么?
我希望我没有使您的模型过于复杂,但为了测试查询(如下所示),我使用了以下模型定义:
模特:
class Room(Base):
__tablename__ = 'room'
id = Column(Integer, primary_key=True)
name = Column(String)
exits = association_proxy(
'lnk_exits', 'to_room',
# creator=lambda v: Exit(to_room=v),
creator=lambda k, v: Exit(direction=k, to_room=v),
)
entries = association_proxy(
'lnk_entries', 'from_room',
# creator=lambda v: Exit(from_room=v),
creator=lambda k, v: Exit(direction=k, from_room=v),
)
class Exit(Base):
__tablename__ = 'exits'
src = Column(Integer, ForeignKey('room.id'), primary_key=True)
dst = Column(Integer, ForeignKey('room.id'), primary_key=True)
direction = Column(DIRECTIONS, primary_key=True)
from_room = relationship(
Room, foreign_keys=[dst],
# backref='lnk_exits',
backref=backref(
"lnk_exits",
collection_class=attribute_mapped_collection("direction"),
cascade="all, delete-orphan",
)
)
to_room = relationship(
Room,
foreign_keys=[src],
# backref='lnk_entries',
backref=backref(
"lnk_entries",
collection_class=attribute_mapped_collection("direction"),
cascade="all, delete-orphan",
)
)
你确实不需要像我那样使用关系,但我喜欢我这样做的方式,因为它允许我处理房间之间的关系下面:
# Insert test data
rooms = [Room(name=name) for name in 'ABCDEFGHI']
session.add_all(rooms)
A, B, C, D, E, F, G, H, I = rooms
A.entries = {'NW': B, 'NE': C, 'S': F}
B.entries = {'SE': A}
C.entries = {'E': D, 'SW': A}
D.entries = {'W': C, 'NE': E}
E.entries = {'SW': D}
F.entries = {'N': A, 'SW': G, 'SE': H}
G.entries = {'NE': F}
H.entries = {'NW': F, 'SE': I}
if True: # add cycle, in which case we get duplicates in the results
B.entries['E'] = C
C.entries['W'] = B
session.commit()
您可以在文档的 Association Proxy
部分阅读更多相关信息。
现在 QUERY
请注意,为了使用下面的查询,您不需要上面的任何关联代理和相关内容。即使有简单的关系 A <--> B
当前查询也会挂起,因为您 CTE
将无限期地来回导航它。所以诀窍是在CTE
中添加level
信息,这样就可以在层级上限制搜索。下面的查询应该让你开始:
# parameters
start_id = session.query(Room).filter(Room.name == 'A').first().id
max_level = 2
# CTE definition
starting_exits = (session.query(Exit, literal(0).label("level"))
.filter(Exit.src == start_id)
.cte(name="starting_exits", recursive=True)
)
start = aliased(starting_exits, name="st")
exit = aliased(Exit, name="e")
joined = (session.query(exit, (start.c.level + 1).label("level"))
.filter(exit.src == start.c.dst)
# @note: below line will avoid simple cycles of 2, which does not always help, but should reduce the result-set significantly already
.filter(exit.dst != start.c.src)
.filter(start.c.level < max_level)
)
cte = start.union_all(joined)
for x in session.query(cte).order_by(cte.c.src, cte.c.dst, cte.c.level):
print(x)
我假设您只对结果查询的第二列 (dst
) 感兴趣,以便获得您可以访问的 Room
中的 id
。为了找到通往该房间的 最短 路径,第四列 (level
) 可能也很有趣。但是您可能仍然有多种方法可以到达同一个目标房间,所以请将这些方法过滤掉 post.
编辑: 使用 cte
获取房间(模型实例)的简单方法是:
# get all Rooms (no duplicates)
s = session.query(cte.c.dst.distinct().label("room_id")).subquery(name="rooms")
q = session.query(Room).join(s, Room.id == s.c.room_id)
for r in q:
print(r)
编辑 2: 要获取出口和房间(模型实例)以重建图形,只需对上面的查询进行另一个连接:
exits = (session.query(Exit, Room)
.join(s, Exit.dst == s.c.room_id)
.join(Room, Room.id == s.c.room_id))
for exit, room in exits:
print exit, room