We Keep Coding

Self referencing many-to-many relationship with extra column in association object

I am new in Sqlalchemy and trying to achieve the following goal with relationship():
There is an User table which stores user data.
Every user is able to invite other user with an invite_code.
Every user keeps a list of invitation, every invitation includes the invite_code and the invitee User
I think the relationship between User and Invitation is one-to-many. Since Invitation contains User, then I think it is probably better to use self-referential relationship to represent the inviter-to-invitaions(invitees) relationship and use an association object to store the invite_code.
I checked the sqlalchemy documentation and the question, tried to implement the classed like this:
from sqlalchemy import Column, Integer, ForeignKey, create_engine, String
from sqlalchemy.orm import relationship, sessionmaker
from sqlalchemy.ext.declarative import declarative_base
Base = declarative_base()
class Invitation(Base):
__tablename__ = 'invitation'
invite_code = Column(Integer)
inviter_id = Column(Integer, ForeignKey('user.id'), primary_key=True)
invitee_id = Column(Integer, ForeignKey('user.id'), primary_key=True)
invitee = relationship('User') #Need HELP here
class User(Base):
__tablename__ = 'user'
id = Column(Integer, primary_key=True)
name = Column(String)
inviters = relationship('User',
secondary='invitation',
primaryjoin=id==Invitation.invitee_id,
secondaryjoin=id==Invitation.inviter_id,
backref='invitees')
invitations = relationship('Invitation')# Need HELP here
def __repr__(self):
return f'User: {self.name}'
if __name__ == '__main__':
engine = create_engine('sqlite://')
Base.metadata.create_all(engine)
Session = sessionmaker(engine)
db = Session()
inviter1 = User(name='inviter1')
inviter2 = User(name='inviter2')
invitee1= User(name='invitee1')
invitee2 = User(name='invitee2')
inviter1.invitees = [invitee1, invitee2]
inviter2.invitees = [invitee1]
db.add(inviter1)
db.add(inviter2)
db.add(invitee1)
db.add(invitee2)
db.commit()
users = db.query(User).all()
for user in users:
print(user)
print(' Inviter: ', user.inviters)
print(' Invitee: ', user.invitees)
print()
If the lines with comment #Need HELP here are deleted, I can get the corresponding inviters and invitees, but cannot get the invite_code. If the #Need HELP here code are added, the error is:
Exception has occurred: AmbiguousForeignKeysError
Could not determine join condition between parent/child tables on relationship Invitation.invitee - there are multiple foreign key paths linking the tables. Specify the 'foreign_keys' argument, providing a list of those columns which should be counted as containing a foreign key reference to the parent table.
Is there a way to add extra data column in association object like association object for many-to-many relationship for self referential table?
Sorry for the too much text, I didn't find any reference document on the web.

Finally, I figured it out with the help of foreign_keys:
from sqlalchemy import Column, Integer, ForeignKey, create_engine, String
from sqlalchemy.orm import relationship, sessionmaker
from sqlalchemy.ext.declarative import declarative_base
Base = declarative_base()
class User(Base):
__tablename__ = 'user'
id = Column(Integer, primary_key=True)
name = Column(String)
sent_invitations = relationship('Invitation', foreign_keys='Invitation.inviter_id', back_populates='inviter', cascade='all, delete')
received_invitations=relationship('Invitation', foreign_keys='Invitation.invitee_id', back_populates='invitee', cascade='all, delete')
def __repr__(self):
return f'User: {self.name}'
class Invitation(Base):
__tablename__ = 'invitation'
id = Column(Integer, primary_key=True)
invite_code = Column(Integer)
inviter_id = Column(Integer, ForeignKey('user.id'))
invitee_id = Column(Integer, ForeignKey('user.id'))
inviter=relationship('User', foreign_keys=[inviter_id], back_populates='sent_invitations')
invitee=relationship('User', foreign_keys=[invitee_id], back_populates='received_invitations')
def __repr__(self):
return f'Invitation: {self.inviter} invited {self.invitee} with {self.invite_code}'
if __name__ == '__main__':
engine = create_engine('sqlite://')
Base.metadata.create_all(engine)
Session = sessionmaker(engine)
db = Session()
inviter1 = User(name='inviter1')
inviter2 = User(name='inviter2')
invitee1= User(name='invitee1')
invitee2 = User(name='invitee2')
invitation1 = Invitation(invite_code=50, inviter=inviter1, invitee=invitee1)
invitation2 = Invitation(invite_code=20, inviter=inviter2, invitee=invitee2)
invitation3 = Invitation(invite_code=22, inviter=inviter1, invitee=inviter2)
invitation4 = Invitation(invite_code=44, inviter=invitee1, invitee=inviter2)
db.add(inviter1)
db.add(inviter2)
db.add(invitee1)
db.add(invitee2)
db.commit()
users = db.query(User).all()
for user in users:
print(user)
print(' sent_invitation: ', user.sent_invitations)
print(' received_invitation: ', user.received_invitations)
print()
invitations = db.query(Invitation).all()
for invitation in invitations:
print(invitation)
db.delete(inviter1)
db.delete(invitee2)
db.commit()

Scrapy - SQLalchemy Foreign Key not created in SQLite

I tried to run Scrapy using itemLoader to collect all the data and put them into SQLite 3. I am success in gathering all the info I wanted but I cannot get the foreign keys to be generated in my ThreadInfo and PostInfo tables using back_populates with foreign key. I did try with back_ref but it also did not work.
All the other info was inserted to SQLite database after my Scrapy finished.
My goal is to have four tables, boardInfo, threadInfo, postInfo, and authorInfo linked to each others.
boardInfo will have one-to-many relationship with threadInfo
threadInfo will have one-to-many relationship with postInfo
authorInfo will have one-to-many relationship with threadInfo and
postInfo.
I used DB Browser for SQLite and found that the values of my foreign keys are Null.
I tried query for the value (threadInfo.boardInfos_id), and it displayed None. I try to fix this for many days and read through the document but cannot solve the issue.
How can I have the foriegn keys generated in my threadInfo and postInfo tables?
Thank you for all guidances and comments.
Here is my models.py
from sqlalchemy import create_engine, Column, Table, ForeignKey, MetaData
from sqlalchemy import Integer, String, Date, DateTime, Float, Boolean, Text
from sqlalchemy.orm import relationship
from sqlalchemy.ext.declarative import declarative_base
from scrapy.utils.project import get_project_settings
Base = declarative_base()
def db_connect():
'''
Performs database connection using database settings from settings.py.
Returns sqlalchemy engine instance
'''
return create_engine(get_project_settings().get('CONNECTION_STRING'))
def create_table(engine):
Base.metadata.create_all(engine)
class BoardInfo(Base):
__tablename__ = 'boardInfos'
id = Column(Integer, primary_key=True)
boardName = Column('boardName', String(100))
threadInfosLink = relationship('ThreadInfo', back_populates='boardInfosLink') # One-to-Many with threadInfo
class ThreadInfo(Base):
__tablename__ = 'threadInfos'
id = Column(Integer, primary_key=True)
threadTitle = Column('threadTitle', String())
threadLink = Column('threadLink', String())
threadAuthor = Column('threadAuthor', String())
threadPost = Column('threadPost', Text())
replyCount = Column('replyCount', Integer)
readCount = Column('readCount', Integer)
boardInfos_id = Column(Integer, ForeignKey('boardInfos.id')) # Many-to-One with boardInfo
boardInfosLink = relationship('BoardInfo', back_populates='threadInfosLink') # Many-to-One with boardInfo
postInfosLink = relationship('PostInfo', back_populates='threadInfosLink') # One-to-Many with postInfo
authorInfos_id = Column(Integer, ForeignKey('authorInfos.id')) # Many-to-One with authorInfo
authorInfosLink = relationship('AuthorInfo', back_populates='threadInfosLink') # Many-to-One with authorInfo
class PostInfo(Base):
__tablename__ = 'postInfos'
id = Column(Integer, primary_key=True)
postOrder = Column('postOrder', Integer, nullable=True)
postAuthor = Column('postAuthor', Text(), nullable=True)
postContent = Column('postContent', Text(), nullable=True)
postTimestamp = Column('postTimestamp', Text(), nullable=True)
threadInfos_id = Column(Integer, ForeignKey('threadInfos.id')) # Many-to-One with threadInfo
threadInfosLink = relationship('ThreadInfo', back_populates='postInfosLink') # Many-to-One with threadInfo
authorInfos_id = Column(Integer, ForeignKey('authorInfos.id')) # Many-to-One with authorInfo
authorInfosLink = relationship('AuthorInfo', back_populates='postInfosLink') # Many-to-One with authorInfo
class AuthorInfo(Base):
__tablename__ = 'authorInfos'
id = Column(Integer, primary_key=True)
threadAuthor = Column('threadAuthor', String())
postInfosLink = relationship('PostInfo', back_populates='authorInfosLink') # One-to-Many with postInfo
threadInfosLink = relationship('ThreadInfo', back_populates='authorInfosLink') # One-to-Many with threadInfo
Here is my pipelines.py
from sqlalchemy import exists, event
from sqlalchemy.orm import sessionmaker
from scrapy.exceptions import DropItem
from .models import db_connect, create_table, BoardInfo, ThreadInfo, PostInfo, AuthorInfo
from sqlalchemy.engine import Engine
from sqlite3 import Connection as SQLite3Connection
import logging
#event.listens_for(Engine, "connect")
def _set_sqlite_pragma(dbapi_connection, connection_record):
if isinstance(dbapi_connection, SQLite3Connection):
cursor = dbapi_connection.cursor()
cursor.execute("PRAGMA foreign_keys=ON;")
# print("####### PRAGMA prog is running!! ######")
cursor.close()
class DuplicatesPipeline(object):
def __init__(self):
'''
Initializes database connection and sessionmaker.
Creates tables.
'''
engine = db_connect()
create_table(engine)
self.Session = sessionmaker(bind=engine)
logging.info('****DuplicatesPipeline: database connected****')
def process_item(self, item, spider):
session = self.Session()
exist_threadLink = session.query(exists().where(ThreadInfo.threadLink == item['threadLink'])).scalar()
exist_thread_replyCount = session.query(ThreadInfo.replyCount).filter_by(threadLink = item['threadLink']).scalar()
if exist_threadLink is True: # threadLink is in DB
if exist_thread_replyCount < item['replyCount']: # check if replyCount is more?
return item
session.close()
else:
raise DropItem('Duplicated item found and replyCount is not changed')
session.close()
else: # New threadLink to be added to BoardPipeline
return item
session.close()
class BoardPipeline(object):
def __init__(self):
'''
Initializes database connection and sessionmaker
Creates tables
'''
engine = db_connect()
create_table(engine)
self.Session = sessionmaker(bind=engine)
def process_item(self, item, spider):
'''
Save scraped info in the database
This method is called for every item pipeline component
'''
session = self.Session()
# Input info to boardInfos
boardInfo = BoardInfo()
boardInfo.boardName = item['boardName']
# Input info to threadInfos
threadInfo = ThreadInfo()
threadInfo.threadTitle = item['threadTitle']
threadInfo.threadLink = item['threadLink']
threadInfo.threadAuthor = item['threadAuthor']
threadInfo.threadPost = item['threadPost']
threadInfo.replyCount = item['replyCount']
threadInfo.readCount = item['readCount']
# Input info to postInfos
# Due to info is in list, so we have to loop and add it.
for num in range(len(item['postOrder'])):
postInfoNum = 'postInfo' + str(num)
postInfoNum = PostInfo()
postInfoNum.postOrder = item['postOrder'][num]
postInfoNum.postAuthor = item['postAuthor'][num]
postInfoNum.postContent = item['postContent'][num]
postInfoNum.postTimestamp = item['postTimestamp'][num]
session.add(postInfoNum)
# Input info to authorInfo
authorInfo = AuthorInfo()
authorInfo.threadAuthor = item['threadAuthor']
# check whether the boardName exists
exist_boardName = session.query(exists().where(BoardInfo.boardName == item['boardName'])).scalar()
if exist_boardName is False: # the current boardName does not exists
session.add(boardInfo)
# check whether the threadAuthor exists
exist_threadAuthor = session.query(exists().where(AuthorInfo.threadAuthor == item['threadAuthor'])).scalar()
if exist_threadAuthor is False: # the current threadAuthor does not exists
session.add(authorInfo)
try:
session.add(threadInfo)
session.commit()
except:
session.rollback()
raise
finally:
session.close()
return item

From the code I can see, it doesn't look to me like you are setting ThreadInfo.authorInfosLink or ThreadInfo.authorInfos_id anywhere (the same goes for all of your FK/relationships).
For the related objects to be attached to a ThreadInfo instance, you need to create them and then attach them something like:
# Input info to authorInfo
authorInfo = AuthorInfo()
authorInfo.threadAuthor = item['threadAuthor']
threadInfo.authorInfosLink = authorInfo
You probably don't want to session.add() each object if it's related via FK. You'll want to:
instantiate a BoardInfo object bi
then instantiate attach your related ThreadInfo object ti
attach your the related object eg bi.threadInfosLink = ti
At the end of all of your chained relationships, you can simply add bi to the session using session.add(bi) -- all of the related objects will be added through their relationships and the FKs will be correct.

Per the discussion in the comments of my other answer, below is how I would rationalize your models to make them make more sense to me.
Notice:
I have removed the unnecessary "Info" everywhere
I have removed explicit column names from your model definitions and will rely instead on SQLAlchemy's ability to infer those for me based on my attribute names
In a "Post" object I do not name the attribute PostContent, it's implied that the content relates to the Post because that's how we're accessing it -- instead simply call the attribute "Post"
I've removed all "Link" terminology -- in places where I think you want a reference to a collection of related objects I've provided a plural attribute of that object as the relationship.
I've left a line in the Post model for you to remove. As you can see, you don't need "author" twice -- once as a related object and once on the Post, that defeats the purpose of the FKs.
With these changes, when you attempt to use these models from your other code it becomes obvious where you need to use .append() and where you simply assign the related object. For a given Board object you know that 'threads' is a collection just based on the attribute name, so you're going to do something like b.threads.append(thread)
from sqlalchemy import create_engine, Column, Table, ForeignKey, MetaData
from sqlalchemy import Integer, String, Date, DateTime, Float, Boolean, Text
from sqlalchemy.orm import relationship
from sqlalchemy.ext.declarative import declarative_base
class Board(Base):
__tablename__ = 'board'
id = Column(Integer, primary_key=True)
name = Column(String(100))
threads = relationship(back_populates='board')
class Thread(Base):
__tablename__ = 'thread'
id = Column(Integer, primary_key=True)
title = Column(String())
link = Column(String())
author = Column(String())
post = Column(Text())
reply_count = Column(Integer)
read_count = Column(Integer)
board_id = Column(Integer, ForeignKey('Board.id'))
board = relationship('Board', back_populates='threads')
posts = relationship('Post', back_populates='threads')
author_id = Column(Integer, ForeignKey('Author.id'))
author = relationship('Author', back_populates='threads')
class Post(Base):
__tablename__ = 'post'
id = Column(Integer, primary_key=True)
order = Column(Integer, nullable=True)
author = Column(Text(), nullable=True) # remove this line and instead use the relationship below
content = Column(Text(), nullable=True)
timestamp = Column(Text(), nullable=True)
thread_id = Column(Integer, ForeignKey('Thread.id'))
thread = relationship('Thread', back_populates='posts')
author_id = Column(Integer, ForeignKey('Author.id'))
author = relationship('Author', back_populates='posts')
class AuthorInfo(Base):
__tablename__ = 'author'
id = Column(Integer, primary_key=True)
name = Column(String())
posts = relationship('Post', back_populates='author')
threads = relationship('Thread', back_populates='author')

SQLAlchemy listen for attribute change in many-to-many relationship and change other attributes of initiator

I'm new to the SQLAlchemy and I wrote a simple CRUD database layout like this: I have three tables, Customer, Product, and Template. The idea is this: Each customer can have a template of the products he usually orders. When I fetch a particular customer from the database, his template along with all the products should be fetched as well. So I have a one-to-one relationship (customer-template) and one many-to-many relationship (template-product). Now, a template should contain fields such as quantity of a particular product, along with its net, gross and tax values. I'd like to have a listener on the quantity column, such that when the quantity of a particular product is changed, other attributes I mentioned will be changed too. So the code I wrote is as follows (please, if you can, also verify whether all the relationships are written appropriately)
from sqlalchemy import event
from sqlalchemy.orm import relationship, exc, column_property, sessionmaker
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm.query import Query as _Query
from sqlalchemy import Table, Column, Integer, String, Boolean, ForeignKey, UniqueConstraint, create_engine, Numeric
from decimal import *
# decimal operations settings
getcontext().prec = 6
getcontext().rounding = ROUND_HALF_UP
engine = create_engine('sqlite:///test.db')
Base = declarative_base()
# Initalize the database if it is not already.
Base.metadata.create_all(engine)
# Create a session to handle updates.
Session = sessionmaker(bind=engine)
# customer has a one-to-one relationship with template, where customer is the parent
class Customer(Base):
__tablename__ = "customers"
id = Column(Integer, primary_key=True)
alias = Column(String)
firm_name = Column(String)
last_name = Column(String)
first_name = Column(String)
tax_id = Column(String, nullable=False)
address = Column(String)
postal_code = Column(String)
city = Column(String)
payment = Column(Boolean)
template = relationship("Template", uselist=False, back_populates="customer")
# product has many-to-many relationship with template, where template is the parent
class Product(Base):
__tablename__ = "products"
id = Column(Integer, primary_key=True)
product_name = Column(String, nullable=False)
symbol = Column(String)
unit = Column(String, nullable=False)
unit_net_price = Column(Numeric, nullable=False)
vat_rate = Column(Numeric, nullable=False)
per_month = Column(Boolean, nullable=False)
# association table for the products-template many-to-many relationship
association_table = Table('association', Base.metadata,
Column('product_id', Integer, ForeignKey('product.id')),
Column('template_id', Integer, ForeignKey('template.id'))
)
# template implements one-to-one relationship with customer and many-to-many relationship with product
class Template(Base):
id = Column(Integer, primary_key=True)
customer_id = Column(Integer, ForeignKey("customer.id"))
customer = relationship("Customer", back_populates="template")
products = relationship("Product", secondary=association_table)
quantity = Column(Numeric)
net_val = Column(Numeric)
tax_val = Column(Numeric)
gross_val = Column(Numeric)
# TODO: implement constructor
def __init__(self, **kwargs):
self.quantity = Decimal(0.0)
self.net_val = Decimal(0.0)
self.tax_val = Decimal(0.0)
self.gross_val = Decimal(0.0)
#event.listens_for(Template.quantity, "set")
def quantity_listener(target, value, oldvalue, initiator):
print(target)
print(initiator)
# target.net_val =
# target.tax_val =
# target.gross_val =
Now, I'm unsure how should I get a particular initiator and set its values, since products in Template table is a list (?)

I'd probably do it like this. It's hard to use Sqlalchemy orm in such cases as you can't access the Session object we all got used to use.
#event.listens_for(Template, "after_update")
def quantity_listener(mapper, connection, target):
field = 'quantity'
added, _, deleted = get_history(target, field)
# added is a new values of the specified field
# deleted is the old values
# so we basically check if 'quantity' has changed
# its a tuple btw
if added and deleted and added[0] != deleted[0]:
stmt = Template.update(). \
values(Template.something=Template.something * Template.other_something). \
where(Template.id == target.id)
conn.execute(stmt)
# here goes another updates for products or you can have another
# listener for them
There might be a better way to accomplish this. I can't debug this right now and I can't manage to get your examples working. Hope it helps.

SqlAlchemy Table and Query Issues

Still wrapping my head around SqlAlchemy and have run into a few issues. Not sure if it is because I am creating the relationships incorrectly, querying incorrect, or both.
The general idea is...
one-to-many from location to user (a location can have many users but users can only have one location).
many-to-many between group and user (a user can be a member of many groups and a group can have many members).
Same as #2 above for desc and user.
My tables are created as follows:
Base = declarative_base()
class Location(Base):
__tablename__ = 'location'
id = Column(Integer, primary_key=True)
name = Column(String)
group_user_association_table = Table('group_user_association_table', Base.metadata,
Column('group_id', Integer, ForeignKey('group.id')),
Column('user_id', Integer, ForeignKey('user.id')))
class Group(Base):
__tablename__ = 'group'
id = Column(Integer, primary_key=True)
name = Column(String)
users = relationship('User', secondary=group_user_association_table, backref='group')
desc_user_association_table = Table('desc_user_association', Base.metadata,
Column('desc_id', Integer, ForeignKey('desc.id')),
Column('user_id', Integer, ForeignKey('user.id')))
class Desc(Base):
__tablename__ = 'desc'
id = Column(Integer, primary_key=True)
name = Column(String)
users = relationship('User', secondary=desc_user_association_table, backref='desc')
class User(Base):
__tablename__ = 'user'
id = Column(Integer, primary_key=True)
user_name = Column(String)
location_id = Column(Integer, ForeignKey('location.id'))
groups = Column(String, ForeignKey('group.id'))
descs = Column(String, ForeignKey('desc.id'))
location = relationship('Location', backref='user')
Here are some examples as to how I am creating the data (all being scraped from the web):
location = Location(id=city[1], name=city[0]) #city = ('name', id)
profile = User()
profile.id = int(str(span2class[0].a['href'][7:]))
profile.user_name = str(span2class[0].a.img['alt'])
profile.location_id = location.id
g = Group(id=gid, name=str(group.contents[0])) # add the group to the Group table
self.db_session.add(g)
# Now add the gid to a list that will be added to the profile that eventually gets added to the user table
profile.groups.append(str(gid)) # stick the gid into the list
profile.groups = ','.join(profile.groups) # convert list to csv string
# Repeat basically same thing above for desc
self.db_session.add(profile)
self.db_session.commit()
As far as queries go, I've got some of the basic ones working such as:
for instance in db_session.query(User).all():
print instance.id, instance.user_name
But when it comes to performing a join to get (for example) group.id and group.name for a specific user.id... nothing I've tried has worked. I am guessing that the form would be something like the following:
db_session.query(User, Group).join('users').filter(User.id==42)
but that didn't work.

Joins works from left to right, so you should join on the relationship from User to Group:
db_session.query(User, Group).join(User.group).filter(User.id == 42)
But this return you a list of tuples (<User>, <Group>), so if the user belongs to 2 or more groups, you will receive 2 or more rows.
If you really want to load both the user and its groups in one (SQL) query, a better way would be to load a user, but configure query to preload groups:
u = (session.query(User)
.options(joinedload(User.group))
.get(42)
)
print("User = {}".format(u))
for g in u.group:
print(" Group = {}".format(g))

SQLAlchemy how to update children when old = [1, 2] new = [2, 3]

I have two models:
class Person(Model):
id
name
skills = relationship(Skill)
class Skill(Model):
id
skill
person_id
At the beginning, for example:
jack = Person(name='jack')
jack.skills = [Skill(s) for s in ['python', 'ruby']]
jack.save()
Then, one day, jack lost his skill 'ruby' but earned 'swift'
so his skill is ['python', 'swift'].
My current way of doing this update is:
look for existing skills, i get old = ['python', 'ruby']
get the new list new = ['python', 'swift']
make old, new to set(old), set(new)
unchanged = old.intersection(new), so i get the skill that does not change
I add every skill in set(new - unchanged)
I delete every skill in set(old-unchanged)
Is there a easier way to do this?

Use collection_class=set on the relationship to treat it as a set instead of a list.
Here's a working example of how to relate people with skills. This is a many-to-many relationship, instead of each skill being related to one person_id, each skill can be related to many people through the person_skill table. The relationship collection is a set, and Skill has a __hash__ function to make skills with the same name hash the same.
from sqlalchemy import create_engine, Column, Integer, String, ForeignKey, Table
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import sessionmaker, relationship
engine = create_engine('sqlite:///:memory:', echo=True)
Base = declarative_base(bind=engine)
Session = sessionmaker(bind=engine)
session = Session()
class Person(Base):
__tablename__ = 'person'
id = Column(Integer, primary_key=True)
name = Column(String, nullable=False, unique=True)
# many-to-many relation, as a set
skills = relationship('Skill', 'person_skill', collection_class=set)
class Skill(Base):
__tablename__ = 'skill'
id = Column(Integer, primary_key=True)
name = Column(String, nullable=False, unique=True)
def __hash__(self):
# so that the set collection will handle duplicate entries
return hash((self.__class__, self.name))
# many-to-many table, relate a person to a skill
person_skill = Table(
'person_skill', Base.metadata,
Column('person_id', Integer, ForeignKey(Person.id), primary_key=True),
Column('skill_id', Integer, ForeignKey(Skill.id), primary_key=True)
)
# create the tables
Base.metadata.create_all()
# populate some skills and people
s1 = Skill(name='python')
s2 = Skill(name='sqlalchemy')
s3 = Skill(name='questions')
s4 = Skill(name='ruby')
p1 = Person(name='davidism', skills={s1, s2, s4})
p2 = Person(name='user2653947', skills={s3})
session.add_all([p1, p2])
session.commit()
# change some skills on people
p1.skills.discard(s4)
p2.skills.add(s2)
session.commit()
This is not a complete solution. You could for instance plug in the unique object pattern demonstrated in this answer to make sure the skills you create are never duplicated.