Question

I am using Python 2.6.6 and SQLAlchemy 0.6.6 to handle a one to many relationship in my database, and am unsure how to prevent SQLAlchemy from adding new child records in the case where similar data already exists.

Database code:

from sqlalchemy import *
from sqlalchemy.orm import backref, relationship, sessionmaker, create_session
from sqlalchemy.ext.declarative import declarative_base

Base = declarative_base()

# Parent table, no foreign key.
class Author(Base):
    __tablename__ = 'authors'

    id = Column(Integer, primary_key=True)
    username = Column(String)
    author_metadata = relationship('AuthorMetadata', backref='author')

# Child table, many records with same author_id.
class AuthorMetadata(Base):
    __tablename__ = 'author_metadata'

    id = Column(Integer, primary_key=True)
    author_id = Column(Integer, ForeignKey('authors.id'))
    metakey = Column(String)
    metavalue = Column(Text)

Example script:

if __name__ == '__main__':
    engine = create_engine('database_details', pool_recycle=90)
    session = create_session(bind=engine)

    author = session.query(Author).filter_by(username='Godfrey').first()
    if not author:
        author = Author()
    author.username = 'Godfrey'
    author.author_metadata = [
        AuthorMetadata(metakey='location', metavalue='New York'),
        AuthorMetadata(metakey='posts', metavalue='5')]
    session.add(author)
    session.flush()

The first time I run the example script, the following appears in the database (as expected):

dev=# select id from authors where username = 'Godfrey';
  id  
------
 5025
(1 row)

dev=# select id, author_id, metakey, metavalue from author_metadata order by id desc limit 2;
  id   | author_id | metakey  | metavalue 
-------+-----------+----------+-----------
 85090 |      5025 | posts    | 5
 85089 |      5025 | location | New York
(2 rows)

If I run the example script again though, you can see that the existing metadata record’s author ids have been set to null and new records have been inserted:

dev=# select id, author_id, metakey, metavalue from author_metadata order by id desc limit 4;
  id   | author_id | metakey  | metavalue 
-------+-----------+----------+-----------
 85092 |      5025 | posts    | 5
 85091 |      5025 | location | New York
 85090 |           | posts    | 5
 85089 |           | location | New York
(4 rows)

I don’t find this surprising, but am wondering if there is a nice way to be able to communicate to SQLAlchemy that it should only insert/modify/delete author metadata rows if the new list of metadata differs from the existing list.

Answer 1

You could explicitely check the contents of the list and only append new AuthorMetadata objects if they don’t exist, rather than delete the entire collection and re-create it with brand new objects. That would at least avoid discarding the previously created records.

Your use case matches attribute_mapped_collection and association_proxy quite well, so you probably want to go with one of them.