Files
deb-python-sqlalchemy-utils/sqlalchemy_utils/aggregates.py
2014-01-23 10:07:35 +02:00

582 lines
16 KiB
Python

"""
SQLAlchemy-Utils provides way of automatically calculating aggregate values of
related models and saving them to parent model.
This solution is inspired by RoR counter cache,
`counter_culture`_ and `stackoverflow reply by Michael Bayer`_.
Why?
----
Many times you may have situations where you need to calculate dynamically some
aggregate value for given model. Some simple examples include:
- Number of products in a catalog
- Average rating for movie
- Latest forum post
- Total price of orders for given customer
Now all these aggregates can be elegantly implemented with SQLAlchemy
column_property_ function. However when your data grows calculating these
values on the fly might start to hurt the performance of your application. The
more aggregates you are using the more performance penalty you get.
This module provides way of calculating these values automatically and
efficiently at the time of modification rather than on the fly.
Features
--------
* Automatically updates aggregate columns when aggregated values change
* Supports aggregate values through arbitrary number levels of relations
* Highly optimized: uses single query per transaction per aggregate column
* Aggregated columns can be of any data type and use any selectable scalar expression
.. _column_property: http://docs.sqlalchemy.org/en/latest/orm/mapper_config.html#using-column-property
.. _counter_culture: https://github.com/magnusvk/counter_culture
.. _stackoverflow reply by Michael Bayer:
http://stackoverflow.com/questions/13693872/
Simple aggregates
-----------------
::
from sqlalchemy_utils import aggregated
class Thread(Base):
__tablename__ = 'thread'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('comments', sa.Column(sa.Integer))
def comment_count(self):
return sa.func.count('1')
comments = sa.orm.relationship(
'Comment',
backref='thread'
)
class Comment(Base):
__tablename__ = 'comment'
id = sa.Column(sa.Integer, primary_key=True)
content = sa.Column(sa.UnicodeText)
thread_id = sa.Column(sa.Integer, sa.ForeignKey(Thread.id))
thread = Thread(name=u'SQLAlchemy development')
thread.comments.append(Comment(u'Going good!'))
thread.comments.append(Comment(u'Great new features!'))
session.add(thread)
session.commit()
thread.comment_count # 2
Custom aggregate expressions
----------------------------
Aggregate expression can be virtually any SQL expression not just a simple function taking one parameter. You can try things such as subqueries and different kinds of functions.
In the following example we have a Catalog of products where each catalog knows the net worth of its products.
::
from sqlalchemy_utils import aggregated
class Catalog(Base):
__tablename__ = 'catalog'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('products', sa.Column(sa.Integer))
def net_worth(self):
return sa.func.sum(Product.price)
products = sa.orm.relationship('Product')
class Product(Base):
__tablename__ = 'product'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
price = sa.Column(sa.Numeric)
catalog_id = sa.Column(sa.Integer, sa.ForeignKey(Catalog.id))
Now the net_worth column of Catalog model will be automatically whenever:
* A new product is added to the catalog
* A product is deleted from the catalog
* The price of catalog product is changed
::
from decimal import Decimal
product1 = Product(name='Some product', price=Decimal(1000))
product2 = Product(name='Some other product', price=Decimal(500))
catalog = Catalog(
name=u'My first catalog',
products=[
product1,
product2
]
)
session.add(catalog)
session.commit()
session.refresh(catalog)
catalog.net_worth # 1500
session.delete(product2)
session.commit()
session.refresh(catalog)
catalog.net_worth # 1000
product1.price = 2000
session.commit()
session.refresh(catalog)
catalog.net_worth # 2000
Multiple aggregates per class
-----------------------------
Sometimes you may need to define multiple aggregate values for same class. If you need
to define lots of relationships pointing to same class, remember to define the relationships as viewonly when possible.
::
from sqlalchemy_utils import aggregated
class Customer(Base):
__tablename__ = 'customer'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('orders', sa.Column(sa.Integer))
def orders_sum(self):
return sa.func.sum(Order.price)
@aggregated('invoiced_orders', sa.Column(sa.Integer))
def invoiced_orders_sum(self):
return sa.func.sum(Order.price)
orders = sa.orm.relationship('Order')
invoiced_orders = sa.orm.relationship(
'Order',
primaryjoin=
'sa.and_(Order.customer_id == Customer.id, Order.invoiced)',
viewonly=True
)
class Order(Base):
__tablename__ = 'order'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
price = sa.Column(sa.Numeric)
invoiced = sa.Column(sa.Boolean, default=False)
customer_id = sa.Column(sa.Integer, sa.ForeignKey(Customer.id))
Many-to-Many aggregates
-----------------------
Aggregate expressions also support many-to-many relationships. The usual use scenarios includes things such as:
1. Friend count of a user
2. Group count where given user belongs to
::
user_group = sa.Table('user_group', Base.metadata,
sa.Column('user_id', sa.Integer, sa.ForeignKey('user.id')),
sa.Column('group_id', sa.Integer, sa.ForeignKey('group.id'))
)
class User(Base):
__tablename__ = 'user'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('groups', sa.Column(sa.Integer, default=0))
def group_count(self):
return sa.func.count('1')
groups = sa.orm.relationship(
'Group',
backref='users',
secondary=user_group
)
class Group(Base):
__tablename__ = 'group'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
user = User(name=u'John Matrix')
user.groups = [Group(name=u'Group A'), Group(name=u'Group B')]
session.add(user)
session.commit()
session.refresh(user)
user.group_count # 2
Multi-level aggregates
----------------------
Aggregates can span accross multiple relationships. In the following example each
Catalog has a net_worth which is the sum of all products in all categories.
::
from sqlalchemy_utils import aggregated
class Catalog(Base):
__tablename__ = 'catalog'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('categories.products', sa.Column(sa.Integer))
def net_worth(self):
return sa.func.sum(Product.price)
categories = sa.orm.relationship('Product')
class Category(Base):
__tablename__ = 'category'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
catalog_id = sa.Column(sa.Integer, sa.ForeignKey(Catalog.id))
products = sa.orm.relationship('Product')
class Product(Base):
__tablename__ = 'product'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
price = sa.Column(sa.Numeric)
category_id = sa.Column(sa.Integer, sa.ForeignKey(Category.id))
Examples
--------
Average movie rating
^^^^^^^^^^^^^^^^^^^^
::
from sqlalchemy_utils import aggregated
class Movie(Base):
__tablename__ = 'movie'
id = sa.Column(sa.Integer, primary_key=True)
name = sa.Column(sa.Unicode(255))
@aggregated('ratings', sa.Column(sa.Numeric))
def avg_rating(self):
return sa.func.sum(Product.price)
ratings = sa.orm.relationship('Rating')
class Rating(Base):
__tablename__ = 'rating'
id = sa.Column(sa.Integer, primary_key=True)
stars = sa.Column(sa.Integer)
movie_id = sa.Column(sa.Integer, sa.ForeignKey(Movie.id))
movie = Movie('Terminator 2')
movie.ratings.append(Rating(stars=5))
movie.ratings.append(Rating(stars=4))
movie.ratings.append(Rating(stars=3))
session.add(movie)
session.commit()
movie.avg_rating # 4
TODO
----
* Special consideration should be given to `deadlocks`_.
.. _deadlocks:
http://mina.naguib.ca/blog/2010/11/22/postgresql-foreign-key-deadlocks.html
"""
from collections import defaultdict
from weakref import WeakKeyDictionary
import sqlalchemy as sa
import six
from sqlalchemy.ext.declarative import declared_attr
try:
# SQLAlchemy 0.9
from sqlalchemy.sql.functions import _FunctionGenerator
except ImportError:
# SQLAlchemy 0.8
from sqlalchemy.sql.expression import _FunctionGenerator
aggregated_attrs = WeakKeyDictionary(defaultdict(list))
class AggregatedAttribute(declared_attr):
def __init__(
self,
fget,
relationship,
column,
*args,
**kwargs
):
super(AggregatedAttribute, self).__init__(fget, *args, **kwargs)
self.__doc__ = fget.__doc__
self.column = column
self.relationship = relationship
def __get__(desc, self, cls):
if cls not in aggregated_attrs:
aggregated_attrs[cls] = [(desc.fget, desc.relationship)]
else:
aggregated_attrs[cls].append((desc.fget, desc.relationship))
return desc.column
class AggregatedValue(object):
def __init__(self, class_, attr, relationships, expr):
self.class_ = class_
self.attr = attr
self.relationships = relationships
self.expr = self.aggregate_expression(expr, class_)
def aggregate_expression(self, expr, class_):
if isinstance(expr, sa.sql.visitors.Visitable):
return expr
elif isinstance(expr, _FunctionGenerator):
return expr(sa.sql.text('1'))
else:
return expr(class_)
@property
def aggregate_query(self):
from_ = self.relationships[0].mapper.class_.__table__
for relationship in self.relationships[0:-1]:
property_ = relationship.property
from_ = (
from_
.join(
property_.parent.class_,
property_.primaryjoin
)
)
query = sa.select(
[self.expr],
from_obj=[from_]
)
query = query.where(self.relationships[-1])
return query.correlate(self.class_).as_scalar()
def update_query(self, objects):
table = self.class_.__table__
query = table.update().values(
{self.attr: self.aggregate_query}
)
if len(self.relationships) == 1:
prop = self.relationships[-1].property
return query.where(self.local_condition(prop, objects))
else:
# Builds query such as:
#
# UPDATE catalog SET product_count = (aggregate_query)
# WHERE id IN (
# SELECT catalog_id
# FROM category
# INNER JOIN sub_category
# ON category.id = sub_category.category_id
# WHERE sub_category.id IN (product_sub_category_ids)
# )
property_ = self.relationships[-1].property
remote_pairs = property_.local_remote_pairs
local = remote_pairs[0][0]
remote = remote_pairs[0][1]
return query.where(
local.in_(
sa.select(
[remote],
from_obj=[self.multi_level_aggregate_query_base]
).where(
self.local_condition(
self.relationships[0].property,
objects
)
)
)
)
@property
def multi_level_aggregate_query_base(self):
property_ = self.relationships[-1].property
from_ = property_.mapper.class_.__table__
for relationship in reversed(self.relationships[1:-1]):
property_ = relationship.property
from_ = (
from_.join(property_.mapper.class_, property_.primaryjoin)
)
return from_
def local_condition(self, prop, objects):
if prop.secondary is not None:
return prop.local_remote_pairs[1][0].in_(
getattr(
obj,
prop.local_remote_pairs[1][0].key
)
for obj in objects
)
else:
return prop.local_remote_pairs[0][0].in_(
getattr(
obj,
prop.local_remote_pairs[0][1].key
)
for obj in objects
)
class AggregationManager(object):
def __init__(self):
self.reset()
def reset(self):
self.generator_registry = defaultdict(list)
def register_listeners(self):
sa.event.listen(
sa.orm.mapper,
'after_configured',
self.update_generator_registry
)
sa.event.listen(
sa.orm.session.Session,
'after_flush',
self.construct_aggregate_queries
)
def update_generator_registry(self):
for class_, attrs in six.iteritems(aggregated_attrs):
for expr, relationship in attrs:
relationships = []
rel_class = class_
for path_name in relationship.split('.'):
rel = getattr(rel_class, path_name)
relationships.append(rel)
rel_class = rel.mapper.class_
self.generator_registry[rel_class].append(
AggregatedValue(
class_=class_,
attr=expr.__name__,
relationships=list(reversed(relationships)),
expr=expr(class_)
)
)
def construct_aggregate_queries(self, session, ctx):
object_dict = defaultdict(list)
for obj in session:
class_ = obj.__class__
if class_ in self.generator_registry:
object_dict[class_].append(obj)
for class_, objects in six.iteritems(object_dict):
for aggregate_value in self.generator_registry[class_]:
query = aggregate_value.update_query(objects)
session.execute(query)
manager = AggregationManager()
manager.register_listeners()
def aggregated(
relationship,
column
):
"""
Decorator that generates an aggregated attribute. The decorated function
should return an aggregate select expression.
:param relationship:
Defines the relationship of which the aggregate is calculated from.
The class needs to have given relationship in order to calculate the
aggregate.
:param column:
SQLAlchemy Column object. The column definition of this aggregate
attribute.
"""
def wraps(func):
return AggregatedAttribute(
func,
relationship,
column
)
return wraps