SQLAlchemy filters

Filter, sort and paginate SQLAlchemy query objects. Ideal for exposing these actions over a REST API.

Filtering

Assuming that we have a SQLAlchemy query object:

from sqlalchemy import Column, Integer, String
from sqlalchemy.ext.declarative import declarative_base


class Base(object):
    id = Column(Integer, primary_key=True)
    name = Column(String(50), nullable=False)
    count = Column(Integer, nullable=True)

    @hybrid_property
    def count_square(self):
        return self.count * self.count

    @hybrid_method
    def three_times_count(self):
        return self.count * 3


Base = declarative_base(cls=Base)


class Foo(Base):

    __tablename__ = 'foo'

# ...

query = session.query(Foo)

Then we can apply filters to that query object (multiple times):

from sqlalchemy_filters import apply_filters


# `query` should be a SQLAlchemy query object

filter_spec = [{'field': 'name', 'op': '==', 'value': 'name_1'}]
filtered_query = apply_filters(query, filter_spec)

more_filters = [{'field': 'foo_id', 'op': 'is_not_null'}]
filtered_query = apply_filters(filtered_query, more_filters)

result = filtered_query.all()

It is also possible to filter queries that contain multiple models, including joins:

class Bar(Base):

    __tablename__ = 'bar'

    foo_id = Column(Integer, ForeignKey('foo.id'))

query = session.query(Foo).join(Bar)

filter_spec = [
    {'model': 'Foo', 'field': 'name', 'op': '==', 'value': 'name_1'},
    {'model': 'Bar', 'field': 'count', 'op': '>=', 'value': 5},
]
filtered_query = apply_filters(query, filter_spec)

result = filtered_query.all()

apply_filters will attempt to automatically join models to query if they're not already present and a model-specific filter is supplied. For example, the value of filtered_query in the following two code blocks is identical:

query = session.query(Foo).join(Bar)  # join pre-applied to query

filter_spec = [
    {'model': 'Foo', 'field': 'name', 'op': '==', 'value': 'name_1'},
    {'model': 'Bar', 'field': 'count', 'op': '>=', 'value': 5},
]
filtered_query = apply_filters(query, filter_spec)

query = session.query(Foo)  # join to Bar will be automatically applied

filter_spec = [
    {'field': 'name', 'op': '==', 'value': 'name_1'},
    {'model': 'Bar', 'field': 'count', 'op': '>=', 'value': 5},
]
filtered_query = apply_filters(query, filter_spec)

The automatic join is only possible if SQLAlchemy can implictly determine the condition for the join, for example because of a foreign key relationship.

Automatic joins allow flexibility for clients to filter and sort by related objects without specifying all possible joins on the server beforehand. Feature can be explicitly disabled by passing do_auto_join=False argument to the apply_filters call.

Note that first filter of the second block does not specify a model. It is implictly applied to the Foo model because that is the only model in the original query passed to apply_filters.

It is also possible to apply filters to queries defined by fields, functions or select_from clause:

query_alt_1 = session.query(Foo.id, Foo.name)
query_alt_2 = session.query(func.count(Foo.id))
query_alt_3 = session.query().select_from(Foo).add_column(Foo.id)

Hybrid attributes

You can filter by a hybrid attribute: a hybrid property or a hybrid method.

query = session.query(Foo)

filter_spec = [{'field': 'count_square', 'op': '>=', 'value': 25}]
filter_spec = [{'field': 'three_times_count', 'op': '>=', 'value': 15}]

filtered_query = apply_filters(query, filter_spec)
result = filtered_query.all()

Restricted Loads

You can restrict the fields that SQLAlchemy loads from the database by using the apply_loads function:

query = session.query(Foo, Bar).join(Bar)
load_spec = [
    {'model': 'Foo', 'fields': ['name']},
    {'model': 'Bar', 'fields': ['count']}
]
query = apply_loads(query, load_spec)  # will load only Foo.name and Bar.count

The effect of the apply_loads function is to _defer_ the load of any other fields to when/if they're accessed, rather than loading them when the query is executed. It only applies to fields that would be loaded during normal query execution.

Effect on joined queries

The default SQLAlchemy join is lazy, meaning that columns from the joined table are loaded only when required. Therefore apply_loads has limited effect in the following scenario:

query = session.query(Foo).join(Bar)
load_spec = [
    {'model': 'Foo', 'fields': ['name']}
    {'model': 'Bar', 'fields': ['count']}  # ignored
]
query = apply_loads(query, load_spec)  # will load only Foo.name

apply_loads cannot be applied to columns that are loaded as joined eager loads. This is because a joined eager load does not add the joined model to the original query, as explained here

The following would not prevent all columns from Bar being eagerly loaded:

query = session.query(Foo).options(joinedload(Foo.bar))
load_spec = [
    {'model': 'Foo', 'fields': ['name']}
    {'model': 'Bar', 'fields': ['count']}
]
query = apply_loads(query, load_spec)

Automatic Join

In fact, what happens here is that Bar is automatically joined to query, because it is determined that Bar is not part of the original query. The load_spec therefore has no effect because the automatic join results in lazy evaluation.

If you wish to perform a joined load with restricted columns, you must specify the columns as part of the joined load, rather than with apply_loads:

query = session.query(Foo).options(joinedload(Bar).load_only('count'))
load_spec = [
    {'model': 'Foo', 'fields': ['name']}
]
query = apply_loads(query. load_spec)  # will load ony Foo.name and Bar.count

Sort

from sqlalchemy_filters import apply_sort


# `query` should be a SQLAlchemy query object

sort_spec = [
    {'model': 'Foo', 'field': 'name', 'direction': 'asc'},
    {'model': 'Bar', 'field': 'id', 'direction': 'desc'},
]
sorted_query = apply_sort(query, sort_spec)

result = sorted_query.all()

apply_sort will attempt to automatically join models to query if they're not already present and a model-specific sort is supplied. The behaviour is the same as in apply_filters.

This allows flexibility for clients to sort by fields on related objects without specifying all possible joins on the server beforehand.

Hybrid attributes

You can sort by a hybrid attribute: a hybrid property or a hybrid method.

Pagination

from sqlalchemy_filters import apply_pagination


# `query` should be a SQLAlchemy query object

query, pagination = apply_pagination(query, page_number=1, page_size=10)

page_size, page_number, num_pages, total_results = pagination

assert 10 == len(query)
assert 10 == page_size == pagination.page_size
assert 1 == page_number == pagination.page_number
assert 3 == num_pages == pagination.num_pages
assert 22 == total_results == pagination.total_results

Filters format

Filters must be provided in a list and will be applied sequentially. Each filter will be a dictionary element in that list, using the following format:

filter_spec = [
    {'model': 'model_name', 'field': 'field_name', 'op': '==', 'value': 'field_value'},
    {'model': 'model_name', 'field': 'field_2_name', 'op': '!=', 'value': 'field_2_value'},
    # ...
]

The model key is optional if the original query being filtered only applies to one model.

If there is only one filter, the containing list may be omitted:

filter_spec = {'field': 'field_name', 'op': '==', 'value': 'field_value'}

Where field is the name of the field that will be filtered using the operator provided in op (optional, defaults to ==) and the provided value (optional, depending on the operator).

This is the list of operators that can be used:

• is_null
• is_not_null
• ==, eq
• !=, ne
• >, gt
• <, lt
• >=, ge
• <=, le
• like
• ilike
• not_ilike
• in
• not_in
• any
• not_any

any / not_any

PostgreSQL specific operators allow to filter queries on columns of type ARRAY. Use any to filter if a value is present in an array and not_any if it's not.

Boolean Functions

and, or, and not functions can be used and nested within the filter specification:

filter_spec = [
    {
        'or': [
            {
                'and': [
                    {'field': 'field_name', 'op': '==', 'value': 'field_value'},
                    {'field': 'field_2_name', 'op': '!=', 'value': 'field_2_value'},
                ]
            },
            {
                'not': [
                    {'field': 'field_3_name', 'op': '==', 'value': 'field_3_value'}
                ]
            },
        ],
    }
]

Note: or and and must reference a list of at least one element. not must reference a list of exactly one element.

Sort format

Sort elements must be provided as dictionaries in a list and will be applied sequentially:

sort_spec = [
    {'model': 'Foo', 'field': 'name', 'direction': 'asc'},
    {'model': 'Bar', 'field': 'id', 'direction': 'desc'},
    # ...
]

Where field is the name of the field that will be sorted using the provided direction.

The model key is optional if the original query being sorted only applies to one model.

nullsfirst / nullslast

sort_spec = [
    {'model': 'Baz', 'field': 'count', 'direction': 'asc', 'nullsfirst': True},
    {'model': 'Qux', 'field': 'city', 'direction': 'desc', 'nullslast': True},
    # ...
]

nullsfirst is an optional attribute that will place NULL values first if set to True, according to the SQLAlchemy documentation.

nullslast is an optional attribute that will place NULL values last if set to True, according to the SQLAlchemy documentation.

If none of them are provided, then NULL values will be sorted according to the RDBMS being used. SQL defines that NULL values should be placed together when sorting, but it does not specify whether they should be placed first or last.

Even though both nullsfirst and nullslast are part of SQLAlchemy, they will raise an unexpected exception if the RDBMS that is being used does not support them.

At the moment they are supported by PostgreSQL, but they are not supported by SQLite and MySQL.

Running tests

The default configuration uses SQLite, MySQL (if the driver is installed, which is the case when tox is used) and PostgreSQL (if the driver is installed, which is the case when tox is used) to run the tests, with the following URIs:

sqlite+pysqlite:///test_sqlalchemy_filters.db
mysql+mysqlconnector://root:@localhost:3306/test_sqlalchemy_filters
postgresql+psycopg2://postgres:@localhost:5432/test_sqlalchemy_filters?client_encoding=utf8'

A test database will be created, used during the tests and destroyed afterwards for each RDBMS configured.

There are Makefile targets to run docker containers locally for both MySQL and PostgreSQL, using the default ports and configuration:

$ make mysql-container
$ make postgres-container

To run the tests locally:

$ # Create/activate a virtual environment
$ pip install tox
$ tox

There are some other Makefile targets that can be used to run the tests:

There are other Makefile targets to run the tests, but extra dependencies will have to be installed:

$ pip install -U --editable ".[dev,mysql,postgresql]"
$ # using default settings
$ make test
$ make coverage

$ # overriding DB parameters
$ ARGS='--mysql-test-db-uri mysql+mysqlconnector://root:@192.168.99.100:3340/test_sqlalchemy_filters' make test
$ ARGS='--sqlite-test-db-uri sqlite+pysqlite:///test_sqlalchemy_filters.db' make test

$ ARGS='--mysql-test-db-uri mysql+mysqlconnector://root:@192.168.99.100:3340/test_sqlalchemy_filters' make coverage
$ ARGS='--sqlite-test-db-uri sqlite+pysqlite:///test_sqlalchemy_filters.db' make coverage

Database management systems

The following RDBMS are supported (tested):

• SQLite
• MySQL
• PostgreSQL

SQLAlchemy support

The following SQLAlchemy versions are supported: 1.0, 1.1, 1.2, 1.3, 1.4.

Changelog

Consult the CHANGELOG document for fixes and enhancements of each version.

License

Apache 2.0. See LICENSE for details.