A SQLAlchemy like ORM implementation for arangodb
Project description
Welcome to Arango ORM (Python ORM Layer For ArangoDB)
arango_orm is a python ORM layer inspired by SQLAlchemy but aimed to work with the multi-model database ArangoDB. It supports accessing both collections and graphs using the ORM. The actual communication with the database is done using python-arango (the database driver for accessing arangodb from python) and object serialization, validation, etc is handled by pydantic.
Installation
pip install arango-orm
Connecting to a Database
from arango import ArangoClient
from arango_orm import Database
client = ArangoClient(hosts='http://localhost:8529')
test_db = client.db('test', username='test', password='test')
db = Database(test_db)
Define a Collection
from datetime import date
from pydantic import Field
from arango_orm import Collection
class Student(Collection):
__collection__ = 'students'
name: str
dob: date
Create Collection in the Database
db.create_collection(Student)
Drop a Collection
db.drop_collection(Student)
Check if a collection exists
db.has_collection(Student)
db.has_collection('students')
Add Records
from datetime import date
s = Student(name='test', _key='12312', dob=date(year=2016, month=9, day=12))
db.add(s)
print(s._id) # students/12312
Get Total Records in the Collection
db.query(Student).count()
Get Record By Key
s = db.query(Student).by_key('12312')
Update a Record
s = db.query(Student).by_key('12312')
s.name = 'Anonymous'
db.update(s)
Delete a Record
s = db.query(Student).by_key('12312')
db.delete(s)
Get All Records in a Collection
students = db.query(Student).all()
Get First Record Matching the Query
first_student = db.query(Student).first()
Filter Records
Using bind parameters (recommended)
records = db.query(Student).filter("name==@name", name='Anonymous').all()
Using plain condition strings (not safe in case of unsanitized user supplied input)
records = db.query(Student).filter("name=='Anonymous'").all()
Filter Using OR
# Get all documents where student name starts with A or B
records = db.query(Student).filter(
"LIKE(rec.name, 'A%')", prepend_rec_name=False).filter(
"LIKE(rec.name, 'B%')", prepend_rec_name=False, _or=True).all()
Filter, Sort and Limit
# Last 5 students with names starting with A
records = db.query(Student).filter(
"LIKE(rec.name, 'A%')", prepend_rec_name=False).sort("name DESC").limit(5).all()
# Query students with pagination (limit&offset)
page_num, per_page = 2, 10
page = db.query(Student).sort("name DESC").limit(per_page, start_from=(page_num - 1) * per_page)
Fetch Only Some Fields
c = db.query(Student).limit(2).returns('_key', 'name').first()
Update Multiple Records
db.query(Student).filter("name==@name", name='Anonymous').update(name='Mr. Anonymous')
Delete Multiple Records
db.query(Student).filter("LIKE(rec.name, 'test%')", prepend_rec_name=False).delete()
Delete All Records
db.query(Student).delete()
Bulk Create Records
s1 = Student(name='test1', _key='12345', dob=date(year=2016, month=9, day=12))
s2 = Student(name='test2', _key='22346', dob=date(year=2015, month=9, day=12)
car1 = Car(make="Honda", model="Fiat", year=2010)
car2 = Car(make="Honda", model="Skoda", year=2015)
db.bulk_add(entity_list=[p_ref_10, p_ref_11, car1, car2])
Bulk Update Records
p_ref1 = db.query(Person).by_key("12312")
p_ref2 = db.query(Person).by_key("12345")
p_ref1.name = "Bruce"
p_ref2.name = "Eliza"
db.bulk_update(entity_list=[p_ref1, p_ref2])
Query Using AQL
db.add(Student(name='test1', _key='12345', dob=date(year=2016, month=9, day=12)))
db.add(Student(name='test2', _key='22346', dob=date(year=2015, month=9, day=12)))
students = [Student._load(s) for s in db.aql.execute("FOR st IN students RETURN st")]
Reference Fields
Reference fields allow linking documents from another collection class within a collection instance. These are similar in functionality to SQLAlchemy's relationship function.
from arango import ArangoClient
from arango_orm.database import Database
from arango_orm.fields import String
from arango_orm import Collection, Relation, Graph, GraphConnection
from arango_orm.references import relationship, graph_relationship
class Person(Collection):
__collection__ = 'persons'
_index = [{'type': 'hash', 'unique': False, 'fields': ['name']}]
_allow_extra_fields = False # prevent extra properties from saving into DB
_key = String(required=True)
name = String(required=True, allow_none=False)
cars = relationship(__name__ + ".Car", '_key', target_field='owner_key')
def __str__(self):
return "<Person(" + self.name + ")>"
class Car(Collection):
__collection__ = 'cars'
_allow_extra_fields = True
make = String(required=True)
model = String(required=True)
year = Integer(required=True)
owner_key = String()
owner = relationship(Person, 'owner_key', cache=False)
def __str__(self):
return "<Car({} - {} - {})>".format(self.make, self.model, self.year)
client = ArangoClient(hosts='http://localhost:8529')
test_db = client.db('test', username='test', password='test')
db = Database(test_db)
p = Person(_key='kashif', name='Kashif Iftikhar')
db.add(p)
p2 = Person(_key='azeen', name='Azeen Kashif')
db.add(p2)
c1 = Car(make='Honda', model='Civic', year=1984, owner_key='kashif')
db.add(c1)
c2 = Car(make='Mitsubishi', model='Lancer', year=2005, owner_key='kashif')
db.add(c2)
c3 = Car(make='Acme', model='Toy Racer', year=2016, owner_key='azeen')
db.add(c3)
print(c1.owner)
print(c1.owner.name)
print(c2.owner.name)
print(c3.owner.name)
print(p.cars)
print(p.cars[0].make)
print(p2.cars)
Working With Graphs
Working with graphs involves creating collection classes and optionally Edge/Relation classes. Users can use the built-in Relation class for specifying relations but if relations need to contain extra attributes then it's required to create a sub-class of Relation class. Graph functionality is explain below with the help of a university graph example containing students, teachers, subjects and the areas where students and teachers reside in.
First we create some collections and relationships
from typing import Literal
from arango_orm import Collection, Relation, Graph, GraphConnection
class Student(Collection):
__collection__ = 'students'
name: str
age: int | None = None
def __str__(self):
return "<Student({})>".format(self.name)
class Teacher(Collection):
__collection__ = 'teachers'
name: str
def __str__(self):
return "<Teacher({})>".format(self.name)
class Subject(Collection):
__collection__ = 'subjects'
name: str
credit_hours: int
has_labs: bool = True
def __str__(self):
return "<Subject({})>".format(self.name)
class Area(Collection):
__collection__ = 'areas'
class SpecializesIn(Relation):
__collection__ = 'specializes_in'
expertise_level: Literal["expert", "medium", "basic"]
def __str__(self):
return "<SpecializesIn(_key={}, expertise_level={}, _from={}, _to={})>".format(
self.key_, self.expertise_level, self._from, self._to)
Next we sub-class the Graph class to specify the relationships between the various collections
class UniversityGraph(Graph):
__graph__ = 'university_graph'
graph_connections = [
# Using general Relation class for relationship
GraphConnection(Student, Relation("studies"), Subject),
GraphConnection(Teacher, Relation("teaches"), Subject),
# Using specific classes for vertex and edges
GraphConnection(Teacher, SpecializesIn, Subject),
GraphConnection([Teacher, Student], Relation("resides_in"), Area)
]
Now it's time to create the graph. Note that we don't need to create the collections individually, creating the graph will create all collections that it contains
from arango import ArangoClient
from arango_orm.database import Database
client = ArangoClient(hosts='http://localhost:8529')
test_db = client.db('test', username='test', password='test')
db = Database(test_db)
uni_graph = UniversityGraph(connection=db)
db.create_graph(uni_graph)
Now the graph and all it's collections have been created, we can verify their existence:
[c['name'] for c in db.collections()]
db.graphs()
Now let's insert some data into our graph:
students_data = [
Student(_key='S1001', name='John Wayne', age=30),
Student(_key='S1002', name='Lilly Parker', age=22),
Student(_key='S1003', name='Cassandra Nix', age=25),
Student(_key='S1004', name='Peter Parker', age=20)
]
teachers_data = [
Teacher(_key='T001', name='Bruce Wayne'),
Teacher(_key='T002', name='Barry Allen'),
Teacher(_key='T003', name='Amanda Waller')
]
subjects_data = [
Subject(_key='ITP101', name='Introduction to Programming', credit_hours=4, has_labs=True),
Subject(_key='CS102', name='Computer History', credit_hours=3, has_labs=False),
Subject(_key='CSOOP02', name='Object Oriented Programming', credit_hours=3, has_labs=True),
]
areas_data = [
Area(_key="Gotham"),
Area(_key="Metropolis"),
Area(_key="StarCity")
]
for s in students_data:
db.add(s)
for t in teachers_data:
db.add(t)
for s in subjects_data:
db.add(s)
for a in areas_data:
db.add(a)
Next let's add some relations, we can add relations by manually adding the relation/edge record into the edge collection, like:
db.add(SpecializesIn(_from="teachers/T001", _to="subjects/ITP101", expertise_level="medium"))
Or we can use the graph object's relation method to generate a relation document from given objects:
gotham = db.query(Area).by_key("Gotham")
metropolis = db.query(Area).by_key("Metropolis")
star_city = db.query(Area).by_key("StarCity")
john_wayne = db.query(Student).by_key("S1001")
lilly_parker = db.query(Student).by_key("S1002")
cassandra_nix = db.query(Student).by_key("S1003")
peter_parker = db.query(Student).by_key("S1004")
intro_to_prog = db.query(Subject).by_key("ITP101")
comp_history = db.query(Subject).by_key("CS102")
oop = db.query(Subject).by_key("CSOOP02")
barry_allen = db.query(Teacher).by_key("T002")
bruce_wayne = db.query(Teacher).by_key("T001")
amanda_waller = db.query(Teacher).by_key("T003")
db.add(uni_graph.relation(peter_parker, Relation("studies"), oop))
db.add(uni_graph.relation(peter_parker, Relation("studies"), intro_to_prog))
db.add(uni_graph.relation(john_wayne, Relation("studies"), oop))
db.add(uni_graph.relation(john_wayne, Relation("studies"), comp_history))
db.add(uni_graph.relation(lilly_parker, Relation("studies"), intro_to_prog))
db.add(uni_graph.relation(lilly_parker, Relation("studies"), comp_history))
db.add(uni_graph.relation(cassandra_nix, Relation("studies"), oop))
db.add(uni_graph.relation(cassandra_nix, Relation("studies"), intro_to_prog))
db.add(uni_graph.relation(barry_allen, SpecializesIn(expertise_level="expert"), oop))
db.add(uni_graph.relation(barry_allen, SpecializesIn(expertise_level="expert"), intro_to_prog))
db.add(uni_graph.relation(bruce_wayne, SpecializesIn(expertise_level="medium"), oop))
db.add(uni_graph.relation(bruce_wayne, SpecializesIn(expertise_level="expert"), comp_history))
db.add(uni_graph.relation(amanda_waller, SpecializesIn(expertise_level="basic"), intro_to_prog))
db.add(uni_graph.relation(amanda_waller, SpecializesIn(expertise_level="medium"), comp_history))
db.add(uni_graph.relation(bruce_wayne, Relation("teaches"), oop))
db.add(uni_graph.relation(barry_allen, Relation("teaches"), intro_to_prog))
db.add(uni_graph.relation(amanda_waller, Relation("teaches"), comp_history))
db.add(uni_graph.relation(bruce_wayne, Relation("resides_in"), gotham))
db.add(uni_graph.relation(barry_allen, Relation("resides_in"), star_city))
db.add(uni_graph.relation(amanda_waller, Relation("resides_in"), metropolis))
db.add(uni_graph.relation(john_wayne, Relation("resides_in"), gotham))
db.add(uni_graph.relation(lilly_parker, Relation("resides_in"), metropolis))
db.add(uni_graph.relation(cassandra_nix, Relation("resides_in"), star_city))
db.add(uni_graph.relation(peter_parker, Relation("resides_in"), metropolis))
With our graph populated with some sample data, let's explore the ways we can work with the graph.
Expanding Documents
We can expand any Collection (not Relation) object to access the data that is linked to it. We can sepcify which links ('inbound', 'outbound', 'any') to expand and the depth to which those should be expanded to. Let's see all immediate connections that Bruce Wayne has in our graph:
bruce = db.query(Teacher).by_key("T001")
uni_graph.expand(bruce, depth=1, direction='any')
Graph expansion on an object adds a _relations
dictionary that contains all the relations for the object according to the expansion criteria:
bruce._relations
# Returns:
{
'resides_in': [<Relation(_key=4205290, _from=teachers/T001, _to=areas/Gotham)>],
'specializes_in': [<SpecializesIn(_key=4205114, expertise_level=medium, _from=teachers/T001, _to=subjects/ITP101)>,
<SpecializesIn(_key=4205271, expertise_level=expert, _from=teachers/T001, _to=subjects/CS102)>,
<SpecializesIn(_key=4205268, expertise_level=medium, _from=teachers/T001, _to=subjects/CSOOP02)>],
'teaches': [<Relation(_key=4205280, _from=teachers/T001, _to=subjects/CSOOP02)>]
}
We can use _from and _to of a relation object to access the id's for both sides of the link. We also have _object_from and _object_to to access the objects on both sides, for example:
bruce._relations['resides_in'][0]._object_from.name
# 'Bruce Wayne'
bruce._relations['resides_in'][0]._object_to._key
# 'Gotham'
There is also a special attribute called _next
that allows accessing the other side of the relationship irrespective of the relationship direction. For example, for outbound relationships the _object_from
contains the source object while for inbound_relationships _object_to
contains the source object. But if we're only interested in traversal of the graph then it's more useful at times to access the other side of the relationship w.r.t the current object irrespective of it's direction:
bruce._relations['resides_in'][0]._next._key
# 'Gotham'
Let's expand the bruce object to 2 levels and see _next
in more action:
uni_graph.expand(bruce, depth=2)
# All relations of the area where bruce resides in
bruce._relations['resides_in'][0]._object_to._relations
# -> {'resides_in': [<Relation(_key=4205300, _from=students/S1001, _to=areas/Gotham)>]}
# Name of the student that resides in the same area as bruce
bruce._relations['resides_in'][0]._object_to._relations['resides_in'][0]._object_from.name
# 'John Wayne'
# The same action using _next without worrying about direction
bruce._relations['resides_in'][0]._next._relations['resides_in'][0]._next.name
# 'John Wayne'
# Get names of all people that reside in the same area and Bruce Wayne
[p._next.name for p in bruce._relations['resides_in'][0]._next._relations['resides_in']]
# ['John Wayne']
Inheritance Mapping
For inheritance mapping, arango_orm offers you two ways to define it.
1. Discriminator field/mapping:
Discriminator field/mapping are defined at entity level:
class Vehicle(Collection):
__collection__ = "vehicle"
_inheritance_field = "discr"
_inheritance_mapping = {
'Bike': 'moto',
'Truck': 'truck'
}
_key = String()
brand = String()
model = String()
# discr field match what you defined in _inheritance_field
# the field type depends on the values of your _inheritance_mapping
discr = String(required=True)
class Bike(Vehicle):
motor_size = Float()
class Truck(Vehicle):
traction_power = Float()
2. Inheritance mapping resolver:
The inheritance_mapping_resolver
is a function defined at graph level; it allows you to make either a simple test
on a discriminator field, or complex inference
class OwnershipGraph2(Graph):
__graph__ = "ownership_graph"
graph_connections = [
GraphConnection(Owner2, Own2, Vehicle2)
]
def inheritance_mapping_resolver(self, col_name: str, doc_dict: dict = {}):
if col_name == 'vehicle':
if 'traction_power' in doc_dict:
return Truck2
else:
return Bike2
return self.vertices[col_name]
Graph Traversal Using AQL
The graph module also supports traversals using AQL, the results are converted to objects and have the same structure as graph.expand method:
obj = uni_graph.aql("FOR v, e, p IN 1..2 INBOUND 'areas/Gotham' GRAPH 'university_graph' RETURN p")
print(obj._key)
# Gotham
gotham_residents = [rel._next.name for rel in obj._relations['resides_in']]
print(gotham_residents)
# ['Bruce Wayne', 'John Wayne']
For Developers
Running the Test Cases
Set env variables in .env file, then load the env
set -a; source .env; set +a
Afterwards run tests
poetry run pytest
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file arango_orm-1.1.0.tar.gz
.
File metadata
- Download URL: arango_orm-1.1.0.tar.gz
- Upload date:
- Size: 37.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: poetry/1.5.1 CPython/3.11.10 Linux/6.5.0-1025-azure
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 7c47e29fc6314e39e37fef3ea08a9429681c5c99d20f3039bf9a6aac8a49fde2 |
|
MD5 | 563534daa9c946791d563337dcf8795b |
|
BLAKE2b-256 | a5df4de504d0c72db24c7d94b8f595e7beabdf54058e3c8411d6504bbe1bd843 |
File details
Details for the file arango_orm-1.1.0-py3-none-any.whl
.
File metadata
- Download URL: arango_orm-1.1.0-py3-none-any.whl
- Upload date:
- Size: 35.5 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: poetry/1.5.1 CPython/3.11.10 Linux/6.5.0-1025-azure
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | a78d25f4c8329b6af9a9428ca4648e374ef2f6fc712c92b474eb10beefb40087 |
|
MD5 | 222a9d649d7e565509895b6639b80352 |
|
BLAKE2b-256 | f0f6968976e62a4b39883853663657b5b1ff6adf324e167d05d8153f57bbb9cf |