The purpose of an Object-Graph Mapper, OGM for short, is to make interactions with large, complex graph databases more understandable and maintainable.
Similar to an Object-Relational Mapper (ORM) used with relational databases, OGMs bridge a gap between the higher-level object-oriented concepts (classes) in your software and the vertexes and edges in your database.
The pyorient OGM design was heavily inspired by the marvellous SQLAlchemy ORM.
Whether you are starting with an existing OrientDB schema or building one from scratch can make a difference to how you will use the mapper.
If you are working with an existing OrientDB schema, it may be more convenient to let the mapper automatically generate python classes.
On the other hand, and especially if you have already written a lot of python code, perhaps instead you would like the mapper to build the database schema from your python classes.
Whatever approach you take, once you have built up your dataset, naturally you will want to run queries against it. OrientDB provides many ways to get at your data; via SQL it has statements like SELECT and TRAVERSE. It also supports the Gremlin graph traversal language. The mapper currently lacks support for TRAVERSE, and its Gremlin support - though functional - could use work.
pyorient splits the process of connecting to your OrientDB server into two steps; specifying where and how to connect, and then actually connecting.
For these two steps, you will need to bring in two interfaces
from pyorient.ogm import Graph, Config
Graph is central to the OGM. It wraps the lower-level pyorient interface pyorient.OrientDB and does the job of mapping your python classes to a database schema, and vice versa.
To specify which database, and which database schema, Graph accepts a configuration. The Config classmethod, from_url is a convenient way to supply one. Along with a URL, it also requires a username and credential for connecting to the database at that URL.
All of these assume a username and password 'root':
configs = [
'localhost/a'
, 'plocal://localhost:2424/a'
, 'test'
, 'memory://localhost/test'
]
for conf in configs:
Config.from_url(conf, 'root', 'root')
The first two and the last two configs, above, denote the same database. The first called a, and the second called test. A port number of 2424 will be assumed when none is given. See the OrientDB documentation for more information about Paginated Local Storage.
If you browse the OGM tests from the pyorient repository, you will notice an optional argument they use for their Config:
g = self.g = Graph(Config.from_url('hardware', 'root', 'root'
, initial_drop=True))
This is done as a shorthand for clearing the previous database used for the same test.
Python classes will only be mapped to a database schema if they belong to a registry supplied to your Graph.
There are two types of registries, one to indicate a vertex or node type, and the other an edge or relationship type. Adding your python class to a registry is a matter of subclassing:
from pyorient.ogm.declarative import declarative_node, declarative_relationship
Node = declarative_node()
Relationship = declarative_relationship()
class Person(Node):
pass
class Likes(Relationship):
pass
Each call to declarative_node() and declarative_relationship() will create a new registry.
The mapper will preserve inheritance hierarchies of nodes and relationships.
To create the corresponding classes in the database schema, we pass registries to Graph:
self.g.create_all(Node.registry)
self.g.create_all(Relationship.registry)
If the classes already exist in your database schema, and you merely want your python classes to be bound, use the Graph include method, which similarly accepts registries.
Object-Graph Mapping happens on a few different levels. So far we have seen the mapping between python classes and schema classes. The next level of mapping is between objects of those classes, and the actual vertexes and edges in the graph. Brokers work on this level.
When the mapper creates its mappings between your vertex and edge classes, and the corresponding classes in the database, it will create a Broker for each class.
From a coding perspective, brokers allow a shorthand for working with the various types of vertexes and edges in your graph. In terms of software architecture, they can also reduce coupling. They hide the classes that you have mapped, to focus on the interfaces they expose.
class Foo(Node):
element_plural = 'foos'
name = String()
g.include(Node.registry) # Creates 'foos' object, and sets 'objects' attribute.
g.foos.create(name='Bar')
Foo.objects.create(name='Baz')
find_bar = g.foos.query(name='Bar')
Here, foos and Foo.objects are the same thing, an instance of Node.Broker.
Plurals make mapping code a lot more readable. The mapper is actually fairly stubborn about this: if your node class lacks an element_plural, or your relationship class lacks a label, the Graph - here, g - will not be given a broker for your class.
Notice how the broker create method hides whether you are working with a vertex or an edge. Manual use is more verbose:
g.create_vertex(Foo, name='Bar')
find_bar = g.query(Foo, name='Bar')
In the section above on brokers, we used, but did not elaborate much on the one Property of our example vertex class, name.
OrientDB and the pyorient mapper support a range of property types. Some of the more basic ones are:
Numeric Types | Other |
---|---|
Boolean | String |
Byte | Date |
Integer | DateTime |
Short | Binary |
Long | Embedded |
Float | |
Double | |
Decimal |
See the OrientDB documentation for more information on these and other types.
Soon, this documentation will include an overview of querying.
PyOrient does not (yet?) ship with a tool for generating python code from a database schema, but it does go half way there. Using the build_mapping method, you can generate a dictionary of python classes (a registry).
This dictionary is suitable for passing to the Graph include method.
from pyorient.ogm.declarative import declarative_node, declarative_relationship
SchemaNode = declarative_node()
SchemaRelationship = declarative_relationship()
classes_from_schema = g.build_mapping(SchemaNode, SchemaRelationship, auto_plural=True)
g.include(classes_from_schema)
In this example, the dynamically generated vertex types will have SchemaNode at the top of their inheritance tree, and edge types SchemaRelationship. Setting auto_plural=True means the subsequent include will assign brokers to the Graph, g.
If you want more nicely named brokers (with actual plural nouns), you will need a custom post-process of the registry returned by build_mapping.
The mapper has basic support for transactions. Besides being useful for concurrency, batching also reduces network round-trip time.
Starting a new batch is easy:
batch = g.batch()
This new batch will contain all the brokers the Graph g has.
Well, wrappers around those brokers using the same names.
So if you had the following types:
class Animal(Node):
element_type = 'animal'
element_plural = 'animals'
name = String(nullable=False, unique=True)
specie = String(nullable=False)
class Food(Node):
element_type = 'food'
element_plural = 'foods'
name = String(nullable=False, unique=True)
color = String(nullable=False)
class Eats(AnimalsRelationship):
label = 'eats'
... then adding vertexes to your batch would look like:
batch['zombie'] = batch.animals.create(name='Liv',specie='undead')
batch['brains'] = batch.foods.create(name='brains', color='grey')
This creates two batch variables, zombie and brains, that you can refer to later in the batch, say to create a relationship between them:
batch[:] = batch.eats.create(batch[:'zombie'], batch[:'brains']).retry(20)
We use slice syntax, above, when we do not need to refer to that relationship later in the batch.
A more natural, syntactic sugar for creating edges in a batch is:
batch[:] = batch[:'zombie'](Eats) > batch[:'brains']
Batches can optionally return a value. For example, if you want to further manipulate the newly created Food vertex:
brains = batch['$brains']
Otherwise, just use the commit method (or an empty slice - without assignment):
batch.commit()
The special attribute that we have not yet discussed, element_type, is redundant here. More generally, it tells the mapper what name to use for the corresponding schema class.
PyOrient has basic support for the Gremlin graph traversal, through Groovy scripts.
import pathlib
from pyorient.groovy import GroovyScripts
for p in pathlib.Path('scripts/').iterdir():
if p.is_file() and p.suffix == '.groovy':
g.scripts.add(GroovyScripts.from_file(str(p)))
All the functions found in the .groovy files under scripts/ can then be called. If one of these scripts had a find_people function taking one argument, their last name, you could say:
wallaces = g.gremlin('find_people', 'Wallace')
The Scripts.add, and the Graph.gremlin methods take an optional namespace argument, when you need such organisation.