Coverage Report
File Stmts Miss Cover TOTAL 940 0 100%
Flow-based Programming¶
A lightweight framework for flow-based programming in python.
+-------------------+ +---------------------+
| Invite People | | Birthday Party |
|-------------------| |---------------------|
o amount<4> | +----->o attendees<> |
| people o---+ +--->o cake<> |
+-------------------+ | +---------------------+
|
+-------------------+ |
| Bake a cake | |
+-------------------+ |
o type<"Chocolate"> | |
| cake o-----+
+-------------------+
Benefits:
Visualize code
Re-usability
Streamlined code design
Built-in concurrency
Represent workflows one to one in the code
Quick Example¶
Consider this simple example on how to represent the construction of a house with Flowpipe:
from flowpipe import Graph, INode, Node, InputPlug, OutputPlug
class HireWorkers(INode):
"""A node can be derived from the INode interface.
The plugs are defined in the init method.
The compute method received the inputs from any connected upstream nodes.
"""
def __init__(self, amount=None, **kwargs):
super(HireWorkers, self).__init__(**kwargs)
InputPlug('amount', self, amount)
OutputPlug('workers', self)
def compute(self, amount):
workers = ['John', 'Jane', 'Mike', 'Michelle']
print('{0} workers are hired to build the house.'.format(amount))
return {'workers.{0}'.format(i): workers[i] for i in range(amount)}
@Node(outputs=['workers'])
def Build(workers, section):
"""A node can also be created by the Node decorator.outputs
The inputs to the function are turned into InputsPlugs, otuputs are defined
in the decorator itself. The wrapped function is used as the compute method.
"""
print('{0} are building the {1}'.format(', '.join(workers.values()), section))
return {'workers.{0}'.format(i): worker for i, worker in workers.items()}
@Node()
def Party(attendees):
print('{0} and {1} are having a great party!'.format(
', '.join(list(attendees.values())[:-1]), list(attendees.values())[-1]))
# Create a graph with the necessary nodes
graph = Graph(name='How to build a house')
workers = HireWorkers(graph=graph, amount=4)
build_walls = Build(graph=graph, name='Build Walls', section='walls')
build_roof = Build(graph=graph, name='Build Roof', section='roof')
party = Party(graph=graph, name='Housewarming Party')
# Wire up the connections between the nodes
workers.outputs['workers']['0'].connect(build_walls.inputs['workers']['0'])
workers.outputs['workers']['1'].connect(build_walls.inputs['workers']['1'])
workers.outputs['workers']['2'].connect(build_roof.inputs['workers']['0'])
workers.outputs['workers']['3'].connect(build_roof.inputs['workers']['1'])
build_walls.outputs['workers']['0'] >> party.inputs['attendees']['0']
build_walls.outputs['workers']['1'] >> party.inputs['attendees']['2']
build_roof.outputs['workers']['0'] >> party.inputs['attendees']['1']
build_roof.outputs['workers']['1'] >> party.inputs['attendees']['3']
party.inputs['attendees']['4'].value = 'Homeowner'
Visualize the code as a graph or as a listing:
print(graph.name)
print(graph)
print(graph.list_repr())
Output:
How to build a house
+------------------------+ +------------------------+ +---------------------------+
| HireWorkers | | Build Roof | | Housewarming Party |
|------------------------| |------------------------| |---------------------------|
o amount<4> | o section<"roof"> | % attendees |
| workers % % workers | +--->o attendees.0<> |
| workers.0 o-----+--->o workers.0<> | |--->o attendees.1<> |
| workers.1 o-----|--->o workers.1<> | |--->o attendees.2<> |
| workers.2 o-----| | workers % |--->o attendees.3<> |
| workers.3 o-----| | workers.0 o-----| o attendees.4<"Homeowner> |
+------------------------+ | | workers.1 o-----| +---------------------------+
| +------------------------+ |
| +------------------------+ |
| | Build Walls | |
| |------------------------| |
| o section<"walls"> | |
| % workers | |
+--->o workers.0<> | |
+--->o workers.1<> | |
| workers % |
| workers.0 o-----+
| workers.1 o-----+
+------------------------+
Build a House
HireWorkers
[i] amount: 4
[o] workers
[o] workers.0 >> Build Walls.workers.0
[o] workers.1 >> Build Walls.workers.1
[o] workers.2 >> Build Roof.workers.0
[o] workers.3 >> Build Roof.workers.1
Build Roof
[i] section: "roof"
[i] workers
[i] workers.0 << HireWorkers.workers.2
[i] workers.1 << HireWorkers.workers.3
[o] workers
[o] workers.0 >> Housewarming Party.attendees.1
[o] workers.1 >> Housewarming Party.attendees.3
Build Walls
[i] section: "walls"
[i] workers
[i] workers.0 << HireWorkers.workers.0
[i] workers.1 << HireWorkers.workers.1
[o] workers
[o] workers.0 >> Housewarming Party.attendees.0
[o] workers.1 >> Housewarming Party.attendees.2
Housewarming Party
[i] attendees
[i] attendees.0 << Build Walls.workers.0
[i] attendees.1 << Build Roof.workers.0
[i] attendees.2 << Build Walls.workers.1
[i] attendees.3 << Build Roof.workers.1
[i] attendees.4: "Homeowner"
Now build the house:
graph.evaluate(mode='threading') # Options are linear, threading and multiprocessing
Output:
4 workers are hired to build the house.
Michelle, Mike are building the roof
Jane, John are building the walls
Mike, John, Michelle, Jane and Homeowner are having a great party!
(Note: for more elaborate evaluation schemes, see Evaluators)
We now know how to throw a party, so let’s invite some people and re-use these skills for a birthday:
graph = Graph(name='How to throw a birthday party')
@Node(outputs=['people'])
def InvitePeople(amount):
people = ['John', 'Jane', 'Mike', 'Michelle']
d = {'people.{0}'.format(i): people[i] for i in range(amount)}
d['people'] = {people[i]: people[i] for i in range(amount)}
return d
invite = InvitePeople(graph=graph, amount=4)
birthday_party = Party(graph=graph, name='Birthday Party')
invite.outputs['people'] >> birthday_party.inputs['attendees']
print(graph.name)
print(graph)
graph.evaluate()
Output:
How to throw a birthday party
+-------------------+ +---------------------+
| InvitePeople | | Birthday Party |
|-------------------| |---------------------|
o amount<4> | +--->o attendees<> |
| people o-----+ +---------------------+
+-------------------+
Jane, Michelle, Mike and John are having a great party!
More Examples¶
There are more examples for common use cases of flowpipe:
The code for these examples: house_and_birthday.py!
Another simple example: world_clock.py!
How to make use of nested subgraphs: nested_graphs.py!
Using the command pattern with flowpipe successfully: workflow_design_pattern.py!
Use flowpipe on a remote cluster of machines, commonly refered to as a “render farm” in the VFX/Animation industry: vfx_render_farm_conversion.py!
An example graph showcasing a common workflow encountered in the VFX/Animation industry: vfx_rendering.py!
VFX Pipeline¶
If you are working in the VFX/Animation industry, please check out this extensive guide on how to use flowpipe in a vfx pipeline!
Evaluators¶
If your nodes just need sequential, threaded or multiprocessing evaluation, the Graph.evaluate() method will serve you just fine. If you want to take more control over the way your Graph is being evaluated, Evaluators are for you. This can also be used to add, e.g. logging or tracing to node evaluation.
Evaluators allow you to take control of node evaluation order, or their scheduling.
See flowpipe/evaluator.py to see the Graph.evaluate() method’s evaluation schemes.
To use a custom evaluator, subclass flowpipe.evaluator.Evaluator, and provide at least an _evaluate_nodes(self, nodes) method.
This method should take a list of nodes and call their respective node.evalaute() methods (along with any other task you want to do for each node being evaluated).
To use a cusom evaluator, create it and call its Evalator.evaluate() method with the Graph to evaluate as an argument:
from flowpipe.evaluators import LinearEvaluator
# assuming you created a graph to evaluate above, called `graph`
lin_eval = LinearEvaluator()
lin_eval.evaluate(graph)
