Skip to main content

Service Propagation

Basic Knowledge Requirements

Before diving into this code, here's a quick heads-up on what you'll need to be familiar with:

  1. Python Programming: It's important to have a good grasp of Python, especially with concepts like functions, loops, and classes, since the example utilizes these fundamentals.
  2. Asynchronous Programming with asyncio: Familiarity with Python's asyncio for writing concurrent code using the async/await syntax.

In the Service Propagation example, we will show how create a subsystem of services that can be composed together to create a more complex system and leverage bidirectional communication between services to propagate information.

In particular, we will create an Agent service that will have several tasks running concurrently sampling data from different tasks. The Agent service will publish the data from the tasks at fixed rate to a /sample path that will be subscribed by a Supervisor service.

At the same time, the Agent service will listen to a /update_residual path to receive the residual value from another the Supervisor service.

class AgentServer:
def __init__(self, event_service: EventServiceGrpc) -> None:
"""Initialize the service.
Args:
event_service: The event service to use for communication.
"""
self._event_service = event_service
self._event_service.add_request_reply_handler(self.request_reply_handler)

args: dict[str, float] = {}
for arg in self._event_service.config.args:
key, value = arg.split("=")
args[key] = value

# the rate in hertz to send commands
self._rate = float(args["rate"])
self._num_tasks = int(args["num_tasks"])

self._remainder: int = 1e6

async def request_reply_handler(self, event: Event, message) -> None:
"""The callback for handling request/reply messages."""
if event.uri.path == "/update_residual":
self._remainder = message.value
self._event_service.logger.info(f"Remainder: {self._remainder}")

return Empty()

async def run_task(self, task_id: int) -> None:
"""Run the main task."""
while True:
if self._remainder <= 0:
await asyncio.sleep(0.01)
continue

message = Struct()
message["sample"] = random.random()
message["task_id"] = task_id

await self._event_service.publish("/sample", message)
await asyncio.sleep(1.0 / self._rate)
print(f"Published sample {message['sample']} from task {task_id}")

async def serve(self) -> None:
"""Run the service."""
tasks: list[asyncio.Task] = [asyncio.create_task(self.run_task(i)) for i in range(self._num_tasks)]
await asyncio.gather(self._event_service.serve(), *tasks)

In addition, we will create a SupervisorServer service that will be subscribed to the /sample path of the Agent service. The SupervisorServer will listen to the samples and send a request update to the StorageServer service when the sample is above a certain confidence threshold.

class SupervisorServer:
def __init__(self, event_service: EventServiceGrpc, config_list: EventServiceConfigList) -> None:
"""Initialize the service.
Args:
event_service: The event service to use for communication.
"""
self._event_service = event_service

self._clients: dict[str, EventClient] = {
config.name: EventClient(config)
for config in config_list.configs
if config.name != event_service.config.name
}

args: dict[str, float] = {}
for arg in self._event_service.config.args:
key, value = arg.split("=")
args[key] = value

# the rate in hertz to send commands
self._confidence = float(args["confidence"])

async def subscribe(self, subscripton) -> None:
"""Run the main task."""
# create the event client
service_name = subscripton.uri.query.split("=")[-1]
client = self._clients[service_name]

async for event, message in client.subscribe(subscripton, decode=True):
if message["sample"] > self._confidence:
residual = await self._clients["storage"].request_reply("/update_storage", Empty(), decode=True)
self._event_service.logger.info(f"Residual: {residual}")
await client.request_reply("/update_residual", residual)

async def serve(self) -> None:
"""Run the service."""
tasks: list[asyncio.Task] = []
for subscription in self._event_service.config.subscriptions:
tasks.append(asyncio.create_task(self.subscribe(subscription)))
await asyncio.gather(self._event_service.serve(), *tasks)

The last service we will create is a StorageServer service that will be listening to the /update_storage path. The StorageServer will keep track of the storage capacity and will remove from storage at fixed rate.

class StorageServer:
def __init__(self, event_service: EventServiceGrpc) -> None:
"""Initialize the service.
Args:
event_service: The event service to use for communication.
"""
self._event_service = event_service
self._event_service.add_request_reply_handler(self.request_reply_handler)

self._storage: int = 0

args: dict[str, float] = {}
for arg in self._event_service.config.args:
key, value = arg.split("=")
args[key] = value

# the maximum storage capacity
self._max_storage = int(args["max_storage"])

# the batch size to remove from storage
self._batch_size = int(args["batch_size"])

async def request_reply_handler(self, event: Event, message: Empty) -> None:
"""The callback for handling request/reply messages."""
if event.uri.path == "/update_storage":
self._storage += 1
residual: int = self._max_storage - self._storage
self._event_service.logger.info(
f"Storage: {self._storage}/{self._max_storage} ({residual} remaining)")
return Int32Value(value=residual)

return Empty()

async def remove_from_storage(self) -> None:
"""Remove from storage."""
while True:

if self._storage < self._batch_size:
await asyncio.sleep(0.05)
continue

self._storage -= self._batch_size

self._event_service.logger.info(f"Removed from storage: {self._batch_size}/{self._max_storage}")

await asyncio.sleep(0.1)

async def serve(self) -> None:
tasks: list[asyncio.Task] = [
asyncio.create_task(self._event_service.serve()),
asyncio.create_task(self.remove_from_storage()),
]
await asyncio.gather(*tasks)

Requirements​

This example only requires the farm-ng-core package.

pip3 install farm-ng-core
tip

We highly recommend to have some basic knowledge about asyncio, gRPC and protobuf.

Service configuration​

The service configuration is defined in the service_config.json file which describes how each service instance is connected to each other and how they are configured. As an example, the agent-1 service is configured as follows:

{
"name": "agent-1",
"args": [
"rate=1.0",
"num_tasks=3"
],
"subscriptions": [
{
"uri": "event://supervisor-1/sample?service=agent-1"
}
]
}

Run the example​

1. Run the storage service​

python storage.py --service-config service_config.json --service-name storage

2. Run the supervisor services​

python supervisor.py --service-config service_config.json --service-name supervisor-1
python supervisor.py --service-config service_config.json --service-name supervisor-2

3. Run the agent services​

python agent.py --service-config service_config.json --service-name agent-1
python agent.py --service-config service_config.json --service-name agent-2
python agent.py --service-config service_config.json --service-name agent-3

Expected output​

In the storage service terminal, you should see the capacity of the storage which is decreasing as the agent services are publishing samples. As soon as the storage capacity is full, the agent services will stop publishing samples until the supervisor services request the agent services to update the residual value.

Storage: 37/500 (463 remaining)
Storage: 38/500 (462 remaining)
Storage: 39/500 (461 remaining)
Storage: 40/500 (460 remaining)
Storage: 41/500 (459 remaining)
Storage: 42/500 (458 remaining)
Storage: 43/500 (457 remaining)
Storage: 44/500 (456 remaining)
Storage: 45/500 (455 remaining)
Storage: 46/500 (454 remaining)
Storage: 47/500 (453 remaining)
Storage: 48/500 (452 remaining)

In the supervisor service terminal, you should see the residual value decreasing as the agent services are publishing samples. As soon as the residual value is zero, the supervisor services will request the agent services to update the residual value.

INFO:supervisor-1:Residual: value: 455

Residual: value: 454

INFO:supervisor-1:Residual: value: 454

Residual: value: 453

INFO:supervisor-1:Residual: value: 453

Residual: value: 452

Finally, in the agent service terminal, you should see the samples being published at fixed rate. As soon as the residual value is zero, the agent services will stop publishing samples until the supervisor services request the agent services to update the residual value.

Published sample 0.7213814924044664 from task 1
Published sample 0.19269278385883515 from task 2
Published sample 0.41566478402029516 from task 3
Published sample 0.6616156115367178 from task 4
Published sample 0.699833473381781 from task 5
Published sample 0.1892017291616711 from task 6
Published sample 0.6031618279156414 from task 7
Published sample 0.6306753264890216 from task 8
Published sample 0.9169593227439662 from task 9