Skip to content

Actors¤

Actors are a primitive unit of computation that runs autonomously.

Actor Programming Model¤

From Wikipedia

The actor model in computer science is a mathematical model of concurrent computation that treats an actor as the basic building block of concurrent computation. In response to a message it receives, an actor can: make local decisions, create more actors, send more messages, and determine how to respond to the next message received. Actors may modify their own private state, but can only affect each other indirectly through messaging (removing the need for lock-based synchronization).

We won't get into much more detail here because it is outside the scope of this documentation. However, if you are interested in learning more about the actor programming model, here are some useful resources:

Frequenz SDK Actors¤

The Actor class serves as the foundation for creating concurrent tasks and all actors in the SDK inherit from it. This class provides a straightforward way to implement actors. It shares similarities with the traditional actor programming model but also has some unique features:

  • Message Passing: Like traditional actors, our Actor class communicates through message passing. Even when no particular message passing mechanism is enforced, the SDK actors use frequenz.channels for communication.

  • Automatic Restart: If an unhandled exception occurs in an actor's logic (_run() method), the actor will be automatically restarted. This ensures robustness in the face of errors.

  • Simplified Task Management: Actors manage asynchronous tasks using asyncio. You can create and manage tasks within the actor, and the Actor class handles task cancellation and cleanup.

  • Simplified lifecycle management: Actors are [async context managers] and also a run() function is provided to easily run a group of actors and wait for them to finish.

Lifecycle¤

Actors are not started when they are created. There are 3 main ways to start an actor (from most to least recommended):

  1. By using the run() function.
  2. By using the actor as an async context manager.
  3. By using the start() method.

Warning

  1. If an actor raises an unhandled exception, it will be restarted automatically.

  2. Actors manage asyncio.Task objects, so a reference to the actor object must be held for as long as the actor is expected to be running, otherwise its tasks will be cancelled and the actor will stop. For more information, please refer to the Python asyncio documentation.

The run() Function¤

The run() function can start many actors at once and waits for them to finish. If any of the actors is stopped with errors, the errors will be logged.

Example
from frequenz.sdk.actor import Actor, run

class MyActor(Actor):
    async def _run(self) -> None:
        while True:
            print("Hello World!")
            await asyncio.sleep(1)

await run(MyActor()) # (1)!
  1. This line will block until the actor completes its execution or is manually stopped.

Async Context Manager¤

When an actor is used as an async context manager, it is started when the async with block is entered and stopped automatically when the block is exited (even if an unhandled exception is raised).

Example
from frequenz.sdk.actor import Actor

class MyActor(Actor):
    async def _run(self) -> None:
        print("Hello World!")

async with MyActor() as actor: # (1)!
    print("The actor is running")
# (2)!
  1. start() is called automatically when entering the async with block.
  2. stop() is called automatically when exiting the async with block.

The start() Method¤

When using the start() method, the actor is started in the background and the method returns immediately. The actor will continue to run until it is manually stopped or until it completes its execution.

Example
from frequenz.sdk.actor import Actor

class MyActor(Actor):
    async def _run(self) -> None:
        print("Hello World!")

actor = MyActor() # (1)!
actor.start() # (2)!
print("The actor is running") # (3)!
await actor.stop() # (4)!
  1. The actor is created but not started yet.
  2. The actor is started manually, it keeps running in the background.
  3. Danger

    If this function would raise an exception, the actor will never be stopped!

  4. Until the actor is stopped manually.

Warning

This method is not recommended because it is easy to forget to stop the actor manually, especially in error conditions.

Communication¤

The Actor class doesn't impose any specific way to communicate between actors. However, frequenz.channels are always used as the communication mechanism between actors in the SDK.

Implementing an Actor¤

To implement an actor, you must inherit from the Actor class and implement an initializer and the abstract _run() method.

Initialization¤

The initializer is called when the actor is created. The Actor class initializer (__init__) should be always called first in the class we are implementing to make sure actors are properly initialized.

The Actor.__init__() takes one optional argument, a name that will be used to identify the actor in logs. If no name is provided, a default name will be generated, but it is recommended that Actor subclasses can also receive a name as an argument to make it easier to identify individual instances in logs more easily.

The actor initializer normally also accepts as arguments the input channels receivers and output channels senders that will be used for communication. These channels should be created outside the actor and passed to it as arguments to ensure actors can be composed easily.

Example echo actor
from frequenz.channels import Receiver, Sender
from frequenz.sdk.actor import Actor

class EchoActor(Actor):  # (1)!
    def __init__(
            self,
            input: Receiver[int],  # (2)!
            output: Sender[int],  # (3)!
            name: str | None = None,  # (4)!
    ) -> None:
        super().__init__(name=name) # (5)!
        self._input: Receiver[int] = input  # (6)!
        self._output: Sender[int] = output  # (7)!
  1. We define a new actor class called EchoActor that inherits from Actor.

  2. We accept an input argument that will be used to receive messages from a channel.

  3. We accept an output argument that will be used to send messages to a channel.
  4. We accept an optional name argument that will be used to identify the actor in logs.
  5. We call Actor.__init__() to make sure the actor is properly initialized.
  6. We store the input argument in a private attribute to use it later.
  7. We store the output argument in a private attribute to use it later.

The _run() Method¤

The abstract _run() method is called automatically by the base class when the actor is started.

Normally an actor should run forever (or until it is stopped), so it is very common to have an infinite loop in the _run() method, typically receiving messages from one or more channels (receivers), processing them and sending the results to other channels (senders). However, it is worth noting that an actor can also be designed for a one-time execution or a limited number of runs, terminating upon completion.

Example echo actor
from frequenz.channels import Receiver, Sender
from frequenz.sdk.actor import Actor

class EchoActor(Actor):
    def __init__(
            self,
            input: Receiver[int],
            output: Sender[int],
            name: str | None = None,
    ) -> None:
        super().__init__(name=name)
        self._input: Receiver[int] = input
        self._output: Sender[int] = output

    async def _run(self) -> None:  # (1)!
        async for msg in self._input:  # (2)!
            await self._output.send(msg)  # (3)!
  1. We implement the abstract _run() method.
  2. We receive messages from the input channel one by one.
  3. We send the received message to the output channel.

Stopping¤

By default, the stop() method will call the cancel() method (which will cancel all the tasks created by the actor) and will wait for them to finish.

This means that when an actor is stopped, the _run() method will receive a CancelledError exception. You should have this in mind when implementing your actor and make sure to handle this exception properly if you need to do any cleanup.

The actor will handle the CancelledError exception automatically if it is not handled in the _run() method, so if there is no need for extra cleanup, you don't need to worry about it.

If an unhandled exception is raised in the _run() method, the actor will re-run the _run() method automatically. This ensures robustness in the face of errors, but you should also have this in mind if you need to do any cleanup to make sure the re-run doesn't cause any problems.

Tip

You could implement your own stop() method to, for example, send a message to a channel to notify that the actor should stop, or any other more graceful way to stop the actor if you need to make sure it can't be interrupted at any await point.

Spawning Extra Tasks¤

Actors run at least one background task, created automatically by the Actor class. But Actor inherits from BackgroundService, which provides a few methods to create and manage extra tasks.

If your actor needs to spawn extra tasks, you can use BackgroundService facilities to manage the tasks, so they are also automatically stopped when the actor is stopped.

All you need to do is add the newly spawned tasks to the actor's tasks set.

Example
import asyncio
from frequenz.sdk.actor import Actor

class MyActor(Actor):
    async def _run(self) -> None:
        extra_task = asyncio.create_task(self._extra_task())  # (1)!
        self.tasks.add(extra_task)  # (2)!
        while True:  # (3)!
            print("_run() running")
            await asyncio.sleep(1)

    async def _extra_task(self) -> None:
        while True:  # (4)!
            print("_extra_task() running")
            await asyncio.sleep(1.1)

async with MyActor() as actor:  # (5)!
    await asyncio.sleep(3)  # (6)!
# (7)!
  1. We create a new task using asyncio.create_task().
  2. We add the task to the actor's tasks set. This ensures the task will be cancelled and cleaned up when the actor is stopped.
  3. We leave the actor running forever.
  4. The extra task will also run forever.
  5. The actor is started.
  6. We wait for 3 seconds, the actor should print a bunch of "_run() running" and "_extra_task() running" messages while it's running.
  7. The actor is stopped and the extra task is cancelled automatically.

Examples¤

Here are a few simple but complete examples to demonstrate how to create actors and connect them using channels.

Tip

The code examples are annotated with markers (like ), you can click on them to see the step-by-step explanation of what's going on. The annotations are numbered according to the order of execution.

Composing actors¤

This example shows how to create two actors and connect them using broadcast channels.

compose.py
import asyncio

from frequenz.channels import Broadcast, Receiver, Sender
from frequenz.sdk.actor import Actor

class Actor1(Actor):  # (1)!
    def __init__(
        self,
        receiver: Receiver[str],
        output: Sender[str],
        name: str | None = None,
    ) -> None:
        super().__init__(name=name)
        self._receiver = receiver
        self._output = output

    async def _run(self) -> None:
        async for msg in self._receiver:
            await self._output.send(f"Actor1 forwarding: {msg!r}")  # (8)!


class Actor2(Actor):
    def __init__(
        self,
        receiver: Receiver[str],
        output: Sender[str],
        name: str | None = None,
    ) -> None:
        super().__init__(name=name)
        self._receiver = receiver
        self._output = output

    async def _run(self) -> None:
        async for msg in self._receiver:
            await self._output.send(f"Actor2 forwarding: {msg!r}")  # (9)!


async def main() -> None:  # (2)!
    # (4)!
    input_channel: Broadcast[str] = Broadcast("Input to Actor1")
    middle_channel: Broadcast[str] = Broadcast("Actor1 -> Actor2 stream")
    output_channel: Broadcast[str] = Broadcast("Actor2 output")

    input_sender = input_channel.new_sender()
    output_receiver = output_channel.new_receiver()

    async with (  # (5)!
        Actor1(input_channel.new_receiver(), middle_channel.new_sender(), "actor1"),
        Actor2(middle_channel.new_receiver(), output_channel.new_sender(), "actor2"),
    ):
        await input_sender.send("Hello")  # (6)!
        msg = await output_receiver.receive()  # (7)!
        print(msg)  # (10)!
    # (11)!

if __name__ == "__main__":  # (3)!
    asyncio.run(main())
  1. We define 2 actors: Actor1 and Actor2 that will just forward a message from an input channel to an output channel, adding some text.

  2. We define an async main() function within the main logic of our asyncio program.

  3. We start the main() function in the async loop using asyncio.run().

  4. We create a bunch of broadcast channels to connect our actors.

    • input_channel is the input channel for Actor1.
    • middle_channel is the channel that connects Actor1 and Actor2.
    • output_channel is the output channel for Actor2.
  5. We create two actors and use them as async context managers, Actor1 and Actor2, and connect them by creating new senders and receivers from the channels.

    Note

    We don't use the run() function here because we want to stop the actors when we are done with them, but the actors will run forever (as long as the channel is not closed). So the async context manager is a better fit for this example.

  6. We schedule the sending of the message Hello to Actor1 via input_channel.

  7. We receive (await) the response from Actor2 via output_channel. Between this and the previous steps the async calls in the actors will be executed.

  8. Actor1 sends the re-formatted message (Actor1 forwarding: Hello) to Actor2 via the middle_channel.

  9. Actor2 sends the re-formatted message (Actor2 forwarding: "Actor1 forwarding: 'Hello'") to the output_channel.

  10. Finally, we print the received message, which will still be Actor2 forwarding: "Actor1 forwarding: 'Hello'".

  11. The actors are stopped and cleaned up automatically when the async with block ends.

The expected output is:

Actor2 forwarding: "Actor1 forwarding: 'Hello'"
Receiving from multiple channels¤

This example shows how to create an actor that receives messages from multiple broadcast channels using select().

select.py
import asyncio

from frequenz.channels import Broadcast, Receiver, Sender
from frequenz.channels.util import select, selected_from
from frequenz.sdk.actor import Actor, run


class EchoActor(Actor):  # (1)!
    def __init__(
        self,
        receiver_1: Receiver[bool],
        receiver_2: Receiver[bool],
        output: Sender[bool],
        name: str | None = None,
    ) -> None:
        super().__init__(name=name)
        self._receiver_1 = receiver_1
        self._receiver_2 = receiver_2
        self._output = output

    async def _run(self) -> None:  # (2)!
        async for selected in select(self._receiver_1, self._receiver_2):  # (10)!
            if selected_from(selected, self._receiver_1):  # (11)!
                print(f"Received from receiver_1: {selected.value}")
                await self._output.send(selected.value)
                if not selected.value:  # (12)!
                    break
            elif selected_from(selected, self._receiver_2):  # (13)!
                print(f"Received from receiver_2: {selected.value}")
                await self._output.send(selected.value)
                if not selected.value:  # (14)!
                    break
            else:
                assert False, "Unknown selected channel"
        print("EchoActor finished")
    # (15)!


# (3)!
input_channel_1 = Broadcast[bool]("input_channel_1")
input_channel_2 = Broadcast[bool]("input_channel_2")
echo_channel = Broadcast[bool]("echo_channel")

echo_actor = EchoActor(  # (4)!
    input_channel_1.new_receiver(),
    input_channel_2.new_receiver(),
    echo_channel.new_sender(),
    "echo-actor",
)

echo_receiver = echo_channel.new_receiver()  # (5)!

async def main() -> None:  # (6)!
    # (8)!
    await input_channel_1.new_sender().send(True)
    await input_channel_2.new_sender().send(False)

    await run(echo_actor)  # (9)!

    await echo_channel.close()  # (16)!

    async for message in echo_receiver:  # (17)!
        print(f"Received {message=}")


if __name__ == "__main__":  # (7)!
    asyncio.run(main())
  1. We define an EchoActor that receives messages from two channels and sends them to another channel.

  2. We implement the _run() method that will receive messages from the two channels using select() and send them to the output channel. The run() method will stop if a False message is received.

  3. We create the channels that will be used with the actor.

  4. We create the actor and connect it to the channels by creating new receivers and senders from the channels.

  5. We create a receiver for the echo_channel to eventually receive the messages sent by the actor.

  6. We define the main() function that will run the actor.

  7. We start the main() function in the async loop using asyncio.run().

  8. We send a message to each of the input channels. These messages will be queued in the channels until they are consumed by the actor.

  9. We start the actor and wait for it to finish using the run() function.

  10. The select() function will get the first message available from the two channels. The order in which they will be handled is unknown, but in this example we assume that the first message will be from input_channel_1 (True) and the second from input_channel_1 (False).

  11. The selected_from() function will return True for the input_channel_1 receiver. selected.value holds the received message, so "Received from receiver_1: True" will be printed and True will be sent to the output channel.

  12. Since selected.value is True, the loop will continue, going back to the select() function.

  13. The selected_from() function will return False for the input_channel_1 receiver and True for the input_channel_2 receiver. The message stored in selected.value will now be False, so "Received from receiver_2: False" will be printed and False will be sent to the output channel.

  14. Since selected.value is False, the loop will break.

  15. The _run() method will finish normally and the actor will be stopped, so the run() function will return.

  16. We close the echo_channel to make sure the echo_receiver will stop receiving messages after all the queued messages are consumed (otherwise the step 17 will never end!).

  17. We receive the messages sent by the actor to the echo_channel one by one and print them, it should print first Received message=True and then Received message=False.

The expected output is:

Received from receiver_1: True
Received from receiver_2: False
Received message=True
Received message=False