Exception raised when trying to use a dead or unavailable actor.
To create an actor:
To stop an actor, call Actor.stop() or ActorRef.stop().
For example:
import pykka
class MyActor(pykka.ThreadingActor):
def __init__(self, my_arg=None):
super(MyActor, self).__init__()
... # My optional init code with access to start() arguments
def on_start(self):
... # My optional setup code in same context as on_receive()
def on_stop(self):
... # My optional cleanup code in same context as on_receive()
def on_failure(self, exception_type, exception_value, traceback):
... # My optional cleanup code in same context as on_receive()
def on_receive(self, message):
... # My optional message handling code for a plain actor
def a_method(self, ...):
... # My regular method to be used through an ActorProxy
my_actor_ref = MyActor.start(my_arg=...)
my_actor_ref.stop()
Start an actor and register it in the ActorRegistry.
Any arguments passed to start() will be passed on to the class constructor.
Behind the scenes, the following is happening when you call start():
| Returns: | a ActorRef which can be used to access the actor in a safe manner |
|---|
The actor URN string is a universally unique identifier for the actor. It may be used for looking up a specific actor using ActorRegistry.get_by_urn.
The actor’s inbox. Use ActorRef.tell(), ActorRef.ask(), and friends to put messages in the inbox.
The actor’s ActorRef instance.
Stop the actor.
The actor will finish processing any messages already in its queue before stopping. It may not be restarted.
Hook for doing any setup that should be done after the actor is started, but before it starts processing messages.
For ThreadingActor, this method is executed in the actor’s own thread, while __init__() is executed in the thread that created the actor.
If an exception is raised by this method the stack trace will be logged, and the actor will stop.
Hook for doing any cleanup that should be done after the actor has processed the last message, and before the actor stops.
This hook is not called when the actor stops because of an unhandled exception. In that case, the on_failure() hook is called instead.
For ThreadingActor this method is executed in the actor’s own thread, immediately before the thread exits.
If an exception is raised by this method the stack trace will be logged, and the actor will stop.
Hook for doing any cleanup after an unhandled exception is raised, and before the actor stops.
For ThreadingActor this method is executed in the actor’s own thread, immediately before the thread exits.
The method’s arguments are the relevant information from sys.exc_info().
If an exception is raised by this method the stack trace will be logged, and the actor will stop.
May be implemented for the actor to handle regular non-proxy messages.
Messages where the value of the “command” key matches “pykka_*” are reserved for internal use in Pykka.
| Parameters: | message (picklable dict) – the message to handle |
|---|---|
| Returns: | anything that should be sent as a reply to the sender |
ThreadingActor implements Actor using regular Python threads.
This implementation is slower than GeventActor, but can be used in a process with other threads that are not Pykka actors.
A boolean value indicating whether this actor is executed on a thread that is a daemon thread (True) or not (False). This must be set before pykka.Actor.start() is called, otherwise RuntimeError is raised.
The entire Python program exits when no alive non-daemon threads are left. This means that an actor running on a daemon thread may be interrupted at any time, and there is no guarantee that cleanup will be done or that pykka.Actor.on_stop() will be called.
Actors do not inherit the daemon flag from the actor that made it. It always has to be set explicitly for the actor to run on a daemonic thread.
Reference to a running actor which may safely be passed around.
ActorRef instances are returned by Actor.start() and the lookup methods in ActorRegistry. You should never need to create ActorRef instances yourself.
| Parameters: | actor (Actor) – the actor to wrap |
|---|
The class of the referenced actor.
See Actor.actor_urn.
See Actor.actor_inbox.
Check if actor is alive.
This is based on whether the actor is registered in the actor registry or not. The actor is not guaranteed to be alive and responding even though is_alive() returns True.
| Returns: | Returns True if actor is alive, False otherwise. |
|---|
Send message to actor without waiting for any response.
Will generally not block, but if the underlying queue is full it will block until a free slot is available.
| Parameters: | message (picklable dict) – message to send |
|---|---|
| Raise : | pykka.ActorDeadError if actor is not available |
| Returns: | nothing |
Send message to actor and wait for the reply.
The message must be a picklable dict. If block is False, it will immediately return a Future instead of blocking.
If block is True, and timeout is None, as default, the method will block until it gets a reply, potentially forever. If timeout is an integer or float, the method will wait for a reply for timeout seconds, and then raise pykka.Timeout.
| Parameters: |
|
|---|---|
| Raise : | pykka.Timeout if timeout is reached |
| Raise : | pykka.ActorDeadError if actor is not available |
| Returns: | pykka.Future or response |
Send a message to the actor, asking it to stop.
The actor will finish processing any messages already in its queue before stopping. It may not be restarted.
block and timeout works as for ask().
| Returns: | True if actor is stopped. False if actor was already dead. |
|---|
Wraps the ActorRef in an ActorProxy.
Using this method like this:
proxy = AnActor.start().proxy()
is analogous to:
proxy = ActorProxy(AnActor.start())
| Raise : | pykka.ActorDeadError if actor is not available |
|---|---|
| Returns: | pykka.ActorProxy |
An ActorProxy wraps an ActorRef instance. The proxy allows the referenced actor to be used through regular method calls and field access.
You can create an ActorProxy from any ActorRef:
actor_ref = MyActor.start()
actor_proxy = ActorProxy(actor_ref)
You can also get an ActorProxy by using proxy():
actor_proxy = MyActor.start().proxy()
When reading an attribute or getting a return value from a method, you get a Future object back. To get the enclosed value from the future, you must call get() on the returned future:
print actor_proxy.string_attribute.get()
print actor_proxy.count().get() + 1
If you call a method just for it’s side effects and do not care about the return value, you do not need to accept the returned future or call get() on the future. Simply call the method, and it will be executed concurrently with your own code:
actor_proxy.method_with_side_effect()
If you want to block your own code from continuing while the other method is processing, you can use get() to block until it completes:
actor_proxy.method_with_side_effect().get()
An actor can use a proxy to itself to schedule work for itself. The scheduled work will only be done after the current message and all messages already in the inbox are processed.
For example, if an actor can split a time consuming task into multiple parts, and after completing each part can ask itself to start on the next part using proxied calls or messages to itself, it can react faster to other incoming messages as they will be interleaved with the parts of the time consuming task. This is especially useful for being able to stop the actor in the middle of a time consuming task.
To create a proxy to yourself, use the actor’s actor_ref attribute:
proxy_to_myself_in_the_future = self.actor_ref.proxy()
If you create a proxy in your actor’s constructor or on_start method, you can create a nice API for deferring work to yourself in the future:
def __init__(self):
...
self.in_future = self.actor_ref.proxy()
...
def do_work(self):
...
self.in_future.do_more_work()
...
def do_more_work(self):
...
An example of ActorProxy usage:
#! /usr/bin/env python
import pykka
class Adder(pykka.ThreadingActor):
def add_one(self, i):
print('{} is increasing {}'.format(self, i))
return i + 1
class Bookkeeper(pykka.ThreadingActor):
def __init__(self, adder):
super(Bookkeeper, self).__init__()
self.adder = adder
def count_to(self, target):
i = 0
while i < target:
i = self.adder.add_one(i).get()
print('{} got {} back'.format(self, i))
if __name__ == '__main__':
adder = Adder.start().proxy()
bookkeeper = Bookkeeper.start(adder).proxy()
bookkeeper.count_to(10).get()
pykka.ActorRegistry.stop_all()
| Parameters: | actor_ref (pykka.ActorRef) – reference to the actor to proxy |
|---|---|
| Raise : | pykka.ActorDeadError if actor is not available |
The actor’s pykka.ActorRef instance.
Exception raised at future timeout.
A Future is a handle to a value which are available or will be available in the future.
Typically returned by calls to actor methods or accesses to actor fields.
To get hold of the encapsulated value, call Future.get().
Get the value encapsulated by the future.
If the encapsulated value is an exception, it is raised instead of returned.
If timeout is None, as default, the method will block until it gets a reply, potentially forever. If timeout is an integer or float, the method will wait for a reply for timeout seconds, and then raise pykka.Timeout.
The encapsulated value can be retrieved multiple times. The future will only block the first time the value is accessed.
| Parameters: | timeout (float or None) – seconds to wait before timeout |
|---|---|
| Raise : | pykka.Timeout if timeout is reached |
| Raise : | encapsulated value if it is an exception |
| Returns: | encapsulated value if it is not an exception |
Set the encapsulated value.
| Parameters: | value (any picklable object or None) – the encapsulated value or nothing |
|---|
Set an exception as the encapsulated value.
You can pass an exc_info three-tuple, as returned by sys.exc_info(). If you don’t pass exc_info, sys.exc_info() will be called and the value returned by it used.
In other words, if you’re calling set_exception(), without any arguments, from an except block, the exception you’re currently handling will automatically be set on the future.
Changed in version 0.15: Previously, set_exception() accepted an exception instance as its only argument. This still works, but it is deprecated and will be removed in a future release.
| Parameters: | exc_info (three-tuple of (exc_class, exc_instance, traceback)) – the encapsulated exception |
|---|
ThreadingFuture implements Future for use with ThreadingActor.
The future is implemented using a Queue.Queue.
The future does not make a copy of the object which is set() on it. It is the setters responsibility to only pass immutable objects or make a copy of the object before setting it on the future.
Changed in version 0.14: Previously, the encapsulated value was a copy made with copy.deepcopy(), unless the encapsulated value was a future, in which case the original future was encapsulated.
Collect all values encapsulated in the list of futures.
If timeout is not None, the method will wait for a reply for timeout seconds, and then raise pykka.Timeout.
| Parameters: |
|
|---|---|
| Raise : | pykka.Timeout if timeout is reached |
| Returns: | list of results |
Registry which provides easy access to all running actors.
Contains global state, but should be thread-safe.
Broadcast message to all actors of the specified target_class.
If no target_class is specified, the message is broadcasted to all actors.
| Parameters: |
|
|---|
Get ActorRef for all running actors.
| Returns: | list of pykka.ActorRef |
|---|
Get ActorRef for all running actors of the given class, or of any subclass of the given class.
| Parameters: | actor_class (class) – actor class, or any superclass of the actor |
|---|---|
| Returns: | list of pykka.ActorRef |
Get ActorRef for all running actors of the given class name.
| Parameters: | actor_class_name (string) – actor class name |
|---|---|
| Returns: | list of pykka.ActorRef |
Get an actor by its universally unique URN.
| Parameters: | actor_urn (string) – actor URN |
|---|---|
| Returns: | pykka.ActorRef or None if not found |
Register an ActorRef in the registry.
This is done automatically when an actor is started, e.g. by calling Actor.start().
| Parameters: | actor_ref (pykka.ActorRef) – reference to the actor to register |
|---|
Stop all running actors.
block and timeout works as for ActorRef.stop().
If block is True, the actors are guaranteed to be stopped in the reverse of the order they were started in. This is helpful if you have simple dependencies in between your actors, where it is sufficient to shut down actors in a LIFO manner: last started, first stopped.
If you have more complex dependencies in between your actors, you should take care to shut them down in the required order yourself, e.g. by stopping dependees from a dependency’s on_stop() method.
| Returns: | If not blocking, a list with a future for each stop action. If blocking, a list of return values from pykka.ActorRef.stop(). |
|---|
Remove an ActorRef from the registry.
This is done automatically when an actor is stopped, e.g. by calling Actor.stop().
| Parameters: | actor_ref (pykka.ActorRef) – reference to the actor to unregister |
|---|
GeventFuture implements pykka.Future for use with GeventActor.
It encapsulates a gevent.event.AsyncResult object which may be used directly, though it will couple your code with gevent.
The encapsulated gevent.event.AsyncResult
GeventActor implements pykka.Actor using the gevent library. gevent is a coroutine-based Python networking library that uses greenlet to provide a high-level synchronous API on top of libevent event loop.
This is a very fast implementation, but as of gevent 0.13.x it does not work in combination with other threads.
Pykka uses Python’s standard logging module for logging debug statements and any unhandled exceptions in the actors. All log records emitted by Pykka are issued to the logger named “pykka”, or a sublogger of it.
Out of the box, Pykka is set up with logging.NullHandler as the only log record handler. This is the recommended approach for logging in libraries, so that the application developer using the library will have full control over how the log records from the library will be exposed to the application’s users. In other words, if you want to see the log records from Pykka anywhere, you need to add a useful handler to the root logger or the logger named “pykka” to get any log output from Pykka. The defaults provided by logging.basicConfig() is enough to get debug log statements out of Pykka:
import logging
logging.basicConfig(level=logging.DEBUG)
If your application is already using logging, and you want debug log output from your own application, but not from Pykka, you can ignore debug log messages from Pykka by increasing the threshold on the Pykka logger to “info” level or higher:
import logging
logging.basicConfig(level=logging.DEBUG)
logging.getLogger('pykka').setLevel(logging.INFO)
For more details on how to use logging, please refer to the Python standard library documentation.
Logs at INFO level a traceback for each running thread.
This can be a convenient tool for debugging deadlocks.
The function accepts any arguments so that it can easily be used as e.g. a signal handler, but it does not use the arguments for anything.
To use this function as a signal handler, setup logging with a logging.INFO threshold or lower and make your main thread register this with the signal module:
import logging
import signal
import pykka.debug
logging.basicConfig(level=logging.DEBUG)
signal.signal(signal.SIGUSR1, pykka.debug.log_thread_tracebacks)
If your application deadlocks, send the SIGUSR1 signal to the process:
kill -SIGUSR1 <pid of your process>
Signal handler caveats:
The morale is: setup signals using your main thread, start your actors, then let your main thread relax for the rest of your application’s life cycle.
For a complete example of how to use this, see examples/deadlock_debugging.py in Pykka’s source code.