Module: Concurrent::ErlangActor

This module provides actor abstraction that has same behaviour as Erlang actor.

Examples

The simplest example is to use the actor as an asynchronous execution. Although, Promises.future { 1 + 1 } is better suited for that purpose.

actor = Concurrent::ErlangActor.spawn(type: :on_thread, name: 'addition') { 1 + 1 } # => # actor.terminated.value! # => 2 

Let's send some messages and maintain some internal state which is what actors are good for.

actor = Concurrent::ErlangActor.spawn(type: :on_thread, name: 'sum') do sum = 0 # internal state  # receive and sum the messages until the actor gets :done  while true message = receive break if message == :done # if the message is asked and not only told,  # reply with the current sum (has no effect if actor was not asked)  reply sum += message end # The final value of the actor  sum end # => # 

The actor can be either told a message asynchronously, or asked. The ask method will block until actor replies.

# tell returns immediately returning the actor actor.tell(1).tell(1) # => # # blocks, waiting for the answer actor.ask 10 # => 12 # stop the actor actor.tell :done # => # # The final value of the actor actor.terminated.value! # => 12 

Actor types

There are two types of actors. The type is specified when calling spawn as a first argument, Concurrent::ErlangActor.spawn(type: :on_thread, ... or Concurrent::ErlangActor.spawn(type: :on_pool, ....

The main difference is in how receive method returns.

• :on_thread it blocks the thread until message is available, then it returns or calls the provided block first.

• However, :on_pool it has to free up the thread on the receive call back to the pool. Therefore the call to receive ends the execution of current scope. The receive has to be given block or blocks that act as a continuations and are called when there is message available.

Let's have a look at how the bodies of actors differ between the types:

ping = Concurrent::ErlangActor.spawn(type: :on_thread) { reply receive } # => # ping.ask 42 # => 42 

It first calls receive, which blocks the thread of the actor. When it returns the received message is passed an an argument to reply, which replies the same value back to the ask method. Then the actor terminates normally, because there is nothing else to do.

However when running on pool a block with code which should be evaluated after the message is received has to be provided.

ping = Concurrent::ErlangActor.spawn(type: :on_pool) { receive { |m| reply m } } # => # ping.ask 42 # => 42 

It starts by calling receive which will remember the given block for later execution when a message is available and stops executing the current scope. Later when a message becomes available the previously provided block is given the message and called. The result of the block is the final value of the normally terminated actor.

The direct blocking style of :on_thread is simpler to write and more straight forward however it has limitations. Each :on_thread actor creates a Thread taking time and resources. There is also a limited number of threads the Ruby process can create so you may hit the limit and fail to create more threads and therefore actors.

Since the :on_pool actor runs on a poll of threads, its creations is faster and cheaper and it does not create new threads. Therefore there is no limit (only RAM) on how many actors can be created.

To simplify, if you need only few actors :on_thread is fine. However if you will be creating hundreds of actors or they will be short-lived :on_pool should be used.

Receiving messages

Simplest message receive.

actor = Concurrent::ErlangActor.spawn(type: :on_thread) { receive } # => # actor.tell :m # => # actor.terminated.value! # => :m 

which also works for actor on pool, because if no block is given it will use a default block { |v| v }

actor = Concurrent::ErlangActor.spawn(type: :on_pool) { receive { |v| v } } # => # # can simply be following actor = Concurrent::ErlangActor.spawn(type: :on_pool) { receive } # => # actor.tell :m # => # actor.terminated.value! # => :m 

The received message type can be limited.

Concurrent::ErlangActor. spawn(type: :on_thread) { receive(Numeric).succ }. tell('junk'). # ignored message  tell(42). terminated.value! # => 43 

On pool it requires a block.

Concurrent::ErlangActor. spawn(type: :on_pool) { receive(Numeric) { |v| v.succ } }. tell('junk'). # ignored message  tell(42). terminated.value! # => 43 

By the way, the body written for on pool actor will work for on thread actor as well.

Concurrent::ErlangActor. spawn(type: :on_thread) { receive(Numeric) { |v| v.succ } }. tell('junk'). # ignored message  tell(42). terminated.value! # => 43 

The receive method can be also used to dispatch based on the received message.

actor = Concurrent::ErlangActor.spawn(type: :on_thread) do while true receive(on(Symbol) { |s| reply s.to_s }, on(And[Numeric, -> v { v >= 0 }]) { |v| reply v.succ }, # put last works as else  on(ANY) do |v| reply :bad_message terminate [:bad_message, v] end) end end # => # actor.ask 1 # => 2 actor.ask 2 # => 3 actor.ask :value # => "value" # this malformed message will terminate the actor actor.ask -1 # => :bad_message # the actor is no longer alive, so ask fails actor.ask "junk" rescue $! # => #> actor.terminated.result # => [false, nil, [:bad_message, -1]] 

And a same thing for the actor on pool. Since it cannot loop it will call the body method repeatedly.

module Behaviour def body receive(on(Symbol) do |s| reply s.to_s body # call again  end, on(And[Numeric, -> v { v >= 0 }]) do |v| reply v.succ body # call again  end, # put last works as else  on(ANY) do |v| reply :bad_message terminate [:bad_message, v] end) end end # => :body  actor = Concurrent::ErlangActor.spawn(type: :on_pool, environment: Behaviour) { body } # => # actor.ask 1 # => 2 actor.ask 2 # => 3 actor.ask :value # => "value" # this malformed message will terminate the actor actor.ask -1 # => :bad_message # the actor is no longer alive, so ask fails actor.ask "junk" rescue $! # => #> actor.terminated.result # => [false, nil, [:bad_message, -1]] 

Since the behavior is stable in this case we can simplify with the :keep option that will keep the receive rules until another receive is called replacing the kept rules.

actor = Concurrent::ErlangActor.spawn(type: :on_pool) do receive(on(Symbol) { |s| reply s.to_s }, on(And[Numeric, -> v { v >= 0 }]) { |v| reply v.succ }, # put last works as else  on(ANY) do |v| reply :bad_message terminate [:bad_message, v] end, keep: true) end # => # actor.ask 1 # => 2 actor.ask 2 # => 3 actor.ask :value # => "value" # this malformed message will terminate the actor actor.ask -1 # => :bad_message # the actor is no longer alive, so ask fails actor.ask "junk" rescue $! # => #> actor.terminated.result # => [false, nil, [:bad_message, -1]] 

Erlang behaviour

The actor matches Erlang processes in behaviour. Therefore it supports the usual Erlang actor linking, monitoring, exit behaviour, etc.

actor = Concurrent::ErlangActor.spawn(type: :on_thread) do spawn(link: true) do # equivalent of spawn_link in Erlang  terminate :err # equivalent of exit in Erlang  end trap # equivalent of process_flag(trap_exit, true)  receive end # => # actor.terminated.value! # => ## @from= # #, # @reason=:err> 

The methods have same or very similar name to be easily found. The one exception from the original Erlang naming is exit. To avoid clashing with Kernel#exit it's called terminate.

Until there is more information available here, the chapters listed below from a book learn you some Erlang are excellent source of information. The Ruby ErlangActor implementation has same behaviour.

• Links
• It's a trap
• Monitors

If anything behaves differently than in Erlang, please file an issue.

Chapters or points to be added

• More erlang behaviour examples.
• The mailbox can be bounded in size, then the tell and ask will block until there is space available in the mailbox. Useful for building systems with backpressure.
• #tell_op and ask_op method examples, integration with promises.
• Best practice: always use timeout, and do something if the message does not arrive, don't leave the actor stuck.
• Best practice: drop and log unrecognized messages, or be even more defensive and terminate.
• Environment definition for actors.