Async SourceAsyncTask

class Task

A sequence of instructions, defined by a block, which is executed sequentially and managed by the scheduler. A task can be in one of the following states: initialized, running, completed, failed, cancelled or stopped.

stateDiagram-v2 [*] --> Initialized Initialized --> Running : Run Running --> Completed : Return Value Running --> Failed : Exception Completed --> [*] Failed --> [*] Running --> Stopped : Stop Stopped --> [*] Completed --> Stopped : Stop Failed --> Stopped : Stop Initialized --> Stopped : Stop

Example

require 'async'

# Create an asynchronous task that sleeps for 1 second:
Async do |task|
	sleep(1)
end

Signature

public

Since stable-v1.

Nested

Definitions

def self.yield

  • deprecated

Signature

deprecated

With no replacement.

Implementation

def self.yield
	Fiber.scheduler.transfer
end

def yield

Yield back to the reactor and allow other fibers to execute.

Implementation

def yield
	Fiber.scheduler.yield
end

def initialize(parent = Task.current?, finished: nil, **options, &block)

Create a new task.

Signature

parameter reactor Reactor

the reactor this task will run within.

parameter parent Task

the parent task.

Implementation

def initialize(parent = Task.current?, finished: nil, **options, &block)
	super(parent, **options)
	
	# These instance variables are critical to the state of the task.
	# In the initialized state, the @block should be set, but the @fiber should be nil.
	# In the running state, the @fiber should be set.
	# In a finished state, the @block should be nil, and the @fiber should be nil.
	@block = block
	@fiber = nil
	
	@status = :initialized
	@result = nil
	@finished = finished
	
	@defer_stop = nil
end

def sleep(duration = nil)

  • deprecated

Signature

deprecated

Prefer Kernel#sleep except when compatibility with stable-v1 is required.

Implementation

def sleep(duration = nil)
	super
end

def with_timeout(duration, exception = TimeoutError, message = "execution expired", &block)

Execute the given block of code, raising the specified exception if it exceeds the given duration during a non-blocking operation.

Implementation

def with_timeout(duration, exception = TimeoutError, message = "execution expired", &block)
	Fiber.scheduler.with_timeout(duration, exception, message, &block)
end

attr :fiber

def alive?

Whether the internal fiber is alive, i.e. it

Implementation

def alive?
	@fiber&.alive?
end

def finished?

Whether we can remove this node from the reactor graph.

Signature

returns Boolean

Implementation

def finished?
	# If the block is nil and the fiber is nil, it means the task has finished execution. This becomes true after `finish!` is called.
	super && @block.nil? && @fiber.nil?
end

def running?

Whether the task is running.

Signature

returns Boolean

Implementation

def running?
	@status == :running
end

def stopped?

The task has been stopped

Implementation

def stopped?
	@status == :stopped
end

def completed?

The task has completed execution and generated a result.

Implementation

def completed?
	@status == :completed
end

attr :status

def run(*arguments)

Begin the execution of the task.

Implementation

def run(*arguments)
	if @status == :initialized
		@status = :running
		
		schedule do
			@block.call(self, *arguments)
		end
	else
		raise RuntimeError, "Task already running!"
	end
end

def async(*arguments, **options, &block)

Run an asynchronous task as a child of the current task.

Implementation

def async(*arguments, **options, &block)
	raise FinishedError if self.finished?
	
	task = Task.new(self, **options, &block)
	
	task.run(*arguments)
	
	return task
end

def wait

Retrieve the current result of the task. Will cause the caller to wait until result is available. If the task resulted in an unhandled error (derived from StandardError), this will be raised. If the task was stopped, this will return nil.

Conceptually speaking, waiting on a task should return a result, and if it throws an exception, this is certainly an exceptional case that should represent a failure in your program, not an expected outcome. In other words, you should not design your programs to expect exceptions from #wait as a normal flow control, and prefer to catch known exceptions within the task itself and return a result that captures the intention of the failure, e.g. a TimeoutError might simply return nil or false to indicate that the operation did not generate a valid result (as a timeout was an expected outcome of the internal operation in this case).

Signature

raises RuntimeError

If the task's fiber is the current fiber.

returns Object

The final expression/result of the task's block.

Implementation

def wait
	raise "Cannot wait on own fiber!" if Fiber.current.equal?(@fiber)
	
	# `finish!` will set both of these to nil before signaling the condition:
	if @block || @fiber
		@finished ||= Condition.new
		@finished.wait
	end
	
	if @status == :failed
		raise @result
	else
		return @result
	end
end

attr :result

Access the result of the task without waiting. May be nil if the task is not completed. Does not raise exceptions.

def stop(later = false)

Stop the task and all of its children.

If later is false, it means that stop has been invoked directly. When later is true, it means that stop is invoked by stop_children or some other indirect mechanism. In that case, if we encounter the "current" fiber, we can't stop it right away, as it's currently performing #stop. Stopping it immediately would interrupt the current stop traversal, so we need to schedule the stop to occur later.

Signature

parameter later Boolean

Whether to stop the task later, or immediately.

Implementation

def stop(later = false)
	if self.stopped?
		# If the task is already stopped, a `stop` state transition re-enters the same state which is a no-op. However, we will also attempt to stop any running children too. This can happen if the children did not stop correctly the first time around. Doing this should probably be considered a bug, but it's better to be safe than sorry.
		return stopped!
	end
	
	# If we are deferring stop...
	if @defer_stop == false
		# Don't stop now... but update the state so we know we need to stop later.
		@defer_stop = true
		return false
	end
	
	# If the fiber is alive, we need to stop it:
	if @fiber&.alive?
		if self.current?
			# If the fiber is current, and later is `true`, we need to schedule the fiber to be stopped later, as it's currently invoking `stop`:
			if later
				# If the fiber is the current fiber and we want to stop it later, schedule it:
				Fiber.scheduler.push(Stop::Later.new(self))
			else
				# Otherwise, raise the exception directly:
				raise Stop, "Stopping current task!"
			end
		else
			# If the fiber is not curent, we can raise the exception directly:
			begin
				# There is a chance that this will stop the fiber that originally called stop. If that happens, the exception handling in `#stopped` will rescue the exception and re-raise it later.
				Fiber.scheduler.raise(@fiber, Stop)
			rescue FiberError
				# In some cases, this can cause a FiberError (it might be resumed already), so we schedule it to be stopped later:
				Fiber.scheduler.push(Stop::Later.new(self))
			end
		end
	else
		# We are not running, but children might be, so transition directly into stopped state:
		stop!
	end
end

def defer_stop

  • public

Defer the handling of stop. During the execution of the given block, if a stop is requested, it will be deferred until the block exits. This is useful for ensuring graceful shutdown of servers and other long-running tasks. You should wrap the response handling code in a defer_stop block to ensure that the task is stopped when the response is complete but not before.

You can nest calls to defer_stop, but the stop will only be deferred until the outermost block exits.

If stop is invoked a second time, it will be immediately executed.

Signature

yields {}

The block of code to execute.

public

Since stable-v1.

Implementation

def defer_stop
	# Tri-state variable for controlling stop:
	# - nil: defer_stop has not been called.
	# - false: defer_stop has been called and we are not stopping.
	# - true: defer_stop has been called and we will stop when exiting the block.
	if @defer_stop.nil?
		# If we are not deferring stop already, we can defer it now:
		@defer_stop = false
		
		begin
			yield
		rescue Stop
			# If we are exiting due to a stop, we shouldn't try to invoke stop again:
			@defer_stop = nil
			raise
		ensure
			# If we were asked to stop, we should do so now:
			if @defer_stop
				@defer_stop = nil
				raise Stop, "Stopping current task (was deferred)!"
			end
		end
	else
		# If we are deferring stop already, entering it again is a no-op.
		yield
	end
end

def self.current

Lookup the class Async::Task for the current fiber. Raise RuntimeError if none is available.

Signature

returns Task

Implementation

def self.current
	Thread.current[:async_task] or raise RuntimeError, "No async task available!"
end

def self.current?

Check if there is a task defined for the current fiber.

Signature

returns Task | Nil

Implementation

def self.current?
	Thread.current[:async_task]
end

def finish!

Finish the current task, moving any children to the parent.

Implementation

def finish!
	# Don't hold references to the fiber or block after the task has finished:
	@fiber = nil
	@block = nil # If some how we went directly from initialized to finished.
	
	# Attempt to remove this node from the task tree.
	consume
	
	# If this task was being used as a future, signal completion here:
	if @finished
		@finished.signal(self)
		@finished = nil
	end
end

def completed!(result)

State transition into the completed state.

Implementation

def completed!(result)
	@result = result
	@status = :completed
end

def failed!(exception = false, propagate = true)

This is a very tricky aspect of tasks to get right. I've modelled it after Thread but it's slightly different in that the exception can propagate back up through the reactor. If the user writes code which raises an exception, that exception should always be visible, i.e. cause a failure. If it's not visible, such code fails silently and can be very difficult to debug.

Implementation

def failed!(exception = false, propagate = true)
	@result = exception
	@status = :failed
	
	if exception
		if propagate
			raise exception
		elsif @finished.nil?
			# If no one has called wait, we log this as a warning:
			Console.logger.warn(self, "Task may have ended with unhandled exception.", exception)
		else
			Console.logger.debug(self, exception)
		end
	end
end

def set!

Set the current fiber's :async_task to this task.

Implementation

def set!
	# This is actually fiber-local:
	Thread.current[:async_task] = self
end

Discussion