async 2.17.0 → 2.32.0

data/lib/async/scheduler.rb

@@ -1,21 +1,35 @@
 # frozen_string_literal: true
 
 # Released under the MIT License.
-# Copyright, 2020-2024, by Samuel Williams.
+# Copyright, 2020-2025, by Samuel Williams.
 # Copyright, 2020, by Jun Jiang.
 # Copyright, 2021, by Julien Portalier.
+# Copyright, 2025, by Shopify Inc.
 
-require_relative 'clock'
-require_relative 'task'
+require_relative "clock"
+require_relative "task"
+require_relative "timeout"
 
-require 'io/event'
+require "io/event"
 
-require 'console'
-require 'resolv'
+require "console"
+require "resolv"
 
 module Async
+	begin
+		require "fiber/profiler"
+		Profiler = Fiber::Profiler
+	rescue LoadError
+		# Fiber::Profiler is not available.
+		Profiler = nil
+	end
+	
 	# Handles scheduling of fibers. Implements the fiber scheduler interface.
 	class Scheduler < Node
+		WORKER_POOL = ENV.fetch("ASYNC_SCHEDULER_WORKER_POOL", nil).then do |value|
+			value == "true" ? true : nil
+		end
+		
 		# Raised when an operation is attempted on a closed scheduler.
 		class ClosedError < RuntimeError
 			# Create a new error.
@@ -27,20 +41,44 @@ module Async
 		end
 		
 		# Whether the fiber scheduler is supported.
-		# @public Since `stable-v1`.
+		# @public Since *Async v1*.
 		def self.supported?
 			true
 		end
 		
+		# Used to augment the scheduler to add support for blocking operations.
+		module BlockingOperationWait
+			# Wait for the given work to be executed.
+			#
+			# @public Since *Async v2.21* and *Ruby v3.4*.
+			# @asynchronous May be non-blocking.
+			#
+			# @parameter work [Proc] The work to execute on a background thread.
+			# @returns [Object] The result of the work.
+			def blocking_operation_wait(work)
+				@worker_pool.call(work)
+			end
+		end
+		
+		private_constant :BlockingOperationWait
+		
+		if ::IO::Event.const_defined?(:WorkerPool)
+			WorkerPool = ::IO::Event::WorkerPool
+		else
+			WorkerPool = nil
+		end
+		
 		# Create a new scheduler.
 		#
-		# @public Since `stable-v1`.
+		# @public Since *Async v1*.
 		# @parameter parent [Node | Nil] The parent node to use for task hierarchy.
 		# @parameter selector [IO::Event::Selector] The selector to use for event handling.
-		def initialize(parent = nil, selector: nil)
+		def initialize(parent = nil, selector: nil, profiler: Profiler&.default, worker_pool: WORKER_POOL)
 			super(parent)
 			
 			@selector = selector || ::IO::Event::Selector.new(Fiber.current)
+			@profiler = profiler
+			
 			@interrupted = false
 			
 			@blocked = 0
@@ -49,9 +87,20 @@ module Async
 			@idle_time = 0.0
 			
 			@timers = ::IO::Event::Timers.new
+			
+			if worker_pool == true
+				@worker_pool = WorkerPool&.new
+			else
+				@worker_pool = worker_pool
+			end
+			
+			if @worker_pool
+				self.singleton_class.prepend(BlockingOperationWait)
+			end
 		end
 		
 		# Compute the scheduler load according to the busy and idle times that are updated by the run loop.
+		#
 		# @returns [Float] The load of the scheduler. 0.0 means no load, 1.0 means fully loaded or over-loaded.
 		def load
 			total_time = @busy_time + @idle_time
@@ -71,7 +120,7 @@ module Async
 				return @busy_time / total_time
 			end
 		end
-	
+		
 		# Invoked when the fiber scheduler is being closed.
 		#
 		# Executes the run loop until all tasks are finished, then closes the scheduler.
@@ -95,7 +144,7 @@ module Async
 		end
 		
 		# Terminate all child tasks and close the scheduler.
-		# @public Since `stable-v1`.
+		# @public Since *Async v1*.
 		def close
 			self.run_loop do
 				until self.terminate
@@ -111,11 +160,16 @@ module Async
 			
 			selector&.close
 			
+			worker_pool = @worker_pool
+			@worker_pool = nil
+			
+			worker_pool&.close
+			
 			consume
 		end
 		
 		# @returns [Boolean] Whether the scheduler has been closed.
-		# @public Since `stable-v1`.
+		# @public Since *Async v1*.
 		def closed?
 			@selector.nil?
 		end
@@ -167,7 +221,12 @@ module Async
 		end
 		
 		# Invoked when a fiber tries to perform a blocking operation which cannot continue. A corresponding call {unblock} must be performed to allow this fiber to continue.
+		#
+		# @public Since *Async v2*.
 		# @asynchronous May only be called on same thread as fiber scheduler.
+		#
+		# @parameter blocker [Object] The object that is blocking the fiber.
+		# @parameter timeout [Float | Nil] The maximum time to block, or if nil, indefinitely.
 		def block(blocker, timeout)
 			# $stderr.puts "block(#{blocker}, #{Fiber.current}, #{timeout})"
 			fiber = Fiber.current
@@ -190,9 +249,15 @@ module Async
 			timer&.cancel!
 		end
 		
+		# Unblock a fiber that was previously blocked.
+		#
+		# @public Since *Async v2* and *Ruby v3.1*.
 		# @asynchronous May be called from any thread.
+		#
+		# @parameter blocker [Object] The object that was blocking the fiber.
+		# @parameter fiber [Fiber] The fiber to unblock.
 		def unblock(blocker, fiber)
-			# $stderr.puts "unblock(#{blocker}, #{fiber})"
+			# Fiber.blocking{$stderr.puts "unblock(#{blocker}, #{fiber})"}
 			
 			# This operation is protected by the GVL:
 			if selector = @selector
@@ -201,8 +266,15 @@ module Async
 			end
 		end
 		
-		# @asynchronous May be non-blocking..
+		# Sleep for the specified duration.
+		#
+		# @public Since *Async v2* and *Ruby v3.1*.
+		# @asynchronous May be non-blocking.
+		#
+		# @parameter duration [Numeric | Nil] The time in seconds to sleep, or if nil, indefinitely.
 		def kernel_sleep(duration = nil)
+			# Fiber.blocking{$stderr.puts "kernel_sleep(#{duration}, #{Fiber.current})"}
+			
 			if duration
 				self.block(nil, duration)
 			else
@@ -210,7 +282,12 @@ module Async
 			end
 		end
 		
-		# @asynchronous May be non-blocking..
+		# Resolve the address of the given hostname.
+		#
+		# @public Since *Async v2*.
+		# @asynchronous May be non-blocking.
+		#
+		# @parameter hostname [String] The hostname to resolve.
 		def address_resolve(hostname)
 			# On some platforms, hostnames may contain a device-specific suffix (e.g. %en0). We need to strip this before resolving.
 			# See <https://github.com/socketry/async/issues/180> for more details.
@@ -218,18 +295,14 @@ module Async
 			::Resolv.getaddresses(hostname)
 		end
 		
-		
-		if IO.method_defined?(:timeout)
-			private def get_timeout(io)
-				io.timeout
-			end
-		else
-			private def get_timeout(io)
-				nil
-			end
-		end
-		
-		# @asynchronous May be non-blocking..
+		# Wait for the specified IO to become ready for the specified events.
+		#
+		# @public Since *Async v2*.
+		# @asynchronous May be non-blocking.
+		#
+		# @parameter io [IO] The IO object to wait on.
+		# @parameter events [Integer] The events to wait for, e.g. `IO::READABLE`, `IO::WRITABLE`, etc.
+		# @parameter timeout [Float | Nil] The maximum time to wait, or if nil, indefinitely.
 		def io_wait(io, events, timeout = nil)
 			fiber = Fiber.current
 			
@@ -238,7 +311,7 @@ module Async
 				timer = @timers.after(timeout) do
 					fiber.transfer
 				end
-			elsif timeout = get_timeout(io)
+			elsif timeout = io.timeout
 				# Otherwise, if we default to the io's timeout, we raise an exception:
 				timer = @timers.after(timeout) do
 					fiber.raise(::IO::TimeoutError, "Timeout (#{timeout}s) while waiting for IO to become ready!")
@@ -251,10 +324,19 @@ module Async
 		end
 		
 		if ::IO::Event::Support.buffer?
+			# Read from the specified IO into the buffer.
+			#
+			# @public Since *Async v2* and Ruby with `IO::Buffer` support.
+			# @asynchronous May be non-blocking.
+			#
+			# @parameter io [IO] The IO object to read from.
+			# @parameter buffer [IO::Buffer] The buffer to read into.
+			# @parameter length [Integer] The minimum number of bytes to read.
+			# @parameter offset [Integer] The offset within the buffer to read into.
 			def io_read(io, buffer, length, offset = 0)
 				fiber = Fiber.current
 				
-				if timeout = get_timeout(io)
+				if timeout = io.timeout
 					timer = @timers.after(timeout) do
 						fiber.raise(::IO::TimeoutError, "Timeout (#{timeout}s) while waiting for IO to become readable!")
 					end
@@ -266,10 +348,19 @@ module Async
 			end
 			
 			if RUBY_ENGINE != "ruby" || RUBY_VERSION >= "3.3.1"
+				# Write the specified buffer to the IO.
+				#
+				# @public Since *Async v2* and *Ruby v3.3.1* with `IO::Buffer` support.
+				# @asynchronous May be non-blocking.
+				#
+				# @parameter io [IO] The IO object to write to.
+				# @parameter buffer [IO::Buffer] The buffer to write from.
+				# @parameter length [Integer] The minimum number of bytes to write.
+				# @parameter offset [Integer] The offset within the buffer to write from.
 				def io_write(io, buffer, length, offset = 0)
 					fiber = Fiber.current
 					
-					if timeout = get_timeout(io)
+					if timeout = io.timeout
 						timer = @timers.after(timeout) do
 							fiber.raise(::IO::TimeoutError, "Timeout (#{timeout}s) while waiting for IO to become writable!")
 						end
@@ -282,7 +373,39 @@ module Async
 			end
 		end
 		
+		# Used to defer stopping the current task until later.
+		class FiberInterrupt
+			# Create a new stop later operation.
+			#
+			# @parameter task [Task] The task to stop later.
+			def initialize(fiber, exception)
+				@fiber = fiber
+				@exception = exception
+			end
+			
+			# @returns [Boolean] Whether the task is alive.
+			def alive?
+				@fiber.alive?
+			end
+			
+			# Transfer control to the operation - this will stop the task.
+			def transfer
+				# Fiber.blocking{$stderr.puts "FiberInterrupt#transfer(#{@fiber}, #{@exception})"}
+				@fiber.raise(@exception)
+			end
+		end
+		
+		# Raise an exception on the specified fiber, waking up the event loop if necessary.
+		def fiber_interrupt(fiber, exception)
+			# Fiber.blocking{$stderr.puts "fiber_interrupt(#{fiber}, #{exception})"}
+			unblock(nil, FiberInterrupt.new(fiber, exception))
+		end
+		
 		# Wait for the specified process ID to exit.
+		#
+		# @public Since *Async v2*.
+		# @asynchronous May be non-blocking.
+		#
 		# @parameter pid [Integer] The process ID to wait for.
 		# @parameter flags [Integer] A bit-mask of flags suitable for `Process::Status.wait`.
 		# @returns [Process::Status] A process status instance.
@@ -291,6 +414,19 @@ module Async
 			return @selector.process_wait(Fiber.current, pid, flags)
 		end
 		
+		# Wait for the specified IOs to become ready for the specified events.
+		#
+		# @public Since *Async v2.25*.
+		# @asynchronous May be non-blocking.
+		def io_select(...)
+			Thread.new do
+				# Don't make unnecessary output, since we will propagate the exception:
+				Thread.current.report_on_exception = false
+				
+				::IO.select(...)
+			end.value
+		end
+		
 		# Run one iteration of the event loop.
 		#
 		# When terminating the event loop, we already know we are finished. So we don't need to check the task tree. This is a logical requirement because `run_once` ignores transient tasks. For example, a single top level transient task is not enough to keep the reactor running, but during termination we must still process it in order to terminate child tasks.
@@ -335,7 +471,10 @@ module Async
 		end
 		
 		# Run one iteration of the event loop.
-		# Does not handle interrupts.
+		#
+		# @public Since *Async v1*.
+		# @asynchronous Must be invoked from blocking (root) fiber.
+		#
 		# @parameter timeout [Float | Nil] The maximum timeout, or if nil, indefinite.
 		# @returns [Boolean] Whether there is more work to do.
 		def run_once(timeout = nil)
@@ -354,6 +493,7 @@ module Async
 		end
 		
 		# Checks and clears the interrupted state of the scheduler.
+		#
 		# @returns [Boolean] Whether the reactor has been interrupted.
 		private def interrupted?
 			if @interrupted
@@ -368,7 +508,9 @@ module Async
 			return false
 		end
 		
-		# Stop all children, including transient children, ignoring any signals.
+		# Stop all children, including transient children.
+		#
+		# @public Since *Async v1*.
 		def stop
 			@children&.each do |child|
 				child.stop
@@ -387,7 +529,13 @@ module Async
 					end
 				end
 			rescue Interrupt => interrupt
+				# If an interrupt did occur during an iteration of the event loop, we need to handle it. More specifically, `self.stop` is not safe to interrupt without potentially corrupting the task tree.
 				Thread.handle_interrupt(::SignalException => :never) do
+					Console.debug(self) do |buffer|
+						buffer.puts "Scheduler interrupted: #{interrupt.inspect}"
+						self.print_hierarchy(buffer)
+					end
+					
 					self.stop
 				end
 				
@@ -395,55 +543,73 @@ module Async
 			end
 			
 			# If the event loop was interrupted, and we finished exiting normally (due to the interrupt), we need to re-raise the interrupt so that the caller can handle it too.
-			Kernel.raise(interrupt) if interrupt
+			if interrupt
+				Kernel.raise(interrupt)
+			end
 		end
 		
 		# Run the reactor until all tasks are finished. Proxies arguments to {#async} immediately before entering the loop, if a block is provided.
+		#
+		# Forwards all parameters to {#async} if a block is given.
+		#
+		# @public Since *Async v1*.
+		#
+		# @yields {|task| ...} The top level task, if a block is given.
+		# @returns [Task] The initial task that was scheduled into the reactor.
 		def run(...)
 			Kernel.raise ClosedError if @selector.nil?
 			
-			initial_task = self.async(...) if block_given?
-			
-			self.run_loop do
-				run_once
+			begin
+				@profiler&.start
+				
+				initial_task = self.async(...) if block_given?
+				
+				self.run_loop do
+					run_once
+				end
+				
+				return initial_task
+			ensure
+				@profiler&.stop
 			end
-			
-			return initial_task
 		end
 		
-		# Start an asynchronous task within the specified reactor. The task will be
-		# executed until the first blocking call, at which point it will yield and
-		# and this method will return.
+		# Start an asynchronous task within the specified reactor. The task will be executed until the first blocking call, at which point it will yield and and this method will return.
 		#
-		# This is the main entry point for scheduling asynchronus tasks.
+		# @public Since *Async v1*.
+		# @asynchronous May context switch immediately to new task.
+		# @deprecated Use {#run} or {Task#async} instead.
 		#
 		# @yields {|task| ...} Executed within the task.
 		# @returns [Task] The task that was scheduled into the reactor.
-		# @deprecated With no replacement.
 		def async(*arguments, **options, &block)
+			# Since this method is called by `run`, this warning is too excessive:
+			# warn("Async::Scheduler#async is deprecated. Use `run` or `Task#async` instead.", uplevel: 1, category: :deprecated) if $VERBOSE
+			
 			Kernel.raise ClosedError if @selector.nil?
 			
 			task = Task.new(Task.current? || self, **options, &block)
 			
-			# I want to take a moment to explain the logic of this.
-			# When calling an async block, we deterministically execute it until the
-			# first blocking operation. We don't *have* to do this - we could schedule
-			# it for later execution, but it's useful to:
-			# - Fail at the point of the method call where possible.
-			# - Execute determinstically where possible.
-			# - Avoid scheduler overhead if no blocking operation is performed.
 			task.run(*arguments)
 			
-			# Console.debug "Initial execution of task #{fiber} complete (#{result} -> #{fiber.alive?})..."
 			return task
 		end
 		
+		# Create a new fiber and return it without starting execution.
+		# @returns [Fiber] The fiber that was created.
 		def fiber(...)
 			return async(...).fiber
 		end
 		
 		# Invoke the block, but after the specified timeout, raise {TimeoutError} in any currenly blocking operation. If the block runs to completion before the timeout occurs or there are no non-blocking operations after the timeout expires, the code will complete without any exception.
+		#
+		# @public Since *Async v1*.
+		# @asynchronous May raise an exception at any interruption point (e.g. blocking operations).
+		#
 		# @parameter duration [Numeric] The time in seconds, in which the task should complete.
+		# @parameter exception [Class] The exception class to raise.
+		# @parameter message [String] The message to pass to the exception.
+		# @yields {|timeout| ...} The block to execute with a timeout.
 		def with_timeout(duration, exception = TimeoutError, message = "execution expired", &block)
 			fiber = Fiber.current
 			
@@ -453,13 +619,26 @@ module Async
 				end
 			end
 			
-			yield timer
+			if block.arity.zero?
+				yield
+			else
+				yield Timeout.new(@timers, timer)
+			end
 		ensure
 			timer&.cancel!
 		end
 		
+		# Invoke the block, but after the specified timeout, raise the specified exception with the given message. If the block runs to completion before the timeout occurs or there are no non-blocking operations after the timeout expires, the code will complete without any exception.
+		#
+		# @public Since *Async v1* and *Ruby v3.1*. May be invoked from `Timeout.timeout`.
+		# @asynchronous May raise an exception at any interruption point (e.g. blocking operations).
+		#
+		# @parameter duration [Numeric] The time in seconds, in which the task should complete.
+		# @parameter exception [Class] The exception class to raise.
+		# @parameter message [String] The message to pass to the exception.
+		# @yields {|duration| ...} The block to execute with a timeout.
 		def timeout_after(duration, exception, message, &block)
-			with_timeout(duration, exception, message) do |timer|
+			with_timeout(duration, exception, message) do
 				yield duration
 			end
 		end

data/lib/async/semaphore.rb

@@ -1,13 +1,13 @@
 # frozen_string_literal: true
 
 # Released under the MIT License.
-# Copyright, 2018-2023, by Samuel Williams.
+# Copyright, 2018-2024, by Samuel Williams.
 
-require_relative 'list'
+require_relative "list"
 
 module Async
 	# A synchronization primitive, which limits access to a given resource.
-	# @public Since `stable-v1`.
+	# @public Since *Async v1*.
 	class Semaphore
 		# @parameter limit [Integer] The maximum number of times the semaphore can be acquired before it blocks.
 		# @parameter parent [Task | Semaphore | Nil] The parent for holding any children tasks.

data/lib/async/stop.rb

@@ -0,0 +1,82 @@
+# frozen_string_literal: true
+
+# Released under the MIT License.
+# Copyright, 2025, by Samuel Williams.
+
+require "fiber"
+require "console"
+
+module Async
+	# Raised when a task is explicitly stopped.
+	class Stop < Exception
+		# Represents the source of the stop operation.
+		class Cause < Exception
+			if RUBY_VERSION >= "3.4"
+				# @returns [Array(Thread::Backtrace::Location)] The backtrace of the caller.
+				def self.backtrace
+					caller_locations(2..-1)
+				end
+			else
+				# @returns [Array(String)] The backtrace of the caller.
+				def self.backtrace
+					caller(2..-1)
+				end
+			end
+			
+			# Create a new cause of the stop operation, with the given message.
+			#
+			# @parameter message [String] The error message.
+			# @returns [Cause] The cause of the stop operation.
+			def self.for(message = "Task was stopped")
+				instance = self.new(message)
+				instance.set_backtrace(self.backtrace)
+				return instance
+			end
+		end
+		
+		if RUBY_VERSION < "3.5"
+			# Create a new stop operation.
+			#
+			# This is a compatibility method for Ruby versions before 3.5 where cause is not propagated correctly when using {Fiber#raise}
+			#
+			# @parameter message [String | Hash] The error message or a hash containing the cause.
+			def initialize(message = "Task was stopped")
+				if message.is_a?(Hash)
+					@cause = message[:cause]
+					message = "Task was stopped"
+				end
+				
+				super(message)
+			end
+			
+			# @returns [Exception] The cause of the stop operation.
+			#
+			# This is a compatibility method for Ruby versions before 3.5 where cause is not propagated correctly when using {Fiber#raise}, we explicitly capture the cause here.
+			def cause
+				super || @cause
+			end
+		end
+		
+		# Used to defer stopping the current task until later.
+		class Later
+			# Create a new stop later operation.
+			#
+			# @parameter task [Task] The task to stop later.
+			# @parameter cause [Exception] The cause of the stop operation.
+			def initialize(task, cause = nil)
+				@task = task
+				@cause = cause
+			end
+			
+			# @returns [Boolean] Whether the task is alive.
+			def alive?
+				true
+			end
+			
+			# Transfer control to the operation - this will stop the task.
+			def transfer
+				@task.stop(false, cause: @cause)
+			end
+		end
+	end
+end