shoryuken 2.1.1 → 3.1.3

data/lib/shoryuken/launcher.rb

@@ -1,43 +1,97 @@
 module Shoryuken
   class Launcher
-    include Celluloid
     include Util
 
-    trap_exit :actor_died
+    def initialize
+      @managers    = create_managers
+      @shutdowning = Concurrent::AtomicBoolean.new(false)
+    end
 
-    attr_accessor :manager
+    def start
+      logger.info { 'Starting' }
 
-    def initialize
-      @condvar = Celluloid::Condition.new
-      @manager = Shoryuken::Manager.new_link(@condvar)
-      @fetcher = Shoryuken::Fetcher.new_link(manager)
+      start_callback
+      start_managers
+    end
+
+    def stop!
+      initiate_stop
+
+      executor.shutdown
+
+      return if executor.wait_for_termination(Shoryuken.options[:timeout])
+
+      executor.kill
+    end
+
+    def stop
+      fire_event(:quiet, true)
+
+      initiate_stop
+
+      executor.shutdown
+      executor.wait_for_termination
+    end
 
-      @done = false
+    private
 
-      manager.fetcher = @fetcher
+    def executor
+      Concurrent.global_io_executor
     end
 
-    def stop(options = {})
-      watchdog('Launcher#stop') do
-        @done = true
-        @fetcher.terminate if @fetcher.alive?
+    def start_managers
+      @managers.each do |manager|
+        Concurrent::Promise.execute { manager.start }.rescue do |ex|
+          log_manager_failure(ex)
+          start_soft_shutdown
+        end
+      end
+    end
+
+    def start_soft_shutdown
+      Process.kill('USR1', Process.pid) if @shutdowning.make_true
+    end
+
+    def log_manager_failure(ex)
+      return unless ex
+
+      logger.error { "Manager failed: #{ex.message}" }
+      logger.error { ex.backtrace.join("\n") } unless ex.backtrace.nil?
+    end
 
-        manager.async.stop(shutdown: !!options[:shutdown], timeout: Shoryuken.options[:timeout])
-        @condvar.wait
-        manager.terminate
+    def initiate_stop
+      logger.info { 'Shutting down' }
+
+      stop_callback
+    end
+
+    def start_callback
+      if (callback = Shoryuken.start_callback)
+        logger.debug { 'Calling start_callback' }
+        callback.call
       end
+
+      fire_event(:startup)
     end
 
-    def run
-      watchdog('Launcher#run') do
-        manager.async.start
+    def stop_callback
+      if (callback = Shoryuken.stop_callback)
+        logger.debug { 'Calling stop_callback' }
+        callback.call
       end
+
+      fire_event(:shutdown, true)
     end
 
-    def actor_died(actor, reason)
-      return if @done
-      logger.warn { "Shoryuken died due to the following error, cannot recover, process exiting: #{reason}" }
-      exit 1
+    def create_managers
+      Shoryuken.groups.map do |group, options|
+        Shoryuken::Manager.new(
+          Shoryuken::Fetcher.new(group),
+          Shoryuken.polling_strategy(group).new(options[:queues]),
+          options[:concurrency],
+          executor
+        )
+      end
     end
   end
 end

data/lib/shoryuken/logging.rb

@@ -3,7 +3,6 @@ require 'logger'
 
 module Shoryuken
   module Logging
-
     class Pretty < Logger::Formatter
       # Provide a call() method that returns the formatted message.
       def call(severity, time, program_name, message)
@@ -37,9 +36,5 @@ module Shoryuken
     def self.logger=(log)
       @logger = (log ? log : Logger.new('/dev/null'))
     end
-
-    def logger
-      shoryuken::Logging.logger
-    end
   end
 end

data/lib/shoryuken/manager.rb

@@ -1,251 +1,100 @@
-require 'shoryuken/processor'
-require 'shoryuken/fetcher'
-
 module Shoryuken
   class Manager
-    include Celluloid
     include Util
 
-    attr_accessor :fetcher
-
-    trap_exit :processor_died
-
-    def initialize(condvar)
-      @count = Shoryuken.options[:concurrency] || 25
-      raise(ArgumentError, "Concurrency value #{@count} is invalid, it needs to be a positive number") unless @count > 0
-      @queues = Shoryuken.queues.dup.uniq
-      @finished = condvar
+    BATCH_LIMIT = 10
+    # See https://github.com/phstc/shoryuken/issues/348#issuecomment-292847028
+    MIN_DISPATCH_INTERVAL = 0.1
 
-      @done = false
-
-      @busy  = []
-      @ready = @count.times.map { build_processor }
-      @threads = {}
+    def initialize(fetcher, polling_strategy, concurrency, executor)
+      @fetcher          = fetcher
+      @polling_strategy = polling_strategy
+      @max_processors   = concurrency
+      @busy_processors  = Concurrent::AtomicFixnum.new(0)
+      @executor         = executor
     end
 
     def start
-      logger.info { 'Starting' }
-
-      dispatch
-    end
-
-    def stop(options = {})
-      watchdog('Manager#stop died') do
-        @done = true
-
-        if (callback = Shoryuken.stop_callback)
-          logger.info { 'Calling Shoryuken.on_stop block' }
-          callback.call
-        end
-
-        fire_event(:shutdown, true)
-
-        @fetcher.terminate if @fetcher.alive?
-
-        logger.info { "Shutting down #{@ready.size} quiet workers" }
-
-        @ready.each do |processor|
-          processor.terminate if processor.alive?
-        end
-        @ready.clear
-
-        return after(0) { @finished.signal } if @busy.empty?
-
-        if options[:shutdown]
-          hard_shutdown_in(options[:timeout])
-        else
-          soft_shutdown(options[:timeout])
-        end
-      end
-    end
-
-    def processor_done(queue, processor)
-      watchdog('Manager#processor_done died') do
-        logger.debug { "Process done for '#{queue}'" }
-
-        @threads.delete(processor.object_id)
-        @busy.delete processor
-
-        if stopped?
-          processor.terminate if processor.alive?
-          return after(0) { @finished.signal } if @busy.empty?
-        else
-          @ready << processor
-        end
-      end
-    end
-
-    def processor_died(processor, reason)
-      watchdog("Manager#processor_died died") do
-        logger.error { "Process died, reason: #{reason}" }
-
-        @threads.delete(processor.object_id)
-        @busy.delete processor
-
-        if stopped?
-          return after(0) { @finished.signal } if @busy.empty?
-        else
-          @ready << build_processor
-        end
-      end
-    end
-
-    def stopped?
-      @done
-    end
-
-    def assign(queue, sqs_msg)
-      watchdog('Manager#assign died') do
-        logger.debug { "Assigning #{sqs_msg.message_id}" }
-
-        processor = @ready.pop
-        @busy << processor
-
-        processor.async.process(queue, sqs_msg)
-      end
+      dispatch_loop
     end
 
-    def rebalance_queue_weight!(queue)
-      watchdog('Manager#rebalance_queue_weight! died') do
-        if (original = original_queue_weight(queue)) > (current = current_queue_weight(queue))
-          logger.info { "Increasing '#{queue}' weight to #{current + 1}, max: #{original}" }
+    private
 
-          @queues << queue
-        end
-      end
+    def running?
+      @executor.running?
     end
 
-    def pause_queue!(queue)
-      return if !@queues.include?(queue) || Shoryuken.options[:delay].to_f <= 0
+    def dispatch_loop
+      return unless running?
 
-      logger.debug { "Pausing '#{queue}' for #{Shoryuken.options[:delay].to_f} seconds, because it's empty" }
-
-      @queues.delete(queue)
-
-      after(Shoryuken.options[:delay].to_f) { async.restart_queue!(queue) }
+      Concurrent::Promise.execute(
+        executor: @executor
+      ) { dispatch }.then { dispatch_loop }.rescue { |ex| raise ex }
     end
 
-
     def dispatch
-      return if stopped?
-
-      logger.debug { "Ready: #{@ready.size}, Busy: #{@busy.size}, Active Queues: #{unparse_queues(@queues)}" }
+      return unless running?
 
-      if @ready.empty?
-        logger.debug { 'Pausing fetcher, because all processors are busy' }
-
-        dispatch_later
-        return
+      if ready <= 0 || (queue = @polling_strategy.next_queue).nil?
+        return sleep(MIN_DISPATCH_INTERVAL)
       end
 
-      if (queue = next_queue)
-        @fetcher.async.fetch(queue, @ready.size)
-      else
-        logger.debug { 'Pausing fetcher, because all queues are paused' }
+      fire_event(:dispatch)
 
-        @fetcher_paused = true
-      end
-    end
+      logger.debug { "Ready: #{ready}, Busy: #{busy}, Active Queues: #{@polling_strategy.active_queues}" }
 
-    def real_thread(proxy_id, thr)
-      @threads[proxy_id] = thr
+      batched_queue?(queue) ? dispatch_batch(queue) : dispatch_single_messages(queue)
     end
 
-    private
-
-    def dispatch_later
-      @_dispatch_timer ||= after(1) do
-        @_dispatch_timer = nil
-        dispatch
-      end
+    def busy
+      @busy_processors.value
     end
 
-    def build_processor
-      processor = Processor.new_link(current_actor)
-      processor.proxy_id = processor.object_id
-      processor
+    def ready
+      @max_processors - busy
     end
 
-    def restart_queue!(queue)
-      return if stopped?
-
-      unless @queues.include? queue
-        logger.debug { "Restarting '#{queue}'" }
-
-        @queues << queue
-
-        if @fetcher_paused
-          logger.debug { 'Restarting fetcher' }
+    def processor_done
+      @busy_processors.decrement
+    end
 
-          @fetcher_paused = false
+    def assign(queue_name, sqs_msg)
+      return unless running?
 
-          dispatch
-        end
-      end
-    end
+      logger.debug { "Assigning #{sqs_msg.message_id}" }
 
-    def current_queue_weight(queue)
-      queue_weight(@queues, queue)
-    end
+      @busy_processors.increment
 
-    def original_queue_weight(queue)
-      queue_weight(Shoryuken.queues, queue)
+      Concurrent::Promise.execute(
+        executor: @executor
+      ) { Processor.process(queue_name, sqs_msg) }.then { processor_done }.rescue { processor_done }
     end
 
-    def queue_weight(queues, queue)
-      queues.count { |q| q == queue }
+    def dispatch_batch(queue)
+      return if (batch = @fetcher.fetch(queue, BATCH_LIMIT)).none?
+      @polling_strategy.messages_found(queue.name, batch.size)
+      assign(queue.name, patch_batch!(batch))
     end
 
-    def next_queue
-      return nil if @queues.empty?
+    def dispatch_single_messages(queue)
+      messages = @fetcher.fetch(queue, ready)
 
-      # get/remove the first queue in the list
-      queue = @queues.shift
-
-      unless defined?(::ActiveJob) || !Shoryuken.worker_registry.workers(queue).empty?
-        # when no worker registered pause the queue to avoid endless recursion
-        logger.debug { "Pausing '#{queue}' for #{Shoryuken.options[:delay].to_f} seconds, because no workers registered" }
-
-        after(Shoryuken.options[:delay].to_f) { async.restart_queue!(queue) }
-
-        return next_queue
-      end
-
-      # add queue back to the end of the list
-      @queues << queue
-
-      queue
+      @polling_strategy.messages_found(queue.name, messages.size)
+      messages.each { |message| assign(queue.name, message) }
     end
 
-    def soft_shutdown(delay)
-      logger.info { "Waiting for #{@busy.size} busy workers" }
-
-      if @busy.size > 0
-        after(delay) { soft_shutdown(delay) }
-      else
-        @finished.signal
-      end
+    def batched_queue?(queue)
+      Shoryuken.worker_registry.batch_receive_messages?(queue.name)
     end
 
-    def hard_shutdown_in(delay)
-      logger.info { "Waiting for #{@busy.size} busy workers" }
-      logger.info { "Pausing up to #{delay} seconds to allow workers to finish..." }
-
-      after(delay) do
-        watchdog('Manager#hard_shutdown_in died') do
-          if @busy.size > 0
-            logger.info { "Hard shutting down #{@busy.size} busy workers" }
-
-            @busy.each do |processor|
-              if processor.alive? && t = @threads.delete(processor.object_id)
-                t.raise Shutdown
-              end
-            end
-          end
-
-          @finished.signal
+    def patch_batch!(sqs_msgs)
+      sqs_msgs.instance_eval do
+        def message_id
+          "batch-with-#{size}-messages"
         end
       end
+
+      sqs_msgs
     end
   end
 end

data/lib/shoryuken/message.rb

@@ -3,19 +3,10 @@ module Shoryuken
     attr_accessor :client, :queue_url, :queue_name, :data
 
     def initialize(client, queue, data)
-      self.client = client
-      self.data = data
-
-      if queue.is_a?(Shoryuken::Queue)
-        self.queue_url = queue.url
-        self.queue_name = queue.name
-      else
-        # TODO: Remove next major release
-        Shoryuken.logger.warn do
-          '[DEPRECATION] Passing a queue url into Shoryuken::Message is deprecated, please pass the queue itself'
-        end
-        self.queue_url = queue
-      end
+      self.client     = client
+      self.data       = data
+      self.queue_url  = queue.url
+      self.queue_name = queue.name
     end
 
     def delete

data/lib/shoryuken/middleware/chain.rb

@@ -102,32 +102,15 @@ module Shoryuken
 
     class Entry
       attr_reader :klass
+
       def initialize(klass, *args)
         @klass = klass
         @args  = args
-
-        patch_deprecated_middleware!(klass)
       end
 
       def make_new
         @klass.new(*@args)
       end
-
-      private
-
-      def patch_deprecated_middleware!(klass)
-        if klass.instance_method(:call).arity == 3
-          Shoryuken.logger.warn { "[DEPRECATION] #{klass.name}#call(worker_instance, queue, sqs_msg) is deprecated. Please use #{klass.name}#call(worker_instance, queue, sqs_msg, body)" }
-
-          klass.class_eval do
-            alias_method :deprecated_call, :call
-
-            def call(worker_instance, queue, sqs_msg, body = nil, &block)
-              deprecated_call(worker_instance, queue, sqs_msg, &block)
-            end
-          end
-        end
-      end
     end
   end
 end

data/lib/shoryuken/middleware/server/auto_delete.rb

@@ -5,18 +5,13 @@ module Shoryuken
         def call(worker, queue, sqs_msg, body)
           yield
 
-          auto_delete = worker.class.get_shoryuken_options['delete'] || worker.class.get_shoryuken_options['auto_delete']
+          return unless worker.class.auto_delete?
 
-          if auto_delete
-            entries = [sqs_msg].flatten.map.with_index do |message, i|
-              { id: i.to_s, receipt_handle: message.receipt_handle }
-            end
+          entries = [sqs_msg].flatten.map.with_index { |message, i| { id: i.to_s, receipt_handle: message.receipt_handle } }
 
-            Shoryuken::Client.queues(queue).delete_messages(entries: entries)
-          end
+          Shoryuken::Client.queues(queue).delete_messages(entries: entries)
         end
       end
     end
   end
 end
-

data/lib/shoryuken/middleware/server/auto_extend_visibility.rb

@@ -1,33 +1,34 @@
-require 'celluloid/current' unless defined?(Celluloid)
-
 module Shoryuken
   module Middleware
     module Server
       class AutoExtendVisibility
+        include Util
+
         EXTEND_UPFRONT_SECONDS = 5
 
         def call(worker, queue, sqs_msg, body)
-          timer = auto_visibility_timer(worker, queue, sqs_msg, body)
-          begin
-            yield
-          ensure
-            if timer
-              timer.cancel
-              @visibility_extender.terminate
-            end
+          return yield unless worker.class.auto_visibility_timeout?
+
+          if sqs_msg.is_a?(Array)
+            logger.warn { "Auto extend visibility isn't supported for batch workers" }
+            return yield
           end
+
+          timer = auto_visibility_timer(worker, queue, sqs_msg, body)
+          yield
+        ensure
+          timer.kill if timer
         end
 
         private
 
         class MessageVisibilityExtender
-          include Celluloid
           include Util
 
           def auto_extend(worker, queue, sqs_msg, body)
             queue_visibility_timeout = Shoryuken::Client.queues(queue).visibility_timeout
 
-            every(queue_visibility_timeout - EXTEND_UPFRONT_SECONDS) do
+            Concurrent::TimerTask.new(execution_interval: queue_visibility_timeout - EXTEND_UPFRONT_SECONDS) do
               begin
                 logger.debug do
                   "Extending message #{worker_name(worker.class, sqs_msg, body)}/#{queue}/#{sqs_msg.message_id} " \
@@ -35,11 +36,11 @@ module Shoryuken
                 end
 
                 sqs_msg.change_visibility(visibility_timeout: queue_visibility_timeout)
-              rescue => e
+              rescue => ex
                 logger.error do
                   'Could not auto extend the message ' \
                   "#{worker_name(worker.class, sqs_msg, body)}/#{queue}/#{sqs_msg.message_id} " \
-                  "visibility timeout. Error: #{e.message}"
+                  "visibility timeout. Error: #{ex.message}"
                 end
               end
             end
@@ -47,9 +48,7 @@ module Shoryuken
         end
 
         def auto_visibility_timer(worker, queue, sqs_msg, body)
-          return unless worker.class.auto_visibility_timeout?
-          @visibility_extender = MessageVisibilityExtender.new_link
-          @visibility_extender.auto_extend(worker, queue, sqs_msg, body)
+          MessageVisibilityExtender.new.auto_extend(worker, queue, sqs_msg, body).tap(&:execute)
         end
       end
     end

data/lib/shoryuken/middleware/server/exponential_backoff_retry.rb

@@ -5,42 +5,57 @@ module Shoryuken
         include Util
 
         def call(worker, queue, sqs_msg, body)
+          return yield unless worker.class.exponential_backoff?
+
+          if sqs_msg.is_a?(Array)
+            logger.warn { "Exponential backoff isn't supported for batch workers" }
+            return yield
+          end
+
           started_at = Time.now
           yield
-        rescue
-          retry_intervals = Array(worker.class.get_shoryuken_options['retry_intervals'])
+        rescue => ex
+          retry_intervals = worker.class.get_shoryuken_options['retry_intervals']
 
-          if retry_intervals.empty? || !handle_failure(sqs_msg, started_at, retry_intervals)
+          if retry_intervals.nil? || !handle_failure(sqs_msg, started_at, retry_intervals)
             # Re-raise the exception if the job is not going to be exponential backoff retried.
             # This allows custom middleware (like exception notifiers) to be aware of the unhandled failure.
             raise
           end
+
+          logger.warn { "Message #{sqs_msg.message_id} will attempt retry due to error: #{ex.message}" }
+          # since we didn't raise, lets log the backtrace for debugging purposes.
+          logger.debug { ex.backtrace.join("\n") } unless ex.backtrace.nil?
         end
 
         private
 
-        def handle_failure(sqs_msg, started_at, retry_intervals)
-          return unless attempts = sqs_msg.attributes['ApproximateReceiveCount']
-
-          attempts = attempts.to_i - 1
-
-          interval = if attempts < retry_intervals.size
-                       retry_intervals[attempts]
-                     else
-                       retry_intervals.last
-                     end
-
-          # Visibility timeouts are limited to a total 12 hours, starting from the receipt of the message.
-          # We calculate the maximum timeout by subtracting the amount of time since the receipt of the message.
-          #
-          # From the docs:  "Amazon SQS restarts the timeout period using the new value."
-          # http://docs.aws.amazon.com/AWSSimpleQueueService/latest/SQSDeveloperGuide/AboutVT.html#AboutVT-extending-message-visibility-timeout
-          max_timeout = 43200 - (Time.now - started_at).ceil - 1
+        def get_interval(retry_intervals, attempts)
+          return retry_intervals.call(attempts) if retry_intervals.respond_to?(:call)
+
+          if attempts <= (retry_intervals = Array(retry_intervals)).size
+            retry_intervals[attempts - 1]
+          else
+            retry_intervals.last
+          end
+        end
+
+        def next_visibility_timeout(interval, started_at)
+          max_timeout = 43_200 - (Time.now - started_at).ceil - 1
           interval = max_timeout if interval > max_timeout
+          interval.to_i
+        end
+
+        def handle_failure(sqs_msg, started_at, retry_intervals)
+          receive_count = sqs_msg.attributes['ApproximateReceiveCount'].to_i
+
+          return false unless (interval = get_interval(retry_intervals, receive_count))
+
+          sqs_msg.change_visibility(visibility_timeout: next_visibility_timeout(interval.to_i, started_at))
 
-          sqs_msg.change_visibility(visibility_timeout: interval.to_i)
+          logger.info { "Message #{sqs_msg.message_id} failed, will be retried in #{interval} seconds" }
 
-          logger.info { "Message #{sqs_msg.message_id} failed, will be retried in #{interval} seconds." }
+          true
         end
       end
     end