karafka 2.0.0.alpha5 → 2.0.0.beta2

data/lib/karafka/messages/builders/batch_metadata.rb

@@ -8,42 +8,30 @@ module Karafka
         class << self
           # Creates metadata based on the kafka batch data.
           #
-          # @param kafka_batch [Array<Rdkafka::Consumer::Message>] raw fetched messages
+          # @param messages [Array<Karafka::Messages::Message>] messages array
           # @param topic [Karafka::Routing::Topic] topic for which we've fetched the batch
           # @param scheduled_at [Time] moment when the batch was scheduled for processing
           # @return [Karafka::Messages::BatchMetadata] batch metadata object
-          # @note Regarding the time lags: we can use the current time here, as batch metadata is
-          #   created in the worker. So whenever this is being built, it means that the processing
-          #   of this batch has already started.
-          def call(kafka_batch, topic, scheduled_at)
-            now = Time.now
-
+          #
+          # @note We do not set `processed_at` as this needs to be assigned when the batch is
+          #   picked up for processing.
+          def call(messages, topic, scheduled_at)
             Karafka::Messages::BatchMetadata.new(
-              size: kafka_batch.count,
-              first_offset: kafka_batch.first.offset,
-              last_offset: kafka_batch.last.offset,
+              size: messages.count,
+              first_offset: messages.first.offset,
+              last_offset: messages.last.offset,
               deserializer: topic.deserializer,
-              partition: kafka_batch[0].partition,
+              partition: messages.first.partition,
               topic: topic.name,
+              # We go with the assumption that the creation of the whole batch is the last message
+              # creation time
+              created_at: messages.last.timestamp,
+              # When this batch was built and scheduled for execution
               scheduled_at: scheduled_at,
-              # This lag describes how long did it take for a message to be consumed from the
-              # moment it was created
-              consumption_lag: time_distance_in_ms(now, kafka_batch.last.timestamp),
-              # This lag describes how long did a batch have to wait before it was picked up by
-              # one of the workers
-              processing_lag: time_distance_in_ms(now, scheduled_at)
-            ).freeze
-          end
-
-          private
-
-          # Computes time distance in between two times in ms
-          #
-          # @param time1 [Time]
-          # @param time2 [Time]
-          # @return [Integer] distance in between two times in ms
-          def time_distance_in_ms(time1, time2)
-            ((time1 - time2) * 1_000).round
+              # We build the batch metadata when we pick up the job in the worker, thus we can use
+              # current time here
+              processed_at: Time.now
+            )
           end
         end
       end

data/lib/karafka/messages/builders/message.rb

@@ -26,6 +26,7 @@ module Karafka
               received_at: received_at
             ).freeze
 
+            # Karafka messages cannot be frozen because of the lazy deserialization feature
             Karafka::Messages::Message.new(
               kafka_message.payload,
               metadata

data/lib/karafka/messages/builders/messages.rb

@@ -9,27 +9,19 @@ module Karafka
           # Creates messages batch with messages inside based on the incoming messages and the
           # topic from which it comes.
           #
-          # @param kafka_messages [Array<Rdkafka::Consumer::Message>] raw fetched messages
+          # @param messages [Array<Karafka::Messages::Message>] karafka messages array
           # @param topic [Karafka::Routing::Topic] topic for which we're received messages
           # @param received_at [Time] moment in time when the messages were received
           # @return [Karafka::Messages::Messages] messages batch object
-          def call(kafka_messages, topic, received_at)
-            messages_array = kafka_messages.map do |message|
-              Karafka::Messages::Builders::Message.call(
-                message,
-                topic,
-                received_at
-              )
-            end
-
+          def call(messages, topic, received_at)
             metadata = BatchMetadata.call(
-              kafka_messages,
+              messages,
               topic,
               received_at
             ).freeze
 
             Karafka::Messages::Messages.new(
-              messages_array,
+              messages,
               metadata
             ).freeze
           end

data/lib/karafka/pro/active_job/consumer.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Pro
+    module ActiveJob
+      # This Karafka component is a Pro component.
+      # All of the commercial components are present in the lib/karafka/pro directory of this
+      # repository and their usage requires commercial license agreement.
+      #
+      # Karafka has also commercial-friendly license, commercial support and commercial components.
+      #
+      # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+      # of your code to Maciej Mensfeld.
+
+      # Pro ActiveJob consumer that is suppose to handle long-running jobs as well as short
+      # running jobs
+      class Consumer < Karafka::ActiveJob::Consumer
+      end
+    end
+  end
+end

data/lib/karafka/pro/active_job/dispatcher.rb

@@ -1,24 +1,24 @@
 # frozen_string_literal: true
 
-# This Karafka component is a Pro component.
-# All of the commercial components are present in the lib/karafka/pro directory of this repository
-# and their usage requires commercial license agreement.
-#
-# Karafka has also commercial-friendly license, commercial support and commercial components.
-#
-# By sending a pull request to the pro components, you are agreeing to transfer the copyright of
-# your code to Maciej Mensfeld.
-
 module Karafka
   module Pro
     # Karafka Pro ActiveJob components
     module ActiveJob
+      # This Karafka component is a Pro component.
+      # All of the commercial components are present in the lib/karafka/pro directory of this
+      # repository and their usage requires commercial license agreement.
+      #
+      # Karafka has also commercial-friendly license, commercial support and commercial components.
+      #
+      # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+      # of your code to Maciej Mensfeld.
+
       # Pro dispatcher that sends the ActiveJob job to a proper topic based on the queue name
       # and that allows to inject additional options into the producer, effectively allowing for a
       # much better and more granular control over the dispatch and consumption process.
       class Dispatcher < ::Karafka::ActiveJob::Dispatcher
         # Defaults for dispatching
-        # The can be updated by using `#karafka_options` on the job
+        # They can be updated by using `#karafka_options` on the job
         DEFAULTS = {
           dispatch_method: :produce_async,
           # We don't create a dummy proc based partitioner as we would have to evaluate it with

data/lib/karafka/pro/active_job/job_options_contract.rb

@@ -1,17 +1,17 @@
 # frozen_string_literal: true
 
-# This Karafka component is a Pro component.
-# All of the commercial components are present in the lib/karafka/pro directory of this repository
-# and their usage requires commercial license agreement.
-#
-# Karafka has also commercial-friendly license, commercial support and commercial components.
-#
-# By sending a pull request to the pro components, you are agreeing to transfer the copyright of
-# your code to Maciej Mensfeld.
-
 module Karafka
   module Pro
     module ActiveJob
+      # This Karafka component is a Pro component.
+      # All of the commercial components are present in the lib/karafka/pro directory of this
+      # repository and their usage requires commercial license agreement.
+      #
+      # Karafka has also commercial-friendly license, commercial support and commercial components.
+      #
+      # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+      # of your code to Maciej Mensfeld.
+
       # Contract for validating the options that can be altered with `#karafka_options` per job
       # class that works with Pro features.
       class JobOptionsContract < ::Karafka::ActiveJob::JobOptionsContract

data/lib/karafka/pro/loader.rb

@@ -1,15 +1,16 @@
 # frozen_string_literal: true
 
-# This Karafka component is a Pro component.
-# All of the commercial components are present in the lib/karafka/pro directory of this repository
-# and their usage requires commercial license agreement.
-#
-# Karafka has also commercial-friendly license, commercial support and commercial components.
-#
-# By sending a pull request to the pro components, you are agreeing to transfer the copyright of
-# your code to Maciej Mensfeld.
 module Karafka
   module Pro
+    # This Karafka component is a Pro component.
+    # All of the commercial components are present in the lib/karafka/pro directory of this
+    # repository and their usage requires commercial license agreement.
+    #
+    # Karafka has also commercial-friendly license, commercial support and commercial components.
+    #
+    # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+    # of your code to Maciej Mensfeld.
+
     # Loader requires and loads all the pro components only when they are needed
     class Loader
       class << self
@@ -17,11 +18,19 @@ module Karafka
         # @param config [Dry::Configurable::Config] whole app config that we can alter with pro
         #   components
         def setup(config)
+          require_relative 'performance_tracker'
+          require_relative 'scheduler'
+          require_relative 'processing/jobs/consume_non_blocking'
+          require_relative 'active_job/consumer'
           require_relative 'active_job/dispatcher'
           require_relative 'active_job/job_options_contract'
 
+          config.internal.scheduler = Scheduler.new
+          config.internal.active_job.consumer = ActiveJob::Consumer
           config.internal.active_job.dispatcher = ActiveJob::Dispatcher.new
           config.internal.active_job.job_options_contract = ActiveJob::JobOptionsContract.new
+
+          config.monitor.subscribe(PerformanceTracker.instance)
         end
       end
     end

data/lib/karafka/pro/performance_tracker.rb

@@ -0,0 +1,80 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Pro
+    # This Karafka component is a Pro component.
+    # All of the commercial components are present in the lib/karafka/pro directory of this
+    # repository and their usage requires commercial license agreement.
+    #
+    # Karafka has also commercial-friendly license, commercial support and commercial components.
+    #
+    # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+    # of your code to Maciej Mensfeld.
+
+    # Tracker used to keep track of performance metrics
+    # It provides insights that can be used to optimize processing flow
+    class PerformanceTracker
+      include Singleton
+
+      # How many samples do we collect per topic partition
+      SAMPLES_COUNT = 200
+
+      private_constant :SAMPLES_COUNT
+
+      # Builds up nested concurrent hash for data tracking
+      def initialize
+        @processing_times = Concurrent::Hash.new do |topics_hash, topic|
+          topics_hash[topic] = Concurrent::Hash.new do |partitions_hash, partition|
+            # This array does not have to be concurrent because we always access single partition
+            # data via instrumentation that operates in a single thread via consumer
+            partitions_hash[partition] = []
+          end
+        end
+      end
+
+      # @param topic [String]
+      # @param partition [Integer]
+      # @return [Float] p95 processing time of a single message from a single topic partition
+      def processing_time_p95(topic, partition)
+        values = @processing_times[topic][partition]
+
+        return 0 if values.empty?
+        return values.first if values.size == 1
+
+        percentile(0.95, values)
+      end
+
+      # @private
+      # @param event [Dry::Events::Event] event details
+      # Tracks time taken to process a single message of a given topic partition
+      def on_consumer_consumed(event)
+        consumer = event[:caller]
+        messages = consumer.messages
+        topic = messages.metadata.topic
+        partition = messages.metadata.partition
+
+        samples = @processing_times[topic][partition]
+        samples << event[:time] / messages.count
+
+        return unless samples.size > SAMPLES_COUNT
+
+        samples.shift
+      end
+
+      private
+
+      # Computers the requested percentile out of provided values
+      # @param percentile [Float]
+      # @param values [Array<String>] all the values based on which we should
+      # @return [Float] computed percentile
+      def percentile(percentile, values)
+        values_sorted = values.sort
+
+        floor = (percentile * (values_sorted.length - 1) + 1).floor - 1
+        mod = (percentile * (values_sorted.length - 1) + 1).modulo(1)
+
+        values_sorted[floor] + (mod * (values_sorted[floor + 1] - values_sorted[floor]))
+      end
+    end
+  end
+end

data/lib/karafka/pro/processing/jobs/consume_non_blocking.rb

@@ -0,0 +1,38 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Pro
+    # Pro components related to processing part of Karafka
+    module Processing
+      # Pro jobs
+      module Jobs
+        # This Karafka component is a Pro component.
+        # All of the commercial components are present in the lib/karafka/pro directory of this
+        # repository and their usage requires commercial license agreement.
+        #
+        # Karafka has also commercial-friendly license, commercial support and commercial
+        # components.
+        #
+        # By sending a pull request to the pro components, you are agreeing to transfer the
+        # copyright of your code to Maciej Mensfeld.
+
+        # The main job type in a non-blocking variant.
+        # This variant works "like" the regular consumption but pauses the partition for as long
+        # as it is needed until a job is done.
+        #
+        # It can be useful when having long lasting jobs that would exceed `max.poll.interval`
+        # if would block.
+        #
+        # @note It needs to be working with a proper consumer that will handle the partition
+        #   management. This layer of the framework knows nothing about Kafka messages consumption.
+        class ConsumeNonBlocking < ::Karafka::Processing::Jobs::Consume
+          # Releases the blocking lock after it is done with the preparation phase for this job
+          def prepare
+            super
+            @non_blocking = true
+          end
+        end
+      end
+    end
+  end
+end

data/lib/karafka/pro/scheduler.rb

@@ -0,0 +1,54 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Pro
+    # This Karafka component is a Pro component.
+    # All of the commercial components are present in the lib/karafka/pro directory of this
+    # repository and their usage requires commercial license agreement.
+    #
+    # Karafka has also commercial-friendly license, commercial support and commercial components.
+    #
+    # By sending a pull request to the pro components, you are agreeing to transfer the copyright
+    # of your code to Maciej Mensfeld.
+
+    # Optimizes scheduler that takes into consideration of execution time needed to process
+    # messages from given topics partitions. It uses the non-preemptive LJF algorithm
+    #
+    # This scheduler is designed to optimize execution times on jobs that perform IO operations as
+    # when taking IO into consideration, the can achieve optimized parallel processing.
+    #
+    # This scheduler can also work with virtual partitions.
+    #
+    # Aside from consumption jobs, other jobs do not run often, thus we can leave them with
+    # default FIFO scheduler from the default Karafka scheduler
+    class Scheduler < ::Karafka::Scheduler
+      # Schedules jobs in the LJF order for consumption
+      #
+      # @param queue [Karafka::Processing::JobsQueue] queue where we want to put the jobs
+      # @param jobs_array [Array<Karafka::Processing::Jobs::Base>] jobs we want to schedule
+      #
+      def schedule_consumption(queue, jobs_array)
+        pt = PerformanceTracker.instance
+
+        ordered = []
+
+        jobs_array.each do |job|
+          messages = job.messages
+          message = messages.first
+
+          cost = pt.processing_time_p95(message.topic, message.partition) * messages.size
+
+          ordered << [job, cost]
+        end
+
+        ordered.sort_by!(&:last)
+        ordered.reverse!
+        ordered.map!(&:first)
+
+        ordered.each do |job|
+          queue << job
+        end
+      end
+    end
+  end
+end

data/lib/karafka/processing/executor.rb

@@ -4,10 +4,10 @@ module Karafka
   # Namespace that encapsulates all the logic related to processing data.
   module Processing
     # Executors:
-    # - run consumers code with provided messages batch (for `#call`) or run given teardown
-    #   operations when needed from separate threads.
-    # - they re-create consumer instances in case of partitions that were revoked
-    #   and assigned back.
+    # - run consumers code (for `#call`) or run given preparation / teardown operations when needed
+    #   from separate threads.
+    # - they re-create consumer instances in case of partitions that were revoked and assigned
+    #   back.
     #
     # @note Executors are not removed after partition is revoked. They are not that big and will
     #   be re-used in case of a re-claim
@@ -18,24 +18,27 @@ module Karafka
       # @return [String] subscription group id to which a given executor belongs
       attr_reader :group_id
 
+      attr_reader :messages
+
       # @param group_id [String] id of the subscription group to which the executor belongs
       # @param client [Karafka::Connection::Client] kafka client
       # @param topic [Karafka::Routing::Topic] topic for which this executor will run
-      # @param pause [Karafka::TimeTrackers::Pause] fetch pause object for crash pausing
-      def initialize(group_id, client, topic, pause)
+      # @param pause_tracker [Karafka::TimeTrackers::Pause] fetch pause tracker for pausing
+      def initialize(group_id, client, topic, pause_tracker)
         @id = SecureRandom.uuid
         @group_id = group_id
         @client = client
         @topic = topic
-        @pause = pause
+        @pause_tracker = pause_tracker
       end
 
-      # Runs consumer data processing against given batch and handles failures and errors.
+      # Builds the consumer instance, builds messages batch and sets all that is needed to run the
+      # user consumption logic
       #
-      # @param messages [Array<Rdkafka::Consumer::Message>] raw rdkafka messages
+      # @param messages [Array<Karafka::Messages::Message>]
       # @param received_at [Time] the moment we've received the batch (actually the moment we've)
       #   enqueued it, but good enough
-      def consume(messages, received_at)
+      def prepare(messages, received_at)
         # Recreate consumer with each batch if persistence is not enabled
         # We reload the consumers with each batch instead of relying on some external signals
         # when needed for consistency. That way devs may have it on or off and not in this
@@ -49,6 +52,11 @@ module Karafka
           received_at
         )
 
+        consumer.on_prepared
+      end
+
+      # Runs consumer data processing against given batch and handles failures and errors.
+      def consume
         # We run the consumer client logic...
         consumer.on_consume
       end
@@ -86,7 +94,7 @@ module Karafka
           consumer = @topic.consumer.new
           consumer.topic = @topic
           consumer.client = @client
-          consumer.pause = @pause
+          consumer.pause_tracker = @pause_tracker
           consumer.producer = ::Karafka::App.producer
           consumer
         end

data/lib/karafka/processing/executors_buffer.rb

@@ -23,21 +23,29 @@ module Karafka
         partition,
         pause
       )
-        topic = @subscription_group.topics.find { |ktopic| ktopic.name == topic }
+        ktopic = @subscription_group.topics.find(topic)
 
-        topic || raise(Errors::TopicNotFoundError, topic)
+        ktopic || raise(Errors::TopicNotFoundError, topic)
 
-        @buffer[topic][partition] ||= Executor.new(
+        @buffer[ktopic][partition] ||= Executor.new(
           @subscription_group.id,
           @client,
-          topic,
+          ktopic,
           pause
         )
       end
 
-      # Runs the shutdown on all active executors.
-      def shutdown
-        @buffer.values.map(&:values).flatten.each(&:shutdown)
+      # Iterates over all available executors and yields them together with topic and partition
+      # info
+      # @yieldparam [Routing::Topic] karafka routing topic object
+      # @yieldparam [Integer] partition number
+      # @yieldparam [Executor] given executor
+      def each
+        @buffer.each do |ktopic, partitions|
+          partitions.each do |partition, executor|
+            yield(ktopic, partition, executor)
+          end
+        end
       end
 
       # Clears the executors buffer. Useful for critical errors recovery.

data/lib/karafka/processing/jobs/base.rb

@@ -5,6 +5,8 @@ module Karafka
     # Namespace for all the jobs that are suppose to run in workers.
     module Jobs
       # Base class for all the jobs types that are suppose to run in workers threads.
+      # Each job can have 3 main entry-points: `#prepare`, `#call` and `#teardown`
+      # Only `#call` is required.
       class Base
         extend Forwardable
 
@@ -12,6 +14,32 @@ module Karafka
         def_delegators :executor, :id, :group_id
 
         attr_reader :executor
+
+        # Creates a new job instance
+        def initialize
+          # All jobs are blocking by default and they can release the lock when blocking operations
+          # are done (if needed)
+          @non_blocking = false
+        end
+
+        # When redefined can run any code that should run before executing the proper code
+        def prepare; end
+
+        # When redefined can run any code that should run after executing the proper code
+        def teardown; end
+
+        # @return [Boolean] is this a non-blocking job
+        #
+        # @note Blocking job is a job, that will cause the job queue to wait until it is finished
+        #   before removing the lock on new jobs being added
+        #
+        # @note All the jobs are blocking by default
+        #
+        # @note Job **needs** to mark itself as non-blocking only **after** it is done with all
+        #   the blocking things (pausing partition, etc).
+        def non_blocking?
+          @non_blocking
+        end
       end
     end
   end

data/lib/karafka/processing/jobs/consume.rb

@@ -6,10 +6,12 @@ module Karafka
       # The main job type. It runs the executor that triggers given topic partition messages
       # processing in an underlying consumer instance.
       class Consume < Base
+        # @return [Array<Rdkafka::Consumer::Message>] array with messages
+        attr_reader :messages
+
         # @param executor [Karafka::Processing::Executor] executor that is suppose to run a given
         #   job
-        # @param messages [Array<dkafka::Consumer::Message>] array with raw rdkafka messages with
-        #   which we are suppose to work
+        # @param messages [Karafka::Messages::Messages] karafka messages batch
         # @return [Consume]
         def initialize(executor, messages)
           @executor = executor
@@ -18,9 +20,14 @@ module Karafka
           super()
         end
 
-        # Runs the given executor.
+        # Runs the preparations on the executor
+        def prepare
+          executor.prepare(@messages, @created_at)
+        end
+
+        # Runs the given executor
         def call
-          executor.consume(@messages, @created_at)
+          executor.consume
         end
       end
     end