karafka 1.0.1 → 1.4.14

data/lib/karafka/cli.rb

@@ -10,6 +10,8 @@ module Karafka
   class Cli < Thor
     package_name 'Karafka'
 
+    default_task :missingno
+
     class << self
       # Loads all Cli commands into Thor framework
       # This method should be executed before we run Karafka::Cli.start, otherwise we won't
@@ -20,6 +22,12 @@ module Karafka
         end
       end
 
+      # When there is a CLI crash, exit
+      # @return [true]
+      def exit_on_failure?
+        true
+      end
+
       private
 
       # @return [Array<Class>] Array with Cli action classes that can be used as commands
@@ -37,7 +45,7 @@ end
 # This is kinda trick - since we don't have a autoload and other magic stuff
 # like Rails does, so instead this method allows us to replace currently running
 # console with a new one via Kernel.exec. It will start console with new code loaded
-# Yes we know that it is not turbofast, however it is turbo convinient and small
+# Yes, we know that it is not turbo fast, however it is turbo convenient and small
 #
 # Also - the KARAFKA_CONSOLE is used to detect that we're executing the irb session
 # so this method is only available when the Karafka console is running
@@ -47,7 +55,7 @@ end
 if ENV['KARAFKA_CONSOLE']
   # Reloads Karafka irb console session
   def reload!
-    puts "Reloading...\n"
+    Karafka.logger.info "Reloading...\n"
     Kernel.exec Karafka::Cli::Console.command
   end
 end

data/lib/karafka/code_reloader.rb

@@ -0,0 +1,67 @@
+# frozen_string_literal: true
+
+module Karafka
+  # Special type of a listener, that is not an instrumentation one, but one that triggers
+  # code reload in the development mode after each fetched batch (or message)
+  #
+  # Please refer to the development code reload sections for details on the benefits and downsides
+  # of the in-process code reloading
+  class CodeReloader
+    # This mutex is needed as we might have an application that has multiple consumer groups
+    # running in separate threads and we should not trigger reload before fully reloading the app
+    # in previous thread
+    MUTEX = Mutex.new
+
+    private_constant :MUTEX
+
+    # @param reloaders [Array<Object>] any code loaders that we use in this app. Whether it is
+    #  the Rails loader, Zeitwerk or anything else that allows reloading triggering
+    # @param block [Proc] yields given block just before reloading. This can be used to hook custom
+    #   reloading stuff, that ain't reloaders (for example for resetting dry-events registry)
+    def initialize(*reloaders, &block)
+      @reloaders = reloaders
+      @block = block
+    end
+
+    # Binds to the instrumentation events and triggers reload
+    # @param _event [Dry::Event] empty dry event
+    # @note Since we de-register all the user defined objects and redraw routes, it means that
+    #   we won't be able to do a multi-batch buffering in the development mode as each of the
+    #   batches will be buffered on a newly created "per fetch" instance.
+    def on_connection_listener_fetch_loop(_event)
+      reload
+    end
+
+    private
+
+    # Triggers reload of both standard and Rails reloaders as well as expires all internals of
+    # Karafka, so it can be rediscovered and rebuilt
+    def reload
+      MUTEX.synchronize do
+        if @reloaders[0].respond_to?(:execute)
+          reload_with_rails
+        else
+          reload_without_rails
+        end
+      end
+    end
+
+    # Rails reloading procedure
+    def reload_with_rails
+      updatable = @reloaders.select(&:updated?)
+
+      return if updatable.empty?
+
+      updatable.each(&:execute)
+      @block&.call
+      Karafka::App.reload
+    end
+
+    # Zeitwerk and other reloaders
+    def reload_without_rails
+      @reloaders.each(&:reload)
+      @block&.call
+      Karafka::App.reload
+    end
+  end
+end

data/lib/karafka/connection/{config_adapter.rb → api_adapter.rb}

@@ -3,19 +3,23 @@
 module Karafka
   # Namespace for all the things related to Kafka connection
   module Connection
-    # Mapper used to convert our internal settings into ruby-kafka settings
+    # Mapper used to convert our internal settings into ruby-kafka settings based on their
+    # API requirements.
     # Since ruby-kafka has more and more options and there are few "levels" on which
     # we have to apply them (despite the fact, that in Karafka you configure all of it
     # in one place), we have to remap it into what ruby-kafka driver requires
     # @note The good thing about Kafka.new method is that it ignores all options that
     #   do nothing. So we don't have to worry about injecting our internal settings
     #   into the client and breaking stuff
-    module ConfigAdapter
+    module ApiAdapter
       class << self
         # Builds all the configuration settings for Kafka.new method
-        # @param _consumer_group [Karafka::Routing::ConsumerGroup] consumer group details
-        # @return [Hash] hash with all the settings required by Kafka.new method
-        def client(_consumer_group)
+        # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group details
+        # @return [Array<Hash>] Array with all the client arguments including hash with all
+        #   the settings required by Kafka.new method
+        # @note We return array, so we can inject any arguments we want, in case of changes in the
+        #   raw driver
+        def client(consumer_group)
           # This one is a default that takes all the settings except special
           # cases defined in the map
           settings = {
@@ -23,22 +27,29 @@ module Karafka
             client_id: ::Karafka::App.config.client_id
           }
 
-          kafka_configs.each do |setting_name, setting_value|
+          kafka_configs.each_key do |setting_name|
             # All options for config adapter should be ignored as we're just interested
             # in what is left, as we want to pass all the options that are "typical"
-            # and not listed in the config_adapter special cases mapping. All the values
-            # from the config_adapter mapping go somewhere else, not to the client directly
-            next if AttributesMap.config_adapter.values.flatten.include?(setting_name)
+            # and not listed in the api_adapter special cases mapping. All the values
+            # from the api_adapter mapping go somewhere else, not to the client directly
+            next if AttributesMap.api_adapter.values.flatten.include?(setting_name)
 
-            settings[setting_name] = setting_value
+            # Settings for each consumer group are either defined per consumer group or are
+            # inherited from the global/general settings level, thus we don't have to fetch them
+            # from the kafka settings as they are already on a consumer group level
+            settings[setting_name] = consumer_group.public_send(setting_name)
           end
 
-          sanitize(settings)
+          settings_hash = sanitize(settings)
+
+          # Normalization for the way Kafka::Client accepts arguments from  0.5.3
+          [settings_hash.delete(:seed_brokers), settings_hash]
         end
 
         # Builds all the configuration settings for kafka#consumer method
         # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group details
-        # @return [Hash] hash with all the settings required by Kafka#consumer method
+        # @return [Hash] all the consumer keyword arguments including hash with all
+        #   the settings required by Kafka#consumer
         def consumer(consumer_group)
           settings = { group_id: consumer_group.id }
           settings = fetch_for(:consumer, consumer_group, settings)
@@ -48,25 +59,61 @@ module Karafka
         # Builds all the configuration settings for kafka consumer consume_each_batch and
         #   consume_each_message methods
         # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group details
-        # @return [Hash] hash with all the settings required by
+        # @return [Hash] hash with all the arguments required by consuming method
+        #   including all the settings required by
         #   Kafka::Consumer#consume_each_message and Kafka::Consumer#consume_each_batch method
-        def consuming(consumer_group)
-          sanitize(fetch_for(:consuming, consumer_group))
+        def consumption(consumer_group)
+          sanitize(
+            fetch_for(
+              :consumption,
+              consumer_group,
+              automatically_mark_as_processed: consumer_group.automatically_mark_as_consumed
+            )
+          )
         end
 
         # Builds all the configuration settings for kafka consumer#subscribe method
         # @param topic [Karafka::Routing::Topic] topic that holds details for a given subscription
         # @return [Hash] hash with all the settings required by kafka consumer#subscribe method
-        def subscription(topic)
-          settings = fetch_for(:subscription, topic)
+        def subscribe(topic)
+          settings = fetch_for(:subscribe, topic)
           [Karafka::App.config.topic_mapper.outgoing(topic.name), sanitize(settings)]
         end
 
         # Builds all the configuration settings required by kafka consumer#pause method
+        # @param topic [String] topic that we want to pause
+        # @param partition [Integer] number partition that we want to pause
         # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group details
-        # @return [Hash] hash with all the settings required to pause kafka consumer
-        def pausing(consumer_group)
-          { timeout: consumer_group.pause_timeout }
+        # @return [Hash] hash with all the details required to pause kafka consumer
+        def pause(topic, partition, consumer_group)
+          {
+            args: [Karafka::App.config.topic_mapper.outgoing(topic), partition],
+            kwargs: {
+              timeout: consumer_group.pause_timeout,
+              max_timeout: consumer_group.pause_max_timeout,
+              exponential_backoff: consumer_group.pause_exponential_backoff
+            }
+          }
+        end
+
+        # Remaps topic details taking the topic mapper feature into consideration.
+        # @param params [Karafka::Params::Params] params instance
+        # @return [Array] array with all the details needed by ruby-kafka to mark message
+        #   as processed
+        # @note When default empty topic mapper is used, no need for any conversion as the
+        #   internal and external format are exactly the same
+        def mark_message_as_processed(params)
+          # Majority of users don't use custom topic mappers. No need to change anything when it
+          # is a default mapper that does not change anything. Only some cloud providers require
+          # topics to be remapped
+          return [params.metadata] if Karafka::App.config.topic_mapper.is_a?(
+            Karafka::Routing::TopicMapper
+          )
+
+          # @note We don't use tap as it is around 13% slower than non-dup version
+          dupped = params.metadata.dup
+          dupped['topic'] = Karafka::App.config.topic_mapper.outgoing(params.metadata.topic)
+          [dupped]
         end
 
         private
@@ -79,10 +126,12 @@ module Karafka
         def fetch_for(namespace_key, route_layer, preexisting_settings = {})
           kafka_configs.each_key do |setting_name|
             # Ignore settings that are not related to our namespace
-            next unless AttributesMap.config_adapter[namespace_key].include?(setting_name)
+            next unless AttributesMap.api_adapter[namespace_key].include?(setting_name)
+
             # Ignore settings that are already initialized
             # In case they are in preexisting settings fetched differently
-            next if preexisting_settings.keys.include?(setting_name)
+            next if preexisting_settings.key?(setting_name)
+
             # Fetch all the settings from a given layer object. Objects can handle the fallback
             # to the kafka settings, so
             preexisting_settings[setting_name] = route_layer.send(setting_name)

data/lib/karafka/connection/batch_delegator.rb

@@ -0,0 +1,55 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Class that delegates processing of batch received messages for which we listen to
+    # a proper processor
+    module BatchDelegator
+      class << self
+        # Delegates messages (does something with them)
+        # It will either schedule or run a proper processor action for messages
+        # @param group_id [String] group_id of a group from which a given message came
+        # @param kafka_batch [<Kafka::FetchedBatch>] raw messages fetched batch
+        # @note This should be looped to obtain a constant delegating of new messages
+        def call(group_id, kafka_batch)
+          topic = Persistence::Topics.fetch(group_id, kafka_batch.topic)
+          consumer = Persistence::Consumers.fetch(topic, kafka_batch.partition)
+
+          Karafka.monitor.instrument(
+            'connection.batch_delegator.call',
+            caller: self,
+            consumer: consumer,
+            kafka_batch: kafka_batch
+          ) do
+            # Due to how ruby-kafka is built, we have the metadata that is stored on the batch
+            # level only available for batch consuming
+            consumer.batch_metadata = Params::Builders::BatchMetadata.from_kafka_batch(
+              kafka_batch,
+              topic
+            )
+
+            kafka_messages = kafka_batch.messages
+
+            # Depending on a case (persisted or not) we might use new consumer instance per
+            # each batch, or use the same one for all of them (for implementing buffering, etc.)
+            if topic.batch_consuming
+              consumer.params_batch = Params::Builders::ParamsBatch.from_kafka_messages(
+                kafka_messages,
+                topic
+              )
+              consumer.call
+            else
+              kafka_messages.each do |kafka_message|
+                consumer.params_batch = Params::Builders::ParamsBatch.from_kafka_messages(
+                  [kafka_message],
+                  topic
+                )
+                consumer.call
+              end
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/karafka/connection/builder.rb

@@ -0,0 +1,23 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Builder used to construct Kafka client
+    module Builder
+      class << self
+        # Builds a Kafka::Client instance that we use to work with Kafka cluster
+        # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group for which we want
+        #   to have a new Kafka client
+        # @return [::Kafka::Client] returns a Kafka client
+        def call(consumer_group)
+          settings = ApiAdapter.client(consumer_group)
+
+          Kafka.new(
+            settings[0],
+            **settings[1]
+          )
+        end
+      end
+    end
+  end
+end

data/lib/karafka/connection/client.rb

@@ -0,0 +1,120 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Class used as a wrapper around Ruby-Kafka client to simplify additional
+    # features that we provide/might provide in future and to hide the internal implementation
+    class Client
+      extend Forwardable
+
+      %i[
+        seek
+        trigger_heartbeat
+        trigger_heartbeat!
+      ].each do |delegated_method|
+        def_delegator :kafka_consumer, delegated_method
+      end
+
+      # Creates a queue consumer client that will pull the data from Kafka
+      # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group for which
+      #   we create a client
+      # @return [Karafka::Connection::Client] group consumer that can subscribe to
+      #   multiple topics
+      def initialize(consumer_group)
+        @consumer_group = consumer_group
+        Persistence::Client.write(self)
+      end
+
+      # Opens connection, gets messages and calls a block for each of the incoming messages
+      # @yieldparam [Array<Kafka::FetchedMessage>, Symbol] kafka response with an info about
+      #   the type of the fetcher that is being used
+      # @note This will yield with raw messages - no preprocessing or reformatting.
+      def fetch_loop
+        settings = ApiAdapter.consumption(consumer_group)
+
+        if consumer_group.batch_fetching
+          kafka_consumer.each_batch(**settings) { |batch| yield(batch, :batch) }
+        else
+          kafka_consumer.each_message(**settings) { |message| yield(message, :message) }
+        end
+      # @note We catch only the processing errors as any other are considered critical (exceptions)
+      #   and should require a client restart with a backoff
+      rescue Kafka::ProcessingError => e
+        # If there was an error during consumption, we have to log it, pause current partition
+        # and process other things
+        Karafka.monitor.instrument(
+          'connection.client.fetch_loop.error',
+          caller: self,
+          error: e.cause
+        )
+        pause(e.topic, e.partition)
+        retry
+      end
+
+      # Gracefully stops topic consumption
+      # @note Stopping running consumers without a really important reason is not recommended
+      #   as until all the consumers are stopped, the server will keep running serving only
+      #   part of the messages
+      def stop
+        @kafka_consumer&.stop
+        @kafka_consumer = nil
+      end
+
+      # Pauses fetching and consumption of a given topic partition
+      # @param topic [String] topic that we want to pause
+      # @param partition [Integer] number partition that we want to pause
+      def pause(topic, partition)
+        args, kwargs = ApiAdapter.pause(topic, partition, consumer_group).values_at(:args, :kwargs)
+        kafka_consumer.pause(*args, **kwargs)
+      end
+
+      # Marks given message as consumed
+      # @param [Karafka::Params::Params] params message that we want to mark as processed
+      # @note This method won't trigger automatic offsets commits, rather relying on the ruby-kafka
+      #   offsets time-interval based committing
+      def mark_as_consumed(params)
+        kafka_consumer.mark_message_as_processed(
+          *ApiAdapter.mark_message_as_processed(params)
+        )
+      end
+
+      # Marks a given message as consumed and commit the offsets in a blocking way
+      # @param [Karafka::Params::Params] params message that we want to mark as processed
+      # @note This method commits the offset for each manual marking to be sure
+      #   that offset commit happen asap in case of a crash
+      def mark_as_consumed!(params)
+        mark_as_consumed(params)
+        # Trigger an immediate, blocking offset commit in order to minimize the risk of crashing
+        # before the automatic triggers have kicked in.
+        kafka_consumer.commit_offsets
+      end
+
+      private
+
+      attr_reader :consumer_group
+
+      # @return [Kafka::Consumer] returns a ready to consume Kafka consumer
+      #   that is set up to consume from topics of a given consumer group
+      def kafka_consumer
+        # @note We don't cache the connection internally because we cache kafka_consumer that uses
+        #   kafka client object instance
+        @kafka_consumer ||= Builder.call(consumer_group).consumer(
+          **ApiAdapter.consumer(consumer_group)
+        ).tap do |consumer|
+          consumer_group.topics.each do |topic|
+            settings = ApiAdapter.subscribe(topic)
+
+            consumer.subscribe(settings[0], **settings[1])
+          end
+        end
+      rescue Kafka::ConnectionError
+        # If we would not wait it will spam log file with failed
+        # attempts if Kafka is down
+        sleep(consumer_group.reconnect_timeout)
+        # We don't log and just re-raise - this will be logged
+        # down the road
+        raise
+      end
+    end
+  end
+end

data/lib/karafka/connection/listener.rb

@@ -7,12 +7,6 @@ module Karafka
     # @note Listener itself does nothing with the message - it will return to the block
     #   a raw Kafka::FetchedMessage
     class Listener
-      include Celluloid
-
-      execute_block_on_receiver :fetch_loop
-
-      attr_reader :consumer_group
-
       # @param consumer_group [Karafka::Routing::ConsumerGroup] consumer group that holds details
       #   on what topics and with what settings should we listen
       # @return [Karafka::Connection::Listener] listener instance
@@ -20,38 +14,57 @@ module Karafka
         @consumer_group = consumer_group
       end
 
+      # Runs prefetch callbacks and executes the main listener fetch loop
+      def call
+        Karafka.monitor.instrument(
+          'connection.listener.before_fetch_loop',
+          consumer_group: @consumer_group,
+          client: client
+        )
+        fetch_loop
+      end
+
+      private
+
       # Opens connection, gets messages and calls a block for each of the incoming messages
-      # @yieldparam [String] consumer group id
-      # @yieldparam [Array<Kafka::FetchedMessage>] kafka fetched messages
-      # @note This will yield with a raw message - no preprocessing or reformatting
       # @note We catch all the errors here, so they don't affect other listeners (or this one)
       #   so we will be able to listen and consume other incoming messages.
       #   Since it is run inside Karafka::Connection::ActorCluster - catching all the exceptions
-      #   won't crash the whole cluster. Here we mostly focus on catchin the exceptions related to
+      #   won't crash the whole cluster. Here we mostly focus on catching the exceptions related to
       #   Kafka connections / Internet connection issues / Etc. Business logic problems should not
       #   propagate this far
-      def fetch_loop(block)
-        messages_consumer.fetch_loop do |raw_messages|
-          block.call(consumer_group.id, raw_messages)
+      def fetch_loop
+        # @note What happens here is a delegation of processing to a proper processor based
+        #   on the incoming messages characteristics
+        client.fetch_loop do |raw_data, type|
+          Karafka.monitor.instrument('connection.listener.fetch_loop')
+
+          case type
+          when :message
+            MessageDelegator.call(@consumer_group.id, raw_data)
+          when :batch
+            BatchDelegator.call(@consumer_group.id, raw_data)
+          end
         end
         # This is on purpose - see the notes for this method
-        # rubocop:disable RescueException
+        # rubocop:disable Lint/RescueException
       rescue Exception => e
-        # rubocop:enable RescueException
-        Karafka.monitor.notice_error(self.class, e)
-        @messages_consumer&.stop
-        retry if @messages_consumer
+        Karafka.monitor.instrument('connection.listener.fetch_loop.error', caller: self, error: e)
+        # rubocop:enable Lint/RescueException
+        # We can stop client without a problem, as it will reinitialize itself when running the
+        # `fetch_loop` again
+        @client.stop
+        # We need to clear the consumers cache for current connection when fatal error happens and
+        # we reset the connection. Otherwise for consumers with manual offset management, the
+        # persistence might have stored some data that would be reprocessed
+        Karafka::Persistence::Consumers.clear
+        sleep(@consumer_group.reconnect_timeout) && retry
       end
 
-      private
-
-      # @return [Karafka::Connection::MessagesConsumer] wrapped kafka consumer for a given topic
+      # @return [Karafka::Connection::Client] wrapped kafka consuming client for a given topic
       #   consumption
-      # @note It adds consumer into Karafka::Server consumers pool for graceful shutdown on exit
-      def messages_consumer
-        @messages_consumer ||= MessagesConsumer.new(consumer_group).tap do |consumer|
-          Karafka::Server.consumers << consumer if Karafka::Server.consumers
-        end
+      def client
+        @client ||= Client.new(@consumer_group)
       end
     end
   end

data/lib/karafka/connection/message_delegator.rb

@@ -0,0 +1,36 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Class that delegates processing of a single received message for which we listen to
+    # a proper processor
+    module MessageDelegator
+      class << self
+        # Delegates message (does something with it)
+        # It will either schedule or run a proper processor action for the incoming message
+        # @param group_id [String] group_id of a group from which a given message came
+        # @param kafka_message [<Kafka::FetchedMessage>] raw message from kafka
+        # @note This should be looped to obtain a constant delegating of new messages
+        def call(group_id, kafka_message)
+          topic = Persistence::Topics.fetch(group_id, kafka_message.topic)
+          consumer = Persistence::Consumers.fetch(topic, kafka_message.partition)
+
+          Karafka.monitor.instrument(
+            'connection.message_delegator.call',
+            caller: self,
+            consumer: consumer,
+            kafka_message: kafka_message
+          ) do
+            # @note We always get a single message within single delegator, which means that
+            # we don't care if user marked it as a batch consumed or not.
+            consumer.params_batch = Params::Builders::ParamsBatch.from_kafka_messages(
+              [kafka_message],
+              topic
+            )
+            consumer.call
+          end
+        end
+      end
+    end
+  end
+end

data/lib/karafka/consumers/batch_metadata.rb

@@ -0,0 +1,10 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Consumers
+    # Brings the batch metadata into consumers that support batch_fetching
+    module BatchMetadata
+      attr_accessor :batch_metadata
+    end
+  end
+end

data/lib/karafka/consumers/callbacks.rb

@@ -0,0 +1,71 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Consumers
+    # Additional callbacks that can be used to trigger some actions on certain moments like
+    # manual offset management, committing or anything else outside of a standard messages flow
+    # They are not included by default, as we don't want to provide functionalities that are
+    # not required by users by default
+    # Please refer to the wiki callbacks page for more details on how to use them
+    module Callbacks
+      # Types of events on which we run callbacks
+      TYPES = %i[
+        after_fetch
+        after_poll
+        before_poll
+        before_stop
+      ].freeze
+
+      private_constant :TYPES
+
+      # Class methods needed to make callbacks run
+      module ClassMethods
+        TYPES.each do |type|
+          # Creates a callback wrapper
+          #
+          # @param method_name [Symbol, String] method name or nil if we plan to provide a block
+          # @yield A block with a code that should be executed before scheduling
+          # @note We don't have to optimize the key fetching here as those are class methods that
+          #   are evaluated once upon start
+          define_method(type) do |method_name = nil, &block|
+            key = "consumers.#{Helpers::Inflector.map(to_s)}.#{type}"
+            Karafka::App.monitor.register_event(key)
+
+            Karafka::App.monitor.subscribe(key) do |event|
+              context = event[:context]
+
+              if method_name
+                context.send(method_name)
+              else
+                context.instance_eval(&block)
+              end
+            end
+          end
+        end
+      end
+
+      class << self
+        # @param consumer_class [Class] consumer class that we extend with callbacks
+        def included(consumer_class)
+          consumer_class.class_eval do
+            extend ClassMethods
+          end
+        end
+      end
+
+      # Executes the default consumer flow, runs callbacks and if not halted will call process
+      # method of a proper backend. It is here because it interacts with the default Karafka
+      # call flow and needs to be overwritten to support callbacks
+      def call
+        if self.class.respond_to?(:after_fetch)
+          Karafka::App.monitor.instrument(
+            "consumers.#{Helpers::Inflector.map(self.class.to_s)}.after_fetch",
+            context: self
+          )
+        end
+
+        process
+      end
+    end
+  end
+end