karafka 1.2.11

data/lib/karafka/connection/builder.rb

@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Builder used to construct Kafka client
+    module Builder
+      class << self
+        # Builds a Kafka::Cient instance that we use to work with Kafka cluster
+        # @return [::Kafka::Client] returns a Kafka client
+        def call
+          Kafka.new(*ApiAdapter.client)
+        end
+      end
+    end
+  end
+end

data/lib/karafka/connection/client.rb

@@ -0,0 +1,107 @@
+# 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
+
+      def_delegator :kafka_consumer, :seek
+
+      # 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>] kafka fetched messages
+      # @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.messages) }
+        else
+          # always yield an array of messages, so we have consistent API (always a batch)
+          kafka_consumer.each_message(*settings) { |message| yield([message]) }
+        end
+      rescue Kafka::ProcessingError => error
+        # 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: error.cause
+        )
+        pause(error.topic, error.partition)
+        retry
+      end
+
+      # Gracefuly 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)
+        kafka_consumer.pause(*ApiAdapter.pause(topic, partition, consumer_group))
+      end
+
+      # Marks a given message as consumed and commit the offsets
+      # @note In opposite to ruby-kafka, we commit the offset for each manual marking to be sure
+      #   that offset commit happen asap in case of a crash
+      # @param [Karafka::Params::Params] params message that we want to mark as processed
+      def mark_as_consumed(params)
+        kafka_consumer.mark_message_as_processed(
+          *ApiAdapter.mark_message_as_processed(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
+
+      # Triggers a non-optional blocking heartbeat that notifies Kafka about the fact, that this
+      # consumer / client is still up and running
+      def trigger_heartbeat
+        kafka_consumer.trigger_heartbeat!
+      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(
+          *ApiAdapter.consumer(consumer_group)
+        ).tap do |consumer|
+          consumer_group.topics.each do |topic|
+            consumer.subscribe(*ApiAdapter.subscribe(topic))
+          end
+        end
+      rescue Kafka::ConnectionError
+        # If we would not wait it would totally spam log file with failed
+        # attempts if Kafka is down
+        sleep(consumer_group.reconnect_timeout)
+        # We don't log and just reraise - this will be logged
+        # down the road
+        raise
+      end
+    end
+  end
+end

data/lib/karafka/connection/delegator.rb

@@ -0,0 +1,46 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # Class that delegates processing of messages for which we listen to a proper processor
+    module Delegator
+      class << self
+        # Delegates messages (does something with them)
+        # It will either schedule or run a proper processor action for messages
+        # @note This should be looped to obtain a constant delegating of new messages
+        # @note We catch all the errors here, to make sure that none failures
+        #   for a given consumption will affect other consumed messages
+        #   If we wouldn't catch it, it would propagate up until killing the thread
+        # @note It is a one huge method, because of performance reasons. It is much faster then
+        #   using send or invoking additional methods
+        # @param group_id [String] group_id of a group from which a given message came
+        # @param kafka_messages [Array<Kafka::FetchedMessage>] raw messages fetched from kafka
+        def call(group_id, kafka_messages)
+          # @note We always get messages by topic and partition so we can take topic from the
+          # first one and it will be valid for all the messages
+          topic = Persistence::Topic.fetch(group_id, kafka_messages[0].topic)
+          consumer = Persistence::Consumer.fetch(topic, kafka_messages[0].partition)
+
+          Karafka.monitor.instrument(
+            'connection.delegator.call',
+            caller: self,
+            consumer: consumer,
+            kafka_messages: kafka_messages
+          ) do
+            # 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 = kafka_messages
+              consumer.call
+            else
+              kafka_messages.each do |kafka_message|
+                consumer.params_batch = [kafka_message]
+                consumer.call
+              end
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/karafka/connection/listener.rb

@@ -0,0 +1,60 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Connection
+    # A single listener that listens to incoming messages from a single route
+    # @note It does not loop on itself - it needs to be executed in a loop
+    # @note Listener itself does nothing with the message - it will return to the block
+    #   a raw Kafka::FetchedMessage
+    class Listener
+      # @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
+      def initialize(consumer_group)
+        @consumer_group = consumer_group
+      end
+
+      # Runs prefetch callbacks and executes the main listener fetch loop
+      def call
+        Karafka::Callbacks.before_fetch_loop(
+          @consumer_group,
+          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
+      #   Kafka connections / Internet connection issues / Etc. Business logic problems should not
+      #   propagate this far
+      def fetch_loop
+        client.fetch_loop do |raw_messages|
+          # @note What happens here is a delegation of processing to a proper processor based
+          #   on the incoming messages characteristics
+          Karafka::Connection::Delegator.call(@consumer_group.id, raw_messages)
+        end
+        # This is on purpose - see the notes for this method
+        # rubocop:disable RescueException
+      rescue Exception => e
+        Karafka.monitor.instrument('connection.listener.fetch_loop.error', caller: self, error: e)
+        # rubocop:enable RescueException
+        @client.stop
+        sleep(@consumer_group.reconnect_timeout) && retry
+      end
+
+      # @return [Karafka::Connection::Client] wrapped kafka consuming client for a given topic
+      #   consumption
+      def client
+        @client ||= Client.new(@consumer_group)
+      end
+    end
+  end
+end

data/lib/karafka/consumers/callbacks.rb

@@ -0,0 +1,54 @@
+# 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
+
+      # Class methods needed to make callbacks run
+      module ClassMethods
+        TYPES.each do |type|
+          # A 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
+          define_method type do |method_name = nil, &block|
+            set_callback type, :before, method_name || block
+          end
+        end
+      end
+
+      # @param consumer_class [Class] consumer class that we extend with callbacks
+      def self.included(consumer_class)
+        consumer_class.class_eval do
+          extend ClassMethods
+          include ActiveSupport::Callbacks
+
+          # The call method is wrapped with a set of callbacks
+          # We won't run process if any of the callbacks throw abort
+          # @see http://api.rubyonrails.org/classes/ActiveSupport/Callbacks/ClassMethods.html#method-i-get_callbacks
+          TYPES.each { |type| define_callbacks type }
+        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
+        run_callbacks :after_fetch do
+          process
+        end
+      end
+    end
+  end
+end

data/lib/karafka/consumers/includer.rb

@@ -0,0 +1,51 @@
+# frozen_string_literal: true
+
+module Karafka
+  # Additional functionalities for consumers
+  module Consumers
+    # Module used to inject functionalities into a given consumer class, based on the consumer
+    # topic and its settings
+    # We don't need all the behaviors in all the cases, so it is not worth having everything
+    # in all the cases all the time
+    module Includer
+      class << self
+        # @param consumer_class [Class] consumer class, that will get some functionalities
+        #   based on the topic under which it operates
+        def call(consumer_class)
+          topic = consumer_class.topic
+
+          bind_backend(consumer_class, topic)
+          bind_params(consumer_class, topic)
+          bind_responders(consumer_class, topic)
+        end
+
+        private
+
+        # Figures out backend for a given consumer class, based on the topic backend and
+        #   includes it into the consumer class
+        # @param consumer_class [Class] consumer class
+        # @param topic [Karafka::Routing::Topic] topic of a consumer class
+        def bind_backend(consumer_class, topic)
+          backend = Kernel.const_get("::Karafka::Backends::#{topic.backend.to_s.capitalize}")
+          consumer_class.include backend
+        end
+
+        # Adds a single #params support for non batch processed topics
+        # @param consumer_class [Class] consumer class
+        # @param topic [Karafka::Routing::Topic] topic of a consumer class
+        def bind_params(consumer_class, topic)
+          return if topic.batch_consuming
+          consumer_class.include SingleParams
+        end
+
+        # Adds responders support for topics and consumers with responders defined for them
+        # @param consumer_class [Class] consumer class
+        # @param topic [Karafka::Routing::Topic] topic of a consumer class
+        def bind_responders(consumer_class, topic)
+          return unless topic.responder
+          consumer_class.include Responders
+        end
+      end
+    end
+  end
+end

data/lib/karafka/consumers/responders.rb

@@ -0,0 +1,24 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Consumers
+    # Feature that allows us to use responders flow in consumer
+    module Responders
+      # Responds with given data using given responder. This allows us to have a similar way of
+      # defining flows like synchronous protocols
+      # @param data Anything we want to pass to responder based on which we want to trigger further
+      #   Kafka responding
+      def respond_with(*data)
+        Karafka.monitor.instrument(
+          'consumers.responders.respond_with',
+          caller: self,
+          data: data
+        ) do
+          # @note we build a new instance of responder each time, as a long-running (persisted)
+          #   consumers can respond multiple times during the lifecycle
+          topic.responder.new(topic.parser).call(*data)
+        end
+      end
+    end
+  end
+end

data/lib/karafka/consumers/single_params.rb

@@ -0,0 +1,15 @@
+# frozen_string_literal: true
+
+module Karafka
+  module Consumers
+    # Params alias for single message consumption consumers
+    module SingleParams
+      private
+
+      # @return [Karafka::Params::Params] params instance for non batch consumption consumers
+      def params
+        params_batch.first
+      end
+    end
+  end
+end

data/lib/karafka/errors.rb

@@ -0,0 +1,50 @@
+# frozen_string_literal: true
+
+module Karafka
+  # Namespace used to encapsulate all the internal errors of Karafka
+  module Errors
+    # Base class for all the Karafka internal errors
+    BaseError = Class.new(StandardError)
+
+    # Should be raised when we attemp to parse incoming params but parsing fails
+    #   If this error (or its descendant) is detected, we will pass the raw message
+    #   into params and proceed further
+    ParserError = Class.new(BaseError)
+
+    # Raised when router receives topic name which does not correspond with any routes
+    # This can only happen in a case when:
+    #   - you've received a message and we cannot match it with a consumer
+    #   - you've changed the routing, so router can no longer associate your topic to
+    #     any consumer
+    #   - or in a case when you do a lot of metaprogramming and you change routing/etc on runtime
+    #
+    # In case this happens, you will have to create a temporary route that will allow
+    # you to "eat" everything from the Sidekiq queue.
+    # @see https://github.com/karafka/karafka/issues/135
+    NonMatchingRouteError = Class.new(BaseError)
+
+    # Raised when we don't use or use responder not in the way it expected to based on the
+    # topics usage definitions
+    InvalidResponderUsage = Class.new(BaseError)
+
+    # Raised when options that we provide to the responder to respond aren't what the schema
+    # requires
+    InvalidResponderMessageOptions = Class.new(BaseError)
+
+    # Raised when configuration doesn't match with validation schema
+    InvalidConfiguration = Class.new(BaseError)
+
+    # Raised when we try to use Karafka CLI commands (except install) without a bootfile
+    MissingBootFile = Class.new(BaseError)
+
+    # Raised when we want to read a persisted thread messages consumer but it is unavailable
+    # This should never happen and if it does, please contact us
+    MissingClient = Class.new(BaseError)
+
+    # Raised when want to hook up to an event that is not registered and supported
+    UnregisteredMonitorEvent = Class.new(BaseError)
+
+    # Raised when we've waited enough for shutting down an unresponding process
+    ForcefulShutdown = Class.new(BaseError)
+  end
+end

data/lib/karafka/fetcher.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+module Karafka
+  # Class used to run the Karafka consumer and handle shutting down, restarting etc
+  # @note Creating multiple fetchers will result in having multiple connections to the same
+  #   topics, which means that if there are no partitions, it won't use them.
+  class Fetcher
+    class << self
+      # Starts listening on all the listeners asynchronously
+      # Fetch loop should never end, which means that we won't create more actor clusters
+      # so we don't have to terminate them
+      def call
+        threads = listeners.map do |listener|
+          # We abort on exception because there should be an exception handling developed for
+          # each listener running in separate threads, so the exceptions should never leak
+          # and if that happens, it means that something really bad happened and we should stop
+          # the whole process
+          Thread
+            .new { listener.call }
+            .tap { |thread| thread.abort_on_exception = true }
+        end
+
+        # We aggregate threads here for a supervised shutdown process
+        threads.each { |thread| Karafka::Server.consumer_threads << thread }
+        threads.each(&:join)
+      # If anything crashes here, we need to raise the error and crush the runner because it means
+      # that something terrible happened
+      rescue StandardError => e
+        Karafka.monitor.instrument('fetcher.call.error', caller: self, error: e)
+        Karafka::App.stop!
+        raise e
+      end
+
+      private
+
+      # @return [Array<Karafka::Connection::Listener>] listeners that will consume messages
+      def listeners
+        @listeners ||= App.consumer_groups.active.map do |consumer_group|
+          Karafka::Connection::Listener.new(consumer_group)
+        end
+      end
+    end
+  end
+end