rdkafka 0.30.0 → 0.31.0

data/lib/rdkafka/bindings/callbacks.rb

@@ -0,0 +1,100 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    LogCallback = FFI::Function.new(
+      :void, [:pointer, :int, :string, :string]
+    ) do |_client_ptr, level, _level_string, line|
+      severity = case level
+      when 0, 1, 2
+        Logger::FATAL
+      when 3
+        Logger::ERROR
+      when 4
+        Logger::WARN
+      when 5, 6
+        Logger::INFO
+      when 7
+        Logger::DEBUG
+      else
+        Logger::UNKNOWN
+      end
+
+      Rdkafka::Config.ensure_log_thread
+      Rdkafka::Config.log_queue << [severity, "rdkafka: #{line}"]
+    end
+
+    StatsCallback = FFI::Function.new(
+      :int, [:pointer, :string, :int, :pointer]
+    ) do |_client_ptr, json, _json_len, _opaque|
+      if Rdkafka::Config.statistics_callback
+        stats = JSON.parse(json)
+
+        # If user requested statistics callbacks, we can use the statistics data to get the
+        # partitions count for each topic when this data is published. That way we do not have
+        # to query this information when user is using `partition_key`. This takes around 0.02ms
+        # every statistics interval period (most likely every 5 seconds) and saves us from making
+        # any queries to the cluster for the partition count.
+        #
+        # One edge case is if user would set the `statistics.interval.ms` much higher than the
+        # default current partition count refresh (30 seconds). This is taken care of as the lack
+        # of reporting to the partitions cache will cause cache expire and blocking refresh.
+        #
+        # If user sets `topic.metadata.refresh.interval.ms` too high this is on the user.
+        #
+        # Since this cache is shared, having few consumers and/or producers in one process will
+        # automatically improve the querying times even with low refresh times.
+        (stats["topics"] || EMPTY_HASH).each do |topic_name, details|
+          partitions_count = details["partitions"].keys.count { |k| !(k == RD_KAFKA_PARTITION_UA_STR) }
+
+          next unless partitions_count.positive?
+
+          Rdkafka::Producer.partitions_count_cache.set(topic_name, partitions_count)
+        end
+
+        Rdkafka::Config.statistics_callback.call(stats)
+      end
+
+      # Return 0 so librdkafka frees the json string
+      RD_KAFKA_RESP_ERR_NO_ERROR
+    end
+
+    ErrorCallback = FFI::Function.new(
+      :void, [:pointer, :int, :string, :pointer]
+    ) do |client_ptr, err_code, reason, _opaque|
+      if Rdkafka::Config.error_callback
+        instance_name = client_ptr.null? ? nil : Rdkafka::Bindings.rd_kafka_name(client_ptr)
+        error = Rdkafka::RdkafkaError.new(err_code, broker_message: reason, instance_name: instance_name)
+        error.set_backtrace(caller)
+        Rdkafka::Config.error_callback.call(error)
+      end
+    end
+
+    # The OAuth callback is currently global and contextless. This means that the callback will be
+    # called for all instances, and the callback must be able to determine to which instance it is
+    # associated. The instance name will be provided in the callback, allowing the callback to
+    # reference the correct instance.
+    #
+    # An example of how to use the instance name in the callback is given below.
+    # The `refresh_token` is configured as the `oauthbearer_token_refresh_callback`.
+    # `instances` is a map of client names to client instances, maintained by the user.
+    #
+    # ```
+    #   def refresh_token(config, client_name)
+    #     client = instances[client_name]
+    #     client.oauthbearer_set_token(
+    #       token: 'new-token-value',
+    #       lifetime_ms: token-lifetime-ms,
+    #       principal_name: 'principal-name'
+    #     )
+    #   end
+    # ```
+    OAuthbearerTokenRefreshCallback = FFI::Function.new(
+      :void, [:pointer, :string, :pointer]
+    ) do |client_ptr, config, _opaque|
+      if Rdkafka::Config.oauthbearer_token_refresh_callback && !client_ptr.null?
+        Rdkafka::Config.oauthbearer_token_refresh_callback.call(config, Rdkafka::Bindings.rd_kafka_name(client_ptr))
+      end
+    end
+  end
+end

data/lib/rdkafka/bindings/configuration.rb

@@ -0,0 +1,50 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Configuration
+
+    enum :kafka_config_response, [
+      :config_unknown, -2,
+      :config_invalid, -1,
+      :config_ok, 0
+    ]
+
+    attach_function :rd_kafka_conf_new, [], :pointer
+    attach_function :rd_kafka_conf_set, [:pointer, :string, :string, :pointer, :int], :kafka_config_response
+    attach_function :rd_kafka_conf_get, [:pointer, :string, :pointer, :pointer], :kafka_config_response
+    attach_function :rd_kafka_conf, [:pointer], :pointer
+    attach_function :rd_kafka_conf_dump, [:pointer, :pointer], :pointer
+    attach_function :rd_kafka_conf_dump_free, [:pointer, :size_t], :void
+    attach_function :rd_kafka_conf_destroy, [:pointer], :void
+    callback :log_cb, [:pointer, :int, :string, :string], :void
+    attach_function :rd_kafka_conf_set_log_cb, [:pointer, :log_cb], :void
+    attach_function :rd_kafka_conf_set_opaque, [:pointer, :pointer], :void
+    callback :stats_cb, [:pointer, :string, :int, :pointer], :int
+    attach_function :rd_kafka_conf_set_stats_cb, [:pointer, :stats_cb], :void
+    callback :error_cb, [:pointer, :int, :string, :pointer], :void
+    attach_function :rd_kafka_conf_set_error_cb, [:pointer, :error_cb], :void
+    attach_function :rd_kafka_rebalance_protocol, [:pointer], :string
+    callback :oauthbearer_token_refresh_cb, [:pointer, :string, :pointer], :void
+    attach_function :rd_kafka_conf_set_oauthbearer_token_refresh_cb, [:pointer, :oauthbearer_token_refresh_cb], :void
+    attach_function :rd_kafka_oauthbearer_set_token, [:pointer, :string, :int64, :pointer, :pointer, :int, :pointer, :int], :int
+    attach_function :rd_kafka_oauthbearer_set_token_failure, [:pointer, :string], :int
+    # Log queue
+    attach_function :rd_kafka_set_log_queue, [:pointer, :pointer], :void
+    attach_function :rd_kafka_queue_get_main, [:pointer], :pointer
+    attach_function :rd_kafka_queue_get_background, [:pointer], :pointer
+
+    # Queue IO Event Support - for fiber scheduler integration
+    # Enables notifications to a custom FD when queue transitions from empty to non-empty
+    # Arguments:
+    # - queue (rd_kafka_queue_t*) - the queue to monitor
+    # - fd (int) - file descriptor to write to (provide your own pipe/eventfd)
+    # - payload (const void*) - data to write to fd
+    # - size (size_t) - size of payload
+    attach_function :rd_kafka_queue_io_event_enable, [:pointer, :int, :pointer, :size_t], :void
+    # Per topic configs
+    attach_function :rd_kafka_topic_conf_new, [], :pointer
+    attach_function :rd_kafka_topic_conf_destroy, [:pointer], :void
+    attach_function :rd_kafka_topic_conf_set, [:pointer, :string, :string, :pointer, :int], :kafka_config_response
+  end
+end

data/lib/rdkafka/bindings/consumer.rb

@@ -0,0 +1,83 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Consumer
+
+    attach_function :rd_kafka_subscribe, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_unsubscribe, [:pointer], :int, blocking: true
+    attach_function :rd_kafka_subscription, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_assign, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_incremental_assign, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_incremental_unassign, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_assignment, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_assignment_lost, [:pointer], :int, blocking: true
+    attach_function :rd_kafka_committed, [:pointer, :pointer, :int], :int, blocking: true
+    attach_function :rd_kafka_commit, [:pointer, :pointer, :bool], :int, blocking: true
+    attach_function :rd_kafka_poll_set_consumer, [:pointer], :void, blocking: true
+    attach_function :rd_kafka_consumer_poll, [:pointer, :int], :pointer, blocking: true
+    # Non-blocking consumer poll variant (does not release GVL)
+    # More efficient for poll(0) calls in fiber schedulers.
+    attach_function :rd_kafka_consumer_poll_nb, :rd_kafka_consumer_poll, [:pointer, :int], :pointer, blocking: false
+    attach_function :rd_kafka_consumer_close, [:pointer], :void, blocking: true
+    attach_function :rd_kafka_queue_get_consumer, [:pointer], :pointer
+    attach_function :rd_kafka_consume_batch_queue, [:pointer, :int, :pointer, :size_t], :ssize_t, blocking: true
+    attach_function :rd_kafka_consume_batch_queue_nb, :rd_kafka_consume_batch_queue, [:pointer, :int, :pointer, :size_t], :ssize_t, blocking: false
+    attach_function :rd_kafka_offsets_store, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_pause_partitions, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_resume_partitions, [:pointer, :pointer], :int, blocking: true
+    attach_function :rd_kafka_seek, [:pointer, :int32, :int64, :int], :int, blocking: true
+    attach_function :rd_kafka_offsets_for_times, [:pointer, :pointer, :int], :int, blocking: true
+    attach_function :rd_kafka_position, [:pointer, :pointer], :int, blocking: true
+    # those two are used for eos support
+    attach_function :rd_kafka_consumer_group_metadata, [:pointer], :pointer, blocking: true
+    attach_function :rd_kafka_consumer_group_metadata_destroy, [:pointer], :void, blocking: true
+
+    # Headers
+    attach_function :rd_kafka_header_get_all, [:pointer, :size_t, :pointer, :pointer, SizePtr], :int
+    attach_function :rd_kafka_message_headers, [:pointer, :pointer], :int
+
+    # Rebalance
+
+    callback :rebalance_cb_function, [:pointer, :int, :pointer, :pointer], :void
+    attach_function :rd_kafka_conf_set_rebalance_cb, [:pointer, :rebalance_cb_function], :void, blocking: true
+
+    RebalanceCallback = FFI::Function.new(
+      :void, [:pointer, :int, :pointer, :pointer]
+    ) do |client_ptr, code, partitions_ptr, opaque_ptr|
+      case code
+      when RD_KAFKA_RESP_ERR__ASSIGN_PARTITIONS
+        if Rdkafka::Bindings.rd_kafka_rebalance_protocol(client_ptr) == "COOPERATIVE"
+          Rdkafka::Bindings.rd_kafka_incremental_assign(client_ptr, partitions_ptr)
+        else
+          Rdkafka::Bindings.rd_kafka_assign(client_ptr, partitions_ptr)
+        end
+      else # RD_KAFKA_RESP_ERR__REVOKE_PARTITIONS or errors
+        if Rdkafka::Bindings.rd_kafka_rebalance_protocol(client_ptr) == "COOPERATIVE"
+          Rdkafka::Bindings.rd_kafka_incremental_unassign(client_ptr, partitions_ptr)
+        else
+          Rdkafka::Bindings.rd_kafka_assign(client_ptr, FFI::Pointer::NULL)
+        end
+      end
+
+      opaque = Rdkafka::Config.opaques[opaque_ptr.to_i]
+      return unless opaque
+
+      tpl = Rdkafka::Consumer::TopicPartitionList.from_native_tpl(partitions_ptr).freeze
+      begin
+        case code
+        when RD_KAFKA_RESP_ERR__ASSIGN_PARTITIONS
+          opaque.call_on_partitions_assigned(tpl)
+        when RD_KAFKA_RESP_ERR__REVOKE_PARTITIONS
+          opaque.call_on_partitions_revoked(tpl)
+        end
+      rescue Exception => err
+        Rdkafka::Config.logger.error("Unhandled exception: #{err.class} - #{err.message}")
+      end
+    end
+
+    # Watermark offsets
+
+    attach_function :rd_kafka_query_watermark_offsets, [:pointer, :string, :int, :pointer, :pointer, :int], :int, blocking: true
+  end
+end

data/lib/rdkafka/bindings/errors.rb

@@ -0,0 +1,15 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    attach_function :rd_kafka_err2name, [:int], :string
+    attach_function :rd_kafka_err2str, [:int], :string
+    attach_function :rd_kafka_error_is_fatal, [:pointer], :int
+    attach_function :rd_kafka_error_is_retriable, [:pointer], :int
+    attach_function :rd_kafka_error_txn_requires_abort, [:pointer], :int
+    attach_function :rd_kafka_error_destroy, [:pointer], :void
+    attach_function :rd_kafka_error_code, [:pointer], :int32
+    attach_function :rd_kafka_error_string, [:pointer], :pointer
+    attach_function :rd_kafka_get_err_descs, [:pointer, :pointer], :void
+  end
+end

data/lib/rdkafka/bindings/handle.rb

@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Handle
+
+    enum :kafka_type, [
+      :rd_kafka_producer,
+      :rd_kafka_consumer
+    ]
+
+    attach_function :rd_kafka_new, [:kafka_type, :pointer, :pointer, :int], :pointer
+
+    attach_function :rd_kafka_destroy, [:pointer], :void
+  end
+end

data/lib/rdkafka/bindings/messages.rb

@@ -0,0 +1,28 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Messages, topics and topic partition lists
+
+    attach_function :rd_kafka_message_destroy, [:pointer], :void
+    attach_function :rd_kafka_message_timestamp, [:pointer, :pointer], :int64
+    attach_function :rd_kafka_message_latency, [:pointer], :int64
+    attach_function :rd_kafka_message_broker_id, [:pointer], :int32
+    attach_function :rd_kafka_message_status, [:pointer], :int
+
+    # Message persistence status (rd_kafka_msg_status_t)
+    RD_KAFKA_MSG_STATUS_NOT_PERSISTED = 0
+    RD_KAFKA_MSG_STATUS_POSSIBLY_PERSISTED = 1
+    RD_KAFKA_MSG_STATUS_PERSISTED = 2
+
+    attach_function :rd_kafka_topic_new, [:pointer, :string, :pointer], :pointer
+    attach_function :rd_kafka_topic_destroy, [:pointer], :pointer
+    attach_function :rd_kafka_topic_name, [:pointer], :string
+
+    attach_function :rd_kafka_topic_partition_list_new, [:int32], :pointer
+    attach_function :rd_kafka_topic_partition_list_add, [:pointer, :string, :int32], :pointer
+    attach_function :rd_kafka_topic_partition_list_set_offset, [:pointer, :string, :int32, :int64], :void
+    attach_function :rd_kafka_topic_partition_list_destroy, [:pointer], :void
+    attach_function :rd_kafka_topic_partition_list_copy, [:pointer], :pointer
+  end
+end

data/lib/rdkafka/bindings/metadata.rb

@@ -0,0 +1,14 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Metadata
+
+    attach_function :rd_kafka_name, [:pointer], :string
+    attach_function :rd_kafka_memberid, [:pointer], :pointer, blocking: true
+    attach_function :rd_kafka_clusterid, [:pointer, :int], :pointer, blocking: true
+    attach_function :rd_kafka_mem_free, [:pointer, :pointer], :void
+    attach_function :rd_kafka_metadata, [:pointer, :int, :pointer, :pointer, :int], :int, blocking: true
+    attach_function :rd_kafka_metadata_destroy, [:pointer], :void, blocking: true
+  end
+end

data/lib/rdkafka/bindings/polling.rb

@@ -0,0 +1,17 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Polling
+
+    attach_function :rd_kafka_flush, [:pointer, :int], :int, blocking: true
+    attach_function :rd_kafka_poll, [:pointer, :int], :int, blocking: true
+    attach_function :rd_kafka_outq_len, [:pointer], :int, blocking: true
+
+    # Non-blocking poll variants (do not release GVL)
+    # These are more efficient for poll(0) calls in fiber schedulers where GVL
+    # release/reacquire overhead is wasteful since we don't expect to wait.
+    # Uses the same underlying C function but with blocking: false to skip GVL release.
+    attach_function :rd_kafka_poll_nb, :rd_kafka_poll, [:pointer, :int], :int, blocking: false
+  end
+end

data/lib/rdkafka/bindings/producer.rb

@@ -0,0 +1,77 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # Producer
+    RD_KAFKA_VTYPE_END = 0
+    RD_KAFKA_VTYPE_TOPIC = 1
+    RD_KAFKA_VTYPE_RKT = 2
+    RD_KAFKA_VTYPE_PARTITION = 3
+    RD_KAFKA_VTYPE_VALUE = 4
+    RD_KAFKA_VTYPE_KEY = 5
+    RD_KAFKA_VTYPE_OPAQUE = 6
+    RD_KAFKA_VTYPE_MSGFLAGS = 7
+    RD_KAFKA_VTYPE_TIMESTAMP = 8
+    RD_KAFKA_VTYPE_HEADER = 9
+    RD_KAFKA_VTYPE_HEADERS = 10
+    RD_KAFKA_PURGE_F_QUEUE = 1
+    RD_KAFKA_PURGE_F_INFLIGHT = 2
+
+    RD_KAFKA_MSG_F_COPY = 0x2
+
+    attach_function :rd_kafka_producev, [:pointer, :varargs], :int, blocking: true
+    attach_function :rd_kafka_purge, [:pointer, :int], :int, blocking: true
+    callback :delivery_cb, [:pointer, :pointer, :pointer], :void
+    attach_function :rd_kafka_conf_set_dr_msg_cb, [:pointer, :delivery_cb], :void
+
+    # Hash mapping partitioner names to their FFI function symbols
+    # @return [Hash{String => Symbol}]
+    PARTITIONERS = %w[random consistent consistent_random murmur2 murmur2_random fnv1a fnv1a_random].each_with_object({}) do |name, hsh|
+      method_name = :"rd_kafka_msg_partitioner_#{name}"
+      attach_function method_name, [:pointer, :pointer, :size_t, :int32, :pointer, :pointer], :int32
+      hsh[name] = method_name
+    end
+
+    # Calculates the partition for a message based on the partitioner
+    #
+    # @param topic_ptr [FFI::Pointer] pointer to the topic handle
+    # @param str [String] the partition key string
+    # @param partition_count [Integer, nil] number of partitions
+    # @param partitioner [String] name of the partitioner to use
+    # @return [Integer] partition number or RD_KAFKA_PARTITION_UA if unassigned
+    # @raise [Rdkafka::Config::ConfigError] when an unknown partitioner is specified
+    def self.partitioner(topic_ptr, str, partition_count, partitioner = "consistent_random")
+      # Return RD_KAFKA_PARTITION_UA(unassigned partition) when partition count is nil/zero.
+      return RD_KAFKA_PARTITION_UA unless partition_count&.nonzero?
+
+      str_ptr = str.empty? ? FFI::MemoryPointer::NULL : FFI::MemoryPointer.from_string(str)
+      method_name = PARTITIONERS.fetch(partitioner) do
+        raise Rdkafka::Config::ConfigError.new("Unknown partitioner: #{partitioner}")
+      end
+
+      public_send(method_name, topic_ptr, str_ptr, partition_key_length(str), partition_count, nil, nil)
+    end
+
+    # Partition key length as the character count (legacy default).
+    #
+    # @param str [String] the partition key string
+    # @return [Integer]
+    def self.partition_key_size(str)
+      str.size
+    end
+
+    # Partition key length as the byte count, which is what librdkafka hashes (UTF-8 bytes copied
+    # into the key pointer) and what other Kafka clients use.
+    #
+    # @param str [String] the partition key string
+    # @return [Integer]
+    def self.partition_key_bytesize(str)
+      str.bytesize
+    end
+
+    # Length of the partition key passed to librdkafka. Aliased to one of the methods above by
+    # `Rdkafka::Config.partitioner_key_uses_bytesize=`, so the per-message path does not check the
+    # setting on every call.
+    singleton_class.alias_method :partition_key_length, :partition_key_size
+  end
+end

data/lib/rdkafka/bindings/structs.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+module Rdkafka
+  module Bindings
+    # FFI struct representing a Kafka message (rd_kafka_message_t)
+    class Message < FFI::Struct
+      layout :err, :int,
+        :rkt, :pointer,
+        :partition, :int32,
+        :payload, :pointer,
+        :len, :size_t,
+        :key, :pointer,
+        :key_len, :size_t,
+        :offset, :int64,
+        :_private, :pointer
+    end
+
+    # FFI struct representing a topic partition (rd_kafka_topic_partition_t)
+    class TopicPartition < FFI::Struct
+      layout :topic, :string,
+        :partition, :int32,
+        :offset, :int64,
+        :metadata, :pointer,
+        :metadata_size, :size_t,
+        :opaque, :pointer,
+        :err, :int,
+        :_private, :pointer
+    end
+
+    # FFI struct representing a topic partition list (rd_kafka_topic_partition_list_t)
+    class TopicPartitionList < FFI::Struct
+      layout :cnt, :int,
+        :size, :int,
+        :elems, :pointer
+    end
+
+    # FFI struct representing a config resource (rd_kafka_ConfigResource_t)
+    # Structs for management of configurations. Each configuration is attached to a resource
+    # and one resource can have many configuration details. Each resource will also have
+    # separate errors results if obtaining configuration was not possible for any reason
+    class ConfigResource < FFI::Struct
+      layout :type, :int,
+        :name, :string
+    end
+
+    # FFI struct for error description (rd_kafka_err_desc)
+    class NativeErrorDesc < FFI::Struct
+      layout :code, :int,
+        :name, :pointer,
+        :desc, :pointer
+    end
+
+    # FFI struct for native error (rd_kafka_error_t)
+    class NativeError < FFI::Struct
+      layout :code, :int32,
+        :errstr, :pointer,
+        :fatal, :u_int8_t,
+        :retriable, :u_int8_t,
+        :txn_requires_abort, :u_int8_t
+    end
+  end
+end