Class: Rdkafka::Producer

Inherits:
Object
  • Object
show all
Includes:
Helpers::Metadata, Helpers::OAuth, Helpers::Time
Defined in:
lib/rdkafka/producer.rb,
lib/rdkafka/producer/delivery_handle.rb,
lib/rdkafka/producer/delivery_report.rb,
lib/rdkafka/producer/partitions_count_cache.rb

Overview

A producer for Kafka messages. To create a producer set up a Config and call producer on that.

Defined Under Namespace

Classes: DeliveryHandle, DeliveryReport, PartitionsCountCache, TopicHandleCreationError

Constant Summary collapse

@@partitions_count_cache = 
PartitionsCountCache.new

Instance Attribute Summary collapse

Class Method Summary collapse

Instance Method Summary collapse

Methods included from Helpers::Metadata

#metadata

Methods included from Helpers::OAuth

#oauthbearer_set_token, #oauthbearer_set_token_failure

Methods included from Helpers::Time

#monotonic_now, #monotonic_now_ms

Constructor Details

#initialize(native_kafka, partitioner) ⇒ Producer

Returns a new instance of Producer.

Parameters:

  • native_kafka (NativeKafka)
  • partitioner (String, nil) —

    name of the partitioner we want to use or nil to use the "consistent_random" default



56
57
58
59
60
61
62
63
64
 
# File 'lib/rdkafka/producer.rb', line 56

def initialize(native_kafka, partitioner)
  @topics_refs_map = {}
  @topics_configs = {}
  @native_kafka = native_kafka
  @partitioner = partitioner || "consistent_random"

  # Makes sure, that native kafka gets closed before it gets GCed by Ruby
  ObjectSpace.define_finalizer(self, native_kafka.finalizer)
end

Instance Attribute Details

#delivery_callback ⇒ Proc?

Returns the current delivery callback, by default this is nil.

Returns:

  • (Proc, nil) 


44
45
46
 
# File 'lib/rdkafka/producer.rb', line 44

def delivery_callback
  @delivery_callback
end

#delivery_callback_arity ⇒ Integer? (readonly)

Returns the number of arguments accepted by the callback, by default this is nil.

Returns:

  • (Integer, nil) 


50
51
52
 
# File 'lib/rdkafka/producer.rb', line 50

def delivery_callback_arity
  @delivery_callback_arity
end

Class Method Details

.partitions_count_cache ⇒ Rdkafka::Producer::PartitionsCountCache

Note:

It is critical to remember, that not all users may have statistics callbacks enabled, hence we should not make assumption that this cache is always updated from the stats.

Global (process wide) partitions cache. We use it to store number of topics partitions, either from the librdkafka statistics (if enabled) or via direct inline calls every now and then. Since the partitions count can only grow and should be same for all consumers and producers, we can use a global cache as long as we ensure that updates only move up.

Returns:



21
22
23
 
# File 'lib/rdkafka/producer.rb', line 21

def self.partitions_count_cache
  @@partitions_count_cache
end

.partitions_count_cache=(partitions_count_cache) ⇒ Object

Parameters:



26
27
28
 
# File 'lib/rdkafka/producer.rb', line 26

def self.partitions_count_cache=(partitions_count_cache)
  @@partitions_count_cache = partitions_count_cache
end

Instance Method Details

#arity(callback) ⇒ Integer

Figures out the arity of a given block/method

Parameters:

  • callback (#call, Proc)

Returns:

  • (Integer) —

    arity of the provided block/method



516
517
518
519
520
 
# File 'lib/rdkafka/producer.rb', line 516

def arity(callback)
  return callback.arity if callback.respond_to?(:arity)

  callback.method(:call).arity
end

#call_delivery_callback(delivery_report, delivery_handle) ⇒ Object

Calls (if registered) the delivery callback

Parameters:



499
500
501
502
503
504
505
506
507
508
509
510
 
# File 'lib/rdkafka/producer.rb', line 499

def call_delivery_callback(delivery_report, delivery_handle)
  return unless @delivery_callback

  case @delivery_callback_arity
  when 0
    @delivery_callback.call
  when 1
    @delivery_callback.call(delivery_report)
  else
    @delivery_callback.call(delivery_report, delivery_handle)
  end
end

#close ⇒ Object

Close this producer and wait for the internal poll queue to empty.



164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
 
# File 'lib/rdkafka/producer.rb', line 164

def close
  return if closed?
  ObjectSpace.undefine_finalizer(self)

  @native_kafka.close do
    # We need to remove the topics references objects before we destroy the producer,
    # otherwise they would leak out
    @topics_refs_map.each_value do |refs|
      refs.each_value do |ref|
        Rdkafka::Bindings.rd_kafka_topic_destroy(ref)
      end
    end
  end

  @topics_refs_map.clear
end

#closed? ⇒ Boolean

Whether this producer has closed

Returns:

  • (Boolean) 


182
183
184
 
# File 'lib/rdkafka/producer.rb', line 182

def closed?
  @native_kafka.closed?
end

#enable_background_queue_io_events(fd, payload = "\x01") ⇒ nil

Enable IO event notifications for background events

Parameters:

  • fd (Integer) —

    file descriptor to signal (from IO.pipe or eventfd)

  • payload (String) (defaults to: "\x01") —

    data to write to fd (default: "\x01")

Returns:

  • (nil)

Raises:



148
149
150
 
# File 'lib/rdkafka/producer.rb', line 148

def enable_background_queue_io_events(fd, payload = "\x01")
  @native_kafka.enable_background_queue_io_events(fd, payload)
end

#enable_queue_io_events(fd, payload = "\x01") ⇒ nil

Enable IO event notifications for fiber scheduler integration When delivery confirmations arrive, librdkafka will write to your FD

Parameters:

  • fd (Integer) —

    file descriptor to signal (from IO.pipe or eventfd)

  • payload (String) (defaults to: "\x01") —

    data to write to fd (default: "\x01")

Returns:

  • (nil)

Raises:



139
140
141
 
# File 'lib/rdkafka/producer.rb', line 139

def enable_queue_io_events(fd, payload = "\x01")
  @native_kafka.enable_main_queue_io_events(fd, payload)
end

#events_poll_nb_each {|count| ... } ⇒ nil

Note:

This method holds the inner lock until the queue is empty or :stop is returned. Other producer operations (produce, close, etc.) will wait until this method returns.

Note:

This method is thread-safe as it uses @native_kafka.with_inner synchronization

Polls for events in a non-blocking loop, yielding the count after each iteration.

This method processes delivery callbacks in a single GVL/mutex session, which is more efficient than repeated individual polls. It uses non-blocking polls internally (no GVL release between polls).

Yields the count of events processed after each poll iteration, allowing the caller to implement timeout or other termination logic by returning :stop.

Examples:

Drain all pending callbacks

producer.events_poll_nb_each { |_count| }

With timeout control

deadline = monotonic_now + timeout_ms
producer.events_poll_nb_each do |_count|
  :stop if monotonic_now >= deadline
end

Yields:

  • (count) —

    Called after each poll iteration

Yield Parameters:

  • count (Integer) —

    Number of events processed in this iteration

Yield Returns:

  • (Symbol, Object) —

    Return :stop to break the loop, any other value continues

Returns:

  • (nil)

Raises:



295
296
297
298
299
300
301
302
303
304
305
 
# File 'lib/rdkafka/producer.rb', line 295

def events_poll_nb_each
  closed_producer_check(__method__)

  @native_kafka.with_inner do |inner|
    loop do
      count = Rdkafka::Bindings.rd_kafka_poll_nb(inner, 0)
      break if count.zero?
      break if yield(count) == :stop
    end
  end
end

#flush(timeout_ms = Defaults::PRODUCER_FLUSH_TIMEOUT_MS) ⇒ Boolean

Note:

We raise an exception for other errors because based on the librdkafka docs, there should be no other errors.

Note:

For timed_out we do not raise an error to keep it backwards compatible

Wait until all outstanding producer requests are completed, with the given timeout in seconds. Call this before closing a producer to ensure delivery of all messages.

Parameters:

  • timeout_ms (Integer) (defaults to: Defaults::PRODUCER_FLUSH_TIMEOUT_MS) —

    how long should we wait for flush of all messages

Returns:

  • (Boolean) —

    true if no more data and all was flushed, false in case there are still outgoing messages after the timeout



197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
 
# File 'lib/rdkafka/producer.rb', line 197

def flush(timeout_ms = Defaults::PRODUCER_FLUSH_TIMEOUT_MS)
  closed_producer_check(__method__)

  code = nil

  @native_kafka.with_inner do |inner|
    code = Rdkafka::Bindings.rd_kafka_flush(inner, timeout_ms)
  end

  # Early skip not to build the error message
  return true if code.zero?

  error = Rdkafka::RdkafkaError.new(code)

  return false if error.code == :timed_out

  raise(error)
end

#name ⇒ String

Returns producer name.

Returns:

  • (String) —

    producer name



126
127
128
129
130
 
# File 'lib/rdkafka/producer.rb', line 126

def name
  @name ||= @native_kafka.with_inner do |inner|
    ::Rdkafka::Bindings.rd_kafka_name(inner)
  end
end

#partition_count(topic) ⇒ Integer

Note:

If 'allow.auto.create.topics' is set to true in the broker, the topic will be auto-created after returning nil.

Note:

We cache the partition count for a given topic for given time. If statistics are enabled for any producer or consumer, it will take precedence over per instance fetching.

This prevents us in case someone uses partition_key from querying for the count with each message. Instead we query at most once every 30 seconds at most if we have a valid partition count or every 5 seconds in case we were not able to obtain number of partitions.

Partition count for a given topic.

Parameters:

  • topic (String) —

    The topic name.

Returns:

  • (Integer) —

    partition count for a given topic or -1 if it could not be obtained.



321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
 
# File 'lib/rdkafka/producer.rb', line 321

def partition_count(topic)
  closed_producer_check(__method__)

  self.class.partitions_count_cache.get(topic) do
    topic_metadata = nil

    @native_kafka.with_inner do |inner|
      topic_metadata = ::Rdkafka::Metadata.new(inner, topic).topics&.first
    end

    topic_metadata ? topic_metadata[:partition_count] : Rdkafka::Bindings::RD_KAFKA_PARTITION_UA
  rescue Rdkafka::RdkafkaError => e
    # A missing topic is an expected, cacheable outcome (the topic may be auto-created on
    # produce, or simply not exist yet). Returning RD_KAFKA_PARTITION_UA from inside the block
    # caches it, so we don't re-run a blocking metadata query on every produce(partition_key:)
    # to that topic. Anything else is a real error and is re-raised.
    raise(e) unless e.code == :unknown_topic_or_part

    Rdkafka::Bindings::RD_KAFKA_PARTITION_UA
  end
end

#produce(topic:, payload: nil, key: nil, partition: nil, partition_key: nil, timestamp: nil, headers: nil, label: nil, topic_config: EMPTY_HASH, partitioner: @partitioner) ⇒ DeliveryHandle

Produces a message to a Kafka topic. The message is added to rdkafka's queue, call wait on the returned delivery handle to make sure it is delivered.

When no partition is specified the underlying Kafka library picks a partition based on the key. If no key is specified, a random partition will be used. When a timestamp is provided this is used instead of the auto-generated timestamp.

Parameters:

  • topic (String) —

    The topic to produce to

  • payload (String, nil) (defaults to: nil)
  • key (String, nil) (defaults to: nil)
  • partition (Integer, nil) (defaults to: nil) —

    Optional partition to produce to

  • partition_key (String, nil) (defaults to: nil) —

    Optional partition key based on which partition assignment can happen

  • timestamp (Time, Integer, nil) (defaults to: nil) —

    Optional timestamp of this message. Integer timestamp is in milliseconds since Jan 1 1970.

  • headers (Hash{String => String, Array<String>}) (defaults to: nil) —

    Optional message headers. Values can be either a single string or an array of strings to support duplicate headers per KIP-82

  • label (Object, nil) (defaults to: nil) —

    a label that can be assigned when producing a message that will be part of the delivery handle and the delivery report

  • topic_config (Hash) (defaults to: EMPTY_HASH) —

    topic config for given message dispatch. Allows to send messages to topics with different configuration

  • partitioner (String) (defaults to: @partitioner) —

    name of the partitioner to use

Returns:

  • (DeliveryHandle) —

    Delivery handle that can be used to wait for the result of producing this message

Raises:

  • (RdkafkaError) —

    When adding the message to rdkafka's queue failed



362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
 
# File 'lib/rdkafka/producer.rb', line 362

def produce(
  topic:,
  payload: nil,
  key: nil,
  partition: nil,
  partition_key: nil,
  timestamp: nil,
  headers: nil,
  label: nil,
  topic_config: EMPTY_HASH,
  partitioner: @partitioner
)
  closed_producer_check(__method__)

  # Start by checking and converting the input

  # Get payload length
  payload_size = if payload.nil?
    0
  else
    payload.bytesize
  end

  # Get key length
  key_size = if key.nil?
    0
  else
    key.bytesize
  end

  topic_config_hash = topic_config.hash

  # Checks if we have the rdkafka topic reference object ready. It saves us on object
  # allocation and allows to use custom config on demand.
  set_topic_config(topic, topic_config, topic_config_hash) unless @topics_refs_map.dig(topic, topic_config_hash)
  topic_ref = @topics_refs_map.dig(topic, topic_config_hash)

  if partition_key
    partition_count = partition_count(topic)

    # Check if there are no overrides for the partitioner and use the default one only when
    # no per-topic is present.
    selected_partitioner = @topics_configs.dig(topic, topic_config_hash, :partitioner) || partitioner

    # If the topic is not present, set to -1
    if partition_count.positive?
      partition = Rdkafka::Bindings.partitioner(
        topic_ref,
        partition_key,
        partition_count,
        selected_partitioner
      )
    end
  end

  # If partition is nil, use RD_KAFKA_PARTITION_UA to let librdafka set the partition randomly or
  # based on the key when present.
  partition ||= Rdkafka::Bindings::RD_KAFKA_PARTITION_UA

  # If timestamp is nil use 0 and let Kafka set one. If an integer or time use it.
  raw_timestamp = if timestamp.nil?
    0
  elsif timestamp.is_a?(Integer)
    timestamp
  elsif timestamp.is_a?(Time)
    (timestamp.to_i * 1000) + (timestamp.usec / 1000)
  else
    raise TypeError.new("Timestamp has to be nil, an Integer or a Time")
  end

  delivery_handle = DeliveryHandle.new
  delivery_handle.label = label
  delivery_handle.topic = topic
  delivery_handle[:pending] = true
  delivery_handle[:response] = Rdkafka::Bindings::RD_KAFKA_PARTITION_UA
  delivery_handle[:partition] = Rdkafka::Bindings::RD_KAFKA_PARTITION_UA
  delivery_handle[:offset] = Rdkafka::Bindings::RD_KAFKA_PARTITION_UA
  DeliveryHandle.register(delivery_handle)

  begin
    args = [
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_RKT, :pointer, topic_ref,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_MSGFLAGS, :int, Rdkafka::Bindings::RD_KAFKA_MSG_F_COPY,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_VALUE, :buffer_in, payload, :size_t, payload_size,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_KEY, :buffer_in, key, :size_t, key_size,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_PARTITION, :int32, partition,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_TIMESTAMP, :int64, raw_timestamp,
      :int, Rdkafka::Bindings::RD_KAFKA_VTYPE_OPAQUE, :pointer, delivery_handle
    ]

    headers&.each do |key0, value0|
      key = key0.to_s
      if value0.is_a?(Array)
        # Handle array of values per KIP-82
        value0.each do |value|
          value = value.to_s
          args << :int << Rdkafka::Bindings::RD_KAFKA_VTYPE_HEADER
          args << :string << key
          args << :pointer << value
          args << :size_t << value.bytesize
        end
      else
        # Handle single value
        value = value0.to_s
        args << :int << Rdkafka::Bindings::RD_KAFKA_VTYPE_HEADER
        args << :string << key
        args << :pointer << value
        args << :size_t << value.bytesize
      end
    end

    args << :int << Rdkafka::Bindings::RD_KAFKA_VTYPE_END

    # Produce the message
    response = @native_kafka.with_inner do |inner|
      Rdkafka::Bindings.rd_kafka_producev(
        inner,
        *args
      )
    end

    # Raise error if the produce call was not successful
    if response != Rdkafka::Bindings::RD_KAFKA_RESP_ERR_NO_ERROR
      raise RdkafkaError.new(response)
    end
  rescue Exception
    DeliveryHandle.remove(delivery_handle.to_ptr.address)
    raise
  end

  delivery_handle
end

#purge ⇒ Object

Purges the outgoing queue and releases all resources.

Useful when closing the producer with outgoing messages to unstable clusters or when for any other reasons waiting cannot go on anymore. This purges both the queue and all the inflight requests + updates the delivery handles statuses so they can be materialized into purge_queue errors.



222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
 
# File 'lib/rdkafka/producer.rb', line 222

def purge
  closed_producer_check(__method__)

  code = nil

  @native_kafka.with_inner do |inner|
    code = Bindings.rd_kafka_purge(
      inner,
      Bindings::RD_KAFKA_PURGE_F_QUEUE | Bindings::RD_KAFKA_PURGE_F_INFLIGHT
    )
  end

  code.zero? || raise(Rdkafka::RdkafkaError.new(code))

  # Wait for the purge to affect everything
  sleep(Defaults::PRODUCER_PURGE_SLEEP_INTERVAL_MS / 1000.0) until flush(Defaults::PRODUCER_PURGE_FLUSH_TIMEOUT_MS)

  true
end

#queue_size ⇒ Integer Also known as: queue_length

Note:

This method is thread-safe as it uses the @native_kafka.with_inner synchronization

Returns the number of messages and requests waiting to be sent to the broker as well as delivery reports queued for the application.

This provides visibility into the producer's internal queue depth, useful for:

  • Monitoring producer backpressure
  • Implementing custom flow control
  • Debugging message delivery issues
  • Graceful shutdown logic (wait until queue is empty)

Examples:

producer.queue_size #=> 42

Returns:

  • (Integer) —

    the number of messages in the queue

Raises:



258
259
260
261
262
263
264
 
# File 'lib/rdkafka/producer.rb', line 258

def queue_size
  closed_producer_check(__method__)

  @native_kafka.with_inner do |inner|
    Rdkafka::Bindings.rd_kafka_outq_len(inner)
  end
end

#set_topic_config(topic, config, config_hash) ⇒ Object

Note:

It is not allowed to re-set the same topic config twice because of the underlying librdkafka caching

Sets alternative set of configuration details that can be set per topic

Parameters:

  • topic (String) —

    The topic name

  • config (Hash) —

    config we want to use per topic basis

  • config_hash (Integer) —

    hash of the config. We expect it here instead of computing it, because it is already computed during the retrieval attempt in the #produce flow.



74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
 
# File 'lib/rdkafka/producer.rb', line 74

def set_topic_config(topic, config, config_hash)
  # Ensure lock on topic reference just in case
  @native_kafka.with_inner do |inner|
    @topics_refs_map[topic] ||= {}
    @topics_configs[topic] ||= {}

    return if @topics_configs[topic].key?(config_hash)

    # If config is empty, we create an empty reference that will be used with defaults
    rd_topic_config = if config.empty?
      nil
    else
      Rdkafka::Bindings.rd_kafka_topic_conf_new.tap do |topic_config|
        config.each do |key, value|
          error_buffer = FFI::MemoryPointer.new(:char, 256)
          result = Rdkafka::Bindings.rd_kafka_topic_conf_set(
            topic_config,
            key.to_s,
            value.to_s,
            error_buffer,
            256
          )

          unless result == :config_ok
            # rd_kafka_topic_new has not been called yet, so librdkafka has not taken
            # ownership of the config. Destroy it ourselves before raising, otherwise the
            # partially built rd_kafka_topic_conf_t leaks. (On the rd_kafka_topic_new path
            # librdkafka frees the conf on both success and failure, so we never destroy it
            # there.)
            Rdkafka::Bindings.rd_kafka_topic_conf_destroy(topic_config)
            raise Config::ConfigError.new(error_buffer.read_string)
          end
        end
      end
    end

    topic_handle = Bindings.rd_kafka_topic_new(inner, topic, rd_topic_config)

    raise TopicHandleCreationError.new("Error creating topic handle for topic #{topic}") if topic_handle.null?

    @topics_configs[topic][config_hash] = config
    @topics_refs_map[topic][config_hash] = topic_handle
  end
end

#start ⇒ Object

Note:

Not needed to run unless explicit start was disabled

Starts the native Kafka polling thread and kicks off the init polling



121
122
123
 
# File 'lib/rdkafka/producer.rb', line 121

def start
  @native_kafka.start
end