Class: Fluent::Plugin::Buffer

Inherits:
Base
  • Object
show all
Includes:
OwnedByMixin, UniqueId::Mixin, MonitorMixin
Defined in:
lib/fluent/plugin/buffer/chunk.rb,
lib/fluent/plugin/buffer.rb,
lib/fluent/plugin/buffer/file_chunk.rb,
lib/fluent/plugin/buffer/memory_chunk.rb,
lib/fluent/plugin/buffer/file_single_chunk.rb

Overview

fluent/plugin/buffer is already loaded

Direct Known Subclasses

FileBuffer, FileSingleBuffer, MemoryBuffer

Defined Under Namespace

Classes: BufferChunkOverflowError, BufferError, BufferOverflowError, Chunk, FileChunk, FileSingleChunk, MemoryChunk, Metadata, ShouldRetry

Constant Summary collapse

MINIMUM_APPEND_ATTEMPT_RECORDS = 
10
DEFAULT_CHUNK_LIMIT_SIZE =

8MB

8 * 1024 * 1024
DEFAULT_TOTAL_LIMIT_SIZE =

512MB, same with v0.12 (BufferedOutput + buf_memory: 64 x 8MB)

512 * 1024 * 1024
DEFAULT_CHUNK_FULL_THRESHOLD = 
0.95
STATS_KEYS = 
[
  'stage_length',
  'stage_byte_size',
  'queue_length',
  'queue_byte_size',
  'available_buffer_space_ratios',
  'total_queued_size',
  'oldest_timekey',
  'newest_timekey'
]

Constants included from Configurable

Configurable::CONFIG_TYPE_REGISTRY

Instance Attribute Summary collapse

Attributes inherited from Base

#under_plugin_development

Instance Method Summary collapse

Methods included from UniqueId::Mixin

#dump_unique_id_hex, #generate_unique_id

Methods included from OwnedByMixin

#log, #owner, #owner=

Methods inherited from Base

#after_shutdown, #after_shutdown?, #after_start, #after_started?, #before_shutdown, #before_shutdown?, #called_in_test?, #closed?, #configured?, #context_router, #context_router=, #fluentd_worker_id, #has_router?, #inspect, #multi_workers_ready?, #plugin_root_dir, #reloadable_plugin?, #shutdown, #shutdown?, #started?, #stop, #stopped?, #string_safe_encoding, #terminated?

Methods included from SystemConfig::Mixin

#system_config, #system_config_override

Methods included from Configurable

#config, #configure_proxy_generate, #configured_section_create, included, lookup_type, register_type

Constructor Details

#initialize ⇒ Buffer

Returns a new instance of Buffer.



160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
 
# File 'lib/fluent/plugin/buffer.rb', line 160

def initialize
  super

  @chunk_limit_size = nil
  @total_limit_size = nil
  @queue_limit_length = nil
  @chunk_limit_records = nil

  @stage = {}    #=> Hash (metadata -> chunk) : not flushed yet
  @queue = []    #=> Array (chunks)           : already flushed (not written)
  @dequeued = {} #=> Hash (unique_id -> chunk): already written (not purged)
  @queued_num = {} # metadata => int (number of queued chunks)
  @dequeued_num = {} # metadata => int (number of dequeued chunks)

  @stage_size = @queue_size = 0
  @timekeys = Hash.new(0)
  @enable_update_timekeys = false
  @mutex = Mutex.new
end

Instance Attribute Details

#dequeued ⇒ Object (readonly)

Returns the value of attribute dequeued.



158
159
160
 
# File 'lib/fluent/plugin/buffer.rb', line 158

def dequeued
  @dequeued
end

#queue ⇒ Object (readonly)

Returns the value of attribute queue.



158
159
160
 
# File 'lib/fluent/plugin/buffer.rb', line 158

def queue
  @queue
end

#queue_size ⇒ Object

for tests



157
158
159
 
# File 'lib/fluent/plugin/buffer.rb', line 157

def queue_size
  @queue_size
end

#queued_num ⇒ Object (readonly)

Returns the value of attribute queued_num.



158
159
160
 
# File 'lib/fluent/plugin/buffer.rb', line 158

def queued_num
  @queued_num
end

#stage ⇒ Object (readonly)

Returns the value of attribute stage.



158
159
160
 
# File 'lib/fluent/plugin/buffer.rb', line 158

def stage
  @stage
end

#stage_size ⇒ Object

for tests



157
158
159
 
# File 'lib/fluent/plugin/buffer.rb', line 157

def stage_size
  @stage_size
end

Instance Method Details

#chunk_size_full?(chunk) ⇒ Boolean

Returns:

  • (Boolean) 


573
574
575
 
# File 'lib/fluent/plugin/buffer.rb', line 573

def chunk_size_full?(chunk)
  chunk.bytesize >= @chunk_limit_size * @chunk_full_threshold || (@chunk_limit_records && chunk.size >= @chunk_limit_records * @chunk_full_threshold)
end

#chunk_size_over?(chunk) ⇒ Boolean

Returns:

  • (Boolean) 


569
570
571
 
# File 'lib/fluent/plugin/buffer.rb', line 569

def chunk_size_over?(chunk)
  chunk.bytesize > @chunk_limit_size || (@chunk_limit_records && chunk.size > @chunk_limit_records)
end

#clear_queue! ⇒ Object 



551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
 
# File 'lib/fluent/plugin/buffer.rb', line 551

def clear_queue!
  log.on_trace { log.trace "clearing queue", instance: self.object_id }

  synchronize do
    until @queue.empty?
      begin
        q = @queue.shift
        log.trace("purging a chunk in queue"){ {id: dump_unique_id_hex(chunk.unique_id), bytesize: chunk.bytesize, size: chunk.size} }
        q.purge
      rescue => e
        log.error "unexpected error while clearing buffer queue", error_class: e.class, error: e
        log.error_backtrace
      end
    end
    @queue_size = 0
  end
end

#close ⇒ Object 



212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
 
# File 'lib/fluent/plugin/buffer.rb', line 212

def close
  super
  synchronize do
    log.debug "closing buffer", instance: self.object_id
    @dequeued.each_pair do |chunk_id, chunk|
      chunk.close
    end
    until @queue.empty?
      @queue.shift.close
    end
    @stage.each_pair do |metadata, chunk|
      chunk.close
    end
  end
end

#configure(conf) ⇒ Object 



184
185
186
187
188
189
190
 
# File 'lib/fluent/plugin/buffer.rb', line 184

def configure(conf)
  super

  unless @queue_limit_length.nil?
    @total_limit_size = @chunk_limit_size * @queue_limit_length
  end
end

#dequeue_chunk ⇒ Object 



489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
 
# File 'lib/fluent/plugin/buffer.rb', line 489

def dequeue_chunk
  return nil if @queue.empty?
  log.on_trace { log.trace "dequeueing a chunk", instance: self.object_id }

  synchronize do
    chunk = @queue.shift

    # this buffer is dequeued by other thread just before "synchronize" in this thread
    return nil unless chunk

    @dequeued[chunk.unique_id] = chunk
    @queued_num[chunk.metadata] -= 1 # BUG if nil, 0 or subzero
    @dequeued_num[chunk.metadata] ||= 0
    @dequeued_num[chunk.metadata] += 1
    log.trace "chunk dequeued", instance: self.object_id, metadata: chunk.metadata
    chunk
  end
end

#enable_update_timekeys ⇒ Object 



192
193
194
 
# File 'lib/fluent/plugin/buffer.rb', line 192

def enable_update_timekeys
  @enable_update_timekeys = true
end

#enqueue_all(force_enqueue = false) ⇒ Object

At flush_at_shutdown, all staged chunks should be enqueued for buffer flush. Set true to force_enqueue for it.



467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
 
# File 'lib/fluent/plugin/buffer.rb', line 467

def enqueue_all(force_enqueue = false)
  log.on_trace { log.trace "enqueueing all chunks in buffer", instance: self.object_id }
  update_timekeys if @enable_update_timekeys

  if block_given?
    synchronize{ @stage.keys }.each do |metadata|
      return if !force_enqueue && queue_full?
      # NOTE: The following line might cause data race depending on Ruby implementations except CRuby
      # cf. https://github.com/fluent/fluentd/pull/1721#discussion_r146170251
      chunk = @stage[metadata]
      next unless chunk
      v = yield metadata, chunk
      enqueue_chunk(metadata) if v
    end
  else
    synchronize{ @stage.keys }.each do |metadata|
      return if !force_enqueue && queue_full?
      enqueue_chunk(metadata)
    end
  end
end

#enqueue_chunk(metadata) ⇒ Object 



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
 
# File 'lib/fluent/plugin/buffer.rb', line 412

def enqueue_chunk(metadata)
  log.on_trace { log.trace "enqueueing chunk", instance: self.object_id, metadata: metadata }

  chunk = synchronize do
    @stage.delete(metadata)
  end
  return nil unless chunk

  chunk.synchronize do
    synchronize do
      if chunk.empty?
        chunk.close
      else
        chunk.metadata.seq = 0 # metadata.seq should be 0 for counting @queued_num
        @queue << chunk
        @queued_num[metadata] = @queued_num.fetch(metadata, 0) + 1
        chunk.enqueued!
      end
      bytesize = chunk.bytesize
      @stage_size -= bytesize
      @queue_size += bytesize
    end
  end
  nil
end

#enqueue_unstaged_chunk(chunk) ⇒ Object 



438
439
440
441
442
443
444
445
446
447
448
449
450
451
 
# File 'lib/fluent/plugin/buffer.rb', line 438

def enqueue_unstaged_chunk(chunk)
  log.on_trace { log.trace "enqueueing unstaged chunk", instance: self.object_id, metadata: chunk.metadata }

  synchronize do
    chunk.synchronize do
      metadata = chunk.metadata
      metadata.seq = 0 # metadata.seq should be 0 for counting @queued_num
      @queue << chunk
      @queued_num[metadata] = @queued_num.fetch(metadata, 0) + 1
      chunk.enqueued!
    end
    @queue_size += chunk.bytesize
  end
end

#generate_chunk(metadata) ⇒ Object

Raises:

  • (NotImplementedError) 


249
250
251
 
# File 'lib/fluent/plugin/buffer.rb', line 249

def generate_chunk(metadata)
  raise NotImplementedError, "Implement this method in child class"
end

#metadata(timekey: nil, tag: nil, variables: nil) ⇒ Object

Keep this method for existing code



258
259
260
 
# File 'lib/fluent/plugin/buffer.rb', line 258

def metadata(timekey: nil, tag: nil, variables: nil)
  Metadata.new(timekey, tag, variables)
end

#new_metadata(timekey: nil, tag: nil, variables: nil) ⇒ Object 



253
254
255
 
# File 'lib/fluent/plugin/buffer.rb', line 253

def new_metadata(timekey: nil, tag: nil, variables: nil)
  Metadata.new(timekey, tag, variables)
end

#persistent? ⇒ Boolean

Returns:

  • (Boolean) 


180
181
182
 
# File 'lib/fluent/plugin/buffer.rb', line 180

def persistent?
  false
end

#purge_chunk(chunk_id) ⇒ Object 



522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
 
# File 'lib/fluent/plugin/buffer.rb', line 522

def purge_chunk(chunk_id)
  metadata = nil
  synchronize do
    chunk = @dequeued.delete(chunk_id)
    return nil unless chunk # purged by other threads

    metadata = chunk.metadata
    log.on_trace { log.trace "purging a chunk", instance: self.object_id, chunk_id: dump_unique_id_hex(chunk_id), metadata: metadata }

    begin
      bytesize = chunk.bytesize
      chunk.purge
      @queue_size -= bytesize
    rescue => e
      log.error "failed to purge buffer chunk", chunk_id: dump_unique_id_hex(chunk_id), error_class: e.class, error: e
      log.error_backtrace
    end

    @dequeued_num[chunk.metadata] -= 1
    if metadata && !@stage[metadata] && (!@queued_num[metadata] || @queued_num[metadata] < 1) && @dequeued_num[metadata].zero?
      @queued_num.delete(metadata)
      @dequeued_num.delete(metadata)
    end
    log.trace "chunk purged", instance: self.object_id, chunk_id: dump_unique_id_hex(chunk_id), metadata: metadata
  end

  nil
end

#queue_full? ⇒ Boolean

Returns:

  • (Boolean) 


394
395
396
 
# File 'lib/fluent/plugin/buffer.rb', line 394

def queue_full?
  synchronize { @queue.size } >= @queued_chunks_limit_size
end

#queued?(metadata = nil, optimistic: false) ⇒ Boolean

Returns:

  • (Boolean) 


402
403
404
405
406
407
408
409
410
 
# File 'lib/fluent/plugin/buffer.rb', line 402

def queued?(metadata = nil, optimistic: false)
  if optimistic
    optimistic_queued?(metadata)
  else
    synchronize do
      optimistic_queued?(metadata)
    end
  end
end

#queued_records ⇒ Object 



398
399
400
 
# File 'lib/fluent/plugin/buffer.rb', line 398

def queued_records
  synchronize { @queue.reduce(0){|r, chunk| r + chunk.size } }
end

#resume ⇒ Object

TODO: for back pressure feature def used?(ratio)

@total_limit_size * ratio > @stage_size + @queue_size

end

Raises:

  • (NotImplementedError) 


244
245
246
247
 
# File 'lib/fluent/plugin/buffer.rb', line 244

def resume
  # return {}, []
  raise NotImplementedError, "Implement this method in child class"
end

#start ⇒ Object 



196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
 
# File 'lib/fluent/plugin/buffer.rb', line 196

def start
  super

  @stage, @queue = resume
  @stage.each_pair do |metadata, chunk|
    @stage_size += chunk.bytesize
  end
  @queue.each do |chunk|
    @queued_num[chunk.metadata] ||= 0
    @queued_num[chunk.metadata] += 1
    @queue_size += chunk.bytesize
  end
  update_timekeys
  log.debug "buffer started", instance: self.object_id, stage_size: @stage_size, queue_size: @queue_size
end

#statistics ⇒ Object 



767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
 
# File 'lib/fluent/plugin/buffer.rb', line 767

def statistics
  stage_size, queue_size = @stage_size, @queue_size
  buffer_space = 1.0 - ((stage_size + queue_size * 1.0) / @total_limit_size)
  stats = {
    'stage_length' => @stage.size,
    'stage_byte_size' => stage_size,
    'queue_length' => @queue.size,
    'queue_byte_size' => queue_size,
    'available_buffer_space_ratios' => (buffer_space * 100).round(1),
    'total_queued_size' => stage_size + queue_size,
  }

  tkeys = timekeys
  if (m = tkeys.min)
    stats['oldest_timekey'] = m
  end
  if (m = tkeys.max)
    stats['newest_timekey'] = m
  end

  { 'buffer' => stats }
end

#storable? ⇒ Boolean

Returns:

  • (Boolean) 


235
236
237
 
# File 'lib/fluent/plugin/buffer.rb', line 235

def storable?
  @total_limit_size > @stage_size + @queue_size
end

#takeback_chunk(chunk_id) ⇒ Object 



508
509
510
511
512
513
514
515
516
517
518
519
520
 
# File 'lib/fluent/plugin/buffer.rb', line 508

def takeback_chunk(chunk_id)
  log.on_trace { log.trace "taking back a chunk", instance: self.object_id, chunk_id: dump_unique_id_hex(chunk_id) }

  synchronize do
    chunk = @dequeued.delete(chunk_id)
    return false unless chunk # already purged by other thread
    @queue.unshift(chunk)
    log.trace "chunk taken back", instance: self.object_id, chunk_id: dump_unique_id_hex(chunk_id), metadata: chunk.metadata
    @queued_num[chunk.metadata] += 1 # BUG if nil
    @dequeued_num[chunk.metadata] -= 1
  end
  true
end

#terminate ⇒ Object 



228
229
230
231
232
233
 
# File 'lib/fluent/plugin/buffer.rb', line 228

def terminate
  super
  @dequeued = @stage = @queue = @queued_num = nil
  @stage_size = @queue_size = 0
  @timekeys.clear
end

#timekeys ⇒ Object 



262
263
264
 
# File 'lib/fluent/plugin/buffer.rb', line 262

def timekeys
  @timekeys.keys
end

#update_timekeys ⇒ Object 



453
454
455
456
457
458
459
460
461
462
463
464
 
# File 'lib/fluent/plugin/buffer.rb', line 453

def update_timekeys
  synchronize do
    chunks = @stage.values
    chunks.concat(@queue)
    @timekeys = chunks.each_with_object({}) do |chunk, keys|
      if chunk.metadata && chunk.metadata.timekey
        t = chunk.metadata.timekey
        keys[t] = keys.fetch(t, 0) + 1
      end
    end
  end
end

#write(metadata_and_data, format: nil, size: nil, enqueue: false) ⇒ Object

metadata MUST have consistent object_id for each variation data MUST be Array of serialized events, or EventStream metadata_and_data MUST be a hash of { metadata => data }

Raises:



269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
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
 
# File 'lib/fluent/plugin/buffer.rb', line 269

def write(metadata_and_data, format: nil, size: nil, enqueue: false)
  return if metadata_and_data.size < 1
  raise BufferOverflowError, "buffer space has too many data" unless storable?

  log.on_trace { log.trace "writing events into buffer", instance: self.object_id, metadata_size: metadata_and_data.size }

  operated_chunks = []
  unstaged_chunks = {} # metadata => [chunk, chunk, ...]
  chunks_to_enqueue = []
  staged_bytesizes_by_chunk = {}

  begin
    # sort metadata to get lock of chunks in same order with other threads
    metadata_and_data.keys.sort.each do |metadata|
      data = metadata_and_data[metadata]
      write_once(metadata, data, format: format, size: size) do |chunk, adding_bytesize|
        chunk.mon_enter # add lock to prevent to be committed/rollbacked from other threads
        operated_chunks << chunk
        if chunk.staged?
          #
          # https://github.com/fluent/fluentd/issues/2712
          # write_once is supposed to write to a chunk only once
          # but this block **may** run multiple times from write_step_by_step and previous write may be rollbacked
          # So we should be counting the stage_size only for the last successful write
          #
          staged_bytesizes_by_chunk[chunk] = adding_bytesize
        elsif chunk.unstaged?
          unstaged_chunks[metadata] ||= []
          unstaged_chunks[metadata] << chunk
        end
      end
    end

    return if operated_chunks.empty?

    # Now, this thread acquires many locks of chunks... getting buffer-global lock causes dead lock.
    # Any operations needs buffer-global lock (including enqueueing) should be done after releasing locks.

    first_chunk = operated_chunks.shift
    # Following commits for other chunks also can finish successfully if the first commit operation
    # finishes without any exceptions.
    # In most cases, #commit just requires very small disk spaces, so major failure reason are
    # permission errors, disk failures and other permanent(fatal) errors.
    begin
      first_chunk.commit
      if enqueue || first_chunk.unstaged? || chunk_size_full?(first_chunk)
        chunks_to_enqueue << first_chunk
      end
      first_chunk.mon_exit
    rescue
      operated_chunks.unshift(first_chunk)
      raise
    end

    errors = []
    # Buffer plugin estimates there's no serious error cause: will commit for all chunks eigher way
    operated_chunks.each do |chunk|
      begin
        chunk.commit
        if enqueue || chunk.unstaged? || chunk_size_full?(chunk)
          chunks_to_enqueue << chunk
        end
        chunk.mon_exit
      rescue => e
        chunk.rollback
        chunk.mon_exit
        errors << e
      end
    end

    # All locks about chunks are released.

    #
    # Now update the stage, stage_size with proper locking
    # FIX FOR stage_size miscomputation - https://github.com/fluent/fluentd/issues/2712
    #
    staged_bytesizes_by_chunk.each do |chunk, bytesize|
      chunk.synchronize do
        synchronize { @stage_size += bytesize }
        log.on_trace { log.trace { "chunk #{chunk.path} size_added: #{bytesize} new_size: #{chunk.bytesize}" } }
      end
    end

    chunks_to_enqueue.each do |c|
      if c.staged? && (enqueue || chunk_size_full?(c))
        m = c.metadata
        enqueue_chunk(m)
        if unstaged_chunks[m]
          u = unstaged_chunks[m].pop
          u.synchronize do
            if u.unstaged? && !chunk_size_full?(u)
              # `u.metadata.seq` and `m.seq` can be different but Buffer#enqueue_chunk expect them to be the same value
              u.metadata.seq = 0
              synchronize {
                @stage[m] = u.staged!
                @stage_size += u.bytesize
              }
            end
          end
        end
      elsif c.unstaged?
        enqueue_unstaged_chunk(c)
      else
        # previously staged chunk is already enqueued, closed or purged.
        # no problem.
      end
    end

    operated_chunks.clear if errors.empty?

    if errors.size > 0
      log.warn "error occurs in committing chunks: only first one raised", errors: errors.map(&:class)
      raise errors.first
    end
  ensure
    operated_chunks.each do |chunk|
      chunk.rollback rescue nil # nothing possible to do for #rollback failure
      if chunk.unstaged?
        chunk.purge rescue nil # to prevent leakage of unstaged chunks
      end
      chunk.mon_exit rescue nil # this may raise ThreadError for chunks already committed
    end
  end
end

#write_once(metadata, data, format: nil, size: nil, &block) ⇒ Object

write once into a chunk

  1. append whole data into existing chunk

  2. commit it & return unless chunk_size_over?

  3. enqueue existing chunk & retry whole method if chunk was not empty

  4. go to step_by_step writing



585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
 
# File 'lib/fluent/plugin/buffer.rb', line 585

def write_once(metadata, data, format: nil, size: nil, &block)
  return if data.empty?

  stored = false
  adding_bytesize = nil

  chunk = synchronize { @stage[metadata] ||= generate_chunk(metadata).staged! }
  enqueue_chunk_before_retry = false
  chunk.synchronize do
    # retry this method if chunk is already queued (between getting chunk and entering critical section)
    raise ShouldRetry unless chunk.staged?

    empty_chunk = chunk.empty?

    original_bytesize = chunk.bytesize
    begin
      if format
        serialized = format.call(data)
        chunk.concat(serialized, size ? size.call : data.size)
      else
        chunk.append(data, compress: @compress)
      end
      adding_bytesize = chunk.bytesize - original_bytesize

      if chunk_size_over?(chunk)
        if format && empty_chunk
          if chunk.bytesize > @chunk_limit_size
            log.warn "chunk bytes limit exceeds for an emitted event stream: #{adding_bytesize}bytes"
          else
            log.warn "chunk size limit exceeds for an emitted event stream: #{chunk.size}records"
          end
        end
        chunk.rollback

        if format && !empty_chunk
          # Event streams should be appended into a chunk at once
          # as far as possible, to improve performance of formatting.
          # Event stream may be a MessagePackEventStream. We don't want to split it into
          # 2 or more chunks (except for a case that the event stream is larger than chunk limit).
          enqueue_chunk_before_retry = true
          raise ShouldRetry
        end
      else
        stored = true
      end
    rescue
      chunk.rollback
      raise
    end

    if stored
      block.call(chunk, adding_bytesize)
    end
  end

  unless stored
    # try step-by-step appending if data can't be stored into existing a chunk in non-bulk mode
    #
    # 1/10 size of original event stream (splits_count == 10) seems enough small
    # to try emitting events into existing chunk.
    # it does not matter to split event stream into very small splits, because chunks have less
    # overhead to write data many times (even about file buffer chunks).
    write_step_by_step(metadata, data, format, 10, &block)
  end
rescue ShouldRetry
  enqueue_chunk(metadata) if enqueue_chunk_before_retry
  retry
end

#write_step_by_step(metadata, data, format, splits_count, &block) ⇒ Object

  1. split event streams into many (10 -> 100 -> 1000 -> …) chunks

  2. append splits into the staged chunks as much as possible

  3. create unstaged chunk and append rest splits -> repeat it for all splits



661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
 
# File 'lib/fluent/plugin/buffer.rb', line 661

def write_step_by_step(metadata, data, format, splits_count, &block)
  splits = []
  if splits_count > data.size
    splits_count = data.size
  end
  slice_size = if data.size % splits_count == 0
                 data.size / splits_count
               else
                 data.size / (splits_count - 1)
               end
  slice_origin = 0
  while slice_origin < data.size
    splits << data.slice(slice_origin, slice_size)
    slice_origin += slice_size
  end

  # This method will append events into the staged chunk at first.
  # Then, will generate chunks not staged (not queued) to append rest data.
  staged_chunk_used = false
  modified_chunks = []
  modified_metadata = metadata
  get_next_chunk = ->(){
    c = if staged_chunk_used
          # Staging new chunk here is bad idea:
          # Recovering whole state including newly staged chunks is much harder than current implementation.
          modified_metadata = modified_metadata.dup_next
          generate_chunk(modified_metadata)
        else
          synchronize { @stage[modified_metadata] ||= generate_chunk(modified_metadata).staged! }
        end
    modified_chunks << c
    c
  }

  writing_splits_index = 0
  enqueue_chunk_before_retry = false

  while writing_splits_index < splits.size
    chunk = get_next_chunk.call
    chunk.synchronize do
      raise ShouldRetry unless chunk.writable?
      staged_chunk_used = true if chunk.staged?

      original_bytesize = chunk.bytesize
      begin
        while writing_splits_index < splits.size
          split = splits[writing_splits_index]
          if format
            chunk.concat(format.call(split), split.size)
          else
            chunk.append(split, compress: @compress)
          end

          if chunk_size_over?(chunk) # split size is larger than difference between size_full? and size_over?
            chunk.rollback

            if split.size == 1 && original_bytesize == 0
              big_record_size = format ? format.call(split).bytesize : split.first.bytesize
              raise BufferChunkOverflowError, "a #{big_record_size}bytes record is larger than buffer chunk limit size"
            end

            if chunk_size_full?(chunk) || split.size == 1
              enqueue_chunk_before_retry = true
            else
              splits_count *= 10
            end

            raise ShouldRetry
          end

          writing_splits_index += 1

          if chunk_size_full?(chunk)
            break
          end
        end
      rescue
        chunk.purge if chunk.unstaged? # unstaged chunk will leak unless purge it
        raise
      end

      block.call(chunk, chunk.bytesize - original_bytesize)
    end
  end
rescue ShouldRetry
  modified_chunks.each do |mc|
    mc.rollback rescue nil
    if mc.unstaged?
      mc.purge rescue nil
    end
  end
  enqueue_chunk(metadata) if enqueue_chunk_before_retry
  retry
end