Class: LogStash::Pipeline

Inherits:
BasePipeline show all
Defined in:
lib/logstash/pipeline.rb

Constant Summary collapse

MAX_INFLIGHT_WARN_THRESHOLD =
10_000

Instance Attribute Summary collapse

Attributes inherited from BasePipeline

#filters, #inputs, #outputs

Instance Method Summary collapse

Methods inherited from BasePipeline

#non_reloadable_plugins, #reloadable?, #reloadable_plugins?

Constructor Details

#initialize(pipeline_config, namespaced_metric = nil, agent = nil) ⇒ Pipeline

Returns a new instance of Pipeline.



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
118
119
120
121
# File 'lib/logstash/pipeline.rb', line 89

def initialize(pipeline_config, namespaced_metric = nil, agent = nil)
  super
  open_queue

  @worker_threads = []

  @signal_queue = java.util.concurrent.LinkedBlockingQueue.new

  @drain_queue =  settings.get_value("queue.drain") || settings.get("queue.type") == "memory"


  @events_filtered = java.util.concurrent.atomic.LongAdder.new
  @events_consumed = java.util.concurrent.atomic.LongAdder.new

  @input_threads = []
  # @ready requires thread safety since it is typically polled from outside the pipeline thread
  @ready = Concurrent::AtomicBoolean.new(false)
  @running = Concurrent::AtomicBoolean.new(false)
  @flushing = Concurrent::AtomicReference.new(false)
  @outputs_registered = Concurrent::AtomicBoolean.new(false)
  @worker_shutdown = java.util.concurrent.atomic.AtomicBoolean.new(false)

  # @finished_execution signals that the pipeline thread has finished its execution
  # regardless of any exceptions; it will always be true when the thread completes
  @finished_execution = Concurrent::AtomicBoolean.new(false)

  # @finished_run signals that the run methods called in the pipeline thread was completed
  # without errors and it will NOT be set if the run method exits from an exception; this
  # is by design and necessary for the wait_until_started semantic
  @finished_run = Concurrent::AtomicBoolean.new(false)

  @thread = nil
end

Instance Attribute Details

#events_consumedObject (readonly)

Returns the value of attribute events_consumed.



80
81
82
# File 'lib/logstash/pipeline.rb', line 80

def events_consumed
  @events_consumed
end

#events_filteredObject (readonly)

Returns the value of attribute events_filtered.



80
81
82
# File 'lib/logstash/pipeline.rb', line 80

def events_filtered
  @events_filtered
end

#started_atObject (readonly)

Returns the value of attribute started_at.



80
81
82
# File 'lib/logstash/pipeline.rb', line 80

def started_at
  @started_at
end

#threadObject (readonly)

Returns the value of attribute thread.



80
81
82
# File 'lib/logstash/pipeline.rb', line 80

def thread
  @thread
end

#worker_threadsObject (readonly)

Returns the value of attribute worker_threads.



80
81
82
# File 'lib/logstash/pipeline.rb', line 80

def worker_threads
  @worker_threads
end

Instance Method Details

#clear_pipeline_metricsObject



580
581
582
583
584
585
586
587
588
589
590
591
592
# File 'lib/logstash/pipeline.rb', line 580

def clear_pipeline_metrics
  # TODO(ph): I think the metric should also proxy that call correctly to the collector
  # this will simplify everything since the null metric would simply just do a noop
  collector = metric.collector

  unless collector.nil?
    # selectively reset metrics we don't wish to keep after reloading
    # these include metrics about the plugins and number of processed events
    # we want to keep other metrics like reload counts and error messages
    collector.clear("stats/pipelines/#{pipeline_id}/plugins")
    collector.clear("stats/pipelines/#{pipeline_id}/events")
  end
end

#filter(event, &block) ⇒ Object

for backward compatibility in devutils for the rspec helpers, this method is not used in the pipeline anymore.



503
504
505
506
507
# File 'lib/logstash/pipeline.rb', line 503

def filter(event, &block)
  maybe_setup_out_plugins
  # filter_func returns all filtered events, including cancelled ones
  filter_func([event]).each {|e| block.call(e)}
end

#filter_batch(batch) ⇒ Object



357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
# File 'lib/logstash/pipeline.rb', line 357

def filter_batch(batch)
  filter_func(batch.to_a).each do |e|
    #these are both original and generated events
    batch.merge(e) unless e.cancelled?
  end
  filter_queue_client.add_filtered_metrics(batch.filtered_size)
  @events_filtered.add(batch.filteredSize)
rescue Exception => e
  # Plugins authors should manage their own exceptions in the plugin code
  # but if an exception is raised up to the worker thread they are considered
  # fatal and logstash will not recover from this situation.
  #
  # Users need to check their configuration or see if there is a bug in the
  # plugin.
  @logger.error("Exception in pipelineworker, the pipeline stopped processing new events, please check your filter configuration and restart Logstash.",
                default_logging_keys("exception" => e.message, "backtrace" => e.backtrace))

  raise e
end

#filters?Boolean

Returns:

  • (Boolean)


155
156
157
# File 'lib/logstash/pipeline.rb', line 155

def filters?
  return @filters.any?
end

#finished_execution?Boolean

def initialize

Returns:

  • (Boolean)


123
124
125
# File 'lib/logstash/pipeline.rb', line 123

def finished_execution?
  @finished_execution.true?
end

#flushObject



536
537
538
539
540
541
# File 'lib/logstash/pipeline.rb', line 536

def flush
  if @flushing.compare_and_set(false, true)
    @logger.debug? && @logger.debug("Pushing flush onto pipeline", default_logging_keys)
    @signal_queue.put(FLUSH)
  end
end

#flush_filters(options = {}, &block) ⇒ Object

perform filters flush and yield flushed event to the passed block

Parameters:

  • options (Hash) (defaults to: {})

Options Hash (options):

  • :final (Boolean)

    > true to signal a final shutdown flush



512
513
514
515
516
517
518
# File 'lib/logstash/pipeline.rb', line 512

def flush_filters(options = {}, &block)
  flushers = options[:final] ? @shutdown_flushers : @periodic_flushers

  flushers.each do |flusher|
    flusher.call(options, &block)
  end
end

#flush_filters_to_batch(batch, options = {}) ⇒ Object

perform filters flush into the output queue

Parameters:

  • batch (ReadClient::ReadBatch)
  • options (Hash) (defaults to: {})


555
556
557
558
559
560
561
562
563
564
# File 'lib/logstash/pipeline.rb', line 555

def flush_filters_to_batch(batch, options = {})
  flush_filters(options) do |event|
    unless event.cancelled?
      @logger.debug? and @logger.debug("Pushing flushed events", default_logging_keys(:event => event))
      batch.merge(event)
    end
  end

  @flushing.set(false)
end

#inputworker(plugin) ⇒ Object



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
# File 'lib/logstash/pipeline.rb', line 423

def inputworker(plugin)
  Util::set_thread_name("[#{pipeline_id}]<#{plugin.class.config_name}")
  begin
    plugin.run(wrapped_write_client(plugin.id.to_sym))
  rescue => e
    if plugin.stop?
      @logger.debug("Input plugin raised exception during shutdown, ignoring it.",
                    default_logging_keys(:plugin => plugin.class.config_name, :exception => e.message, :backtrace => e.backtrace))
      return
    end

    # otherwise, report error and restart
    @logger.error(I18n.t("logstash.pipeline.worker-error-debug",
                          default_logging_keys(
                            :plugin => plugin.inspect,
                            :error => e.message,
                            :exception => e.class,
                            :stacktrace => e.backtrace.join("\n"))))

    # Assuming the failure that caused this exception is transient,
    # let's sleep for a bit and execute #run again
    sleep(1)
    retry
  ensure
    plugin.do_close
  end
end

#inspectObject

Sometimes we log stuff that will dump the pipeline which may contain sensitive information (like the raw syntax tree which can contain passwords) We want to hide most of what’s in here



597
598
599
600
601
602
603
604
605
# File 'lib/logstash/pipeline.rb', line 597

def inspect
  {
    :pipeline_id => pipeline_id,
    :settings => settings.inspect,
    :ready => @ready,
    :running => @running,
    :flushing => @flushing
  }
end

#output_batch(batch, output_events_map) ⇒ Object

Take an array of events and send them to the correct output



378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
# File 'lib/logstash/pipeline.rb', line 378

def output_batch(batch, output_events_map)
  # Build a mapping of { output_plugin => [events...]}
  batch.to_a.each do |event|
    # We ask the AST to tell us which outputs to send each event to
    # Then, we stick it in the correct bin
    output_func(event).each do |output|
      output_events_map[output].push(event)
    end
  end
  # Now that we have our output to event mapping we can just invoke each output
  # once with its list of events
  output_events_map.each do |output, events|
    output.multi_receive(events)
    events.clear
  end

  filter_queue_client.add_output_metrics(batch.filtered_size)
end

#plugin_threads_infoObject

flush_filters_to_batch



566
567
568
569
570
# File 'lib/logstash/pipeline.rb', line 566

def plugin_threads_info
  input_threads = @input_threads.select {|t| t.alive? }
  worker_threads = @worker_threads.select {|t| t.alive? }
  (input_threads + worker_threads).map {|t| Util.thread_info(t) }
end

#ready?Boolean

Returns:

  • (Boolean)


127
128
129
# File 'lib/logstash/pipeline.rb', line 127

def ready?
  @ready.value
end

#register_plugin(plugin) ⇒ Plugin

register_plugin simply calls the plugin #register method and catches & logs any error

Parameters:

  • plugin (Plugin)

    the plugin to register

Returns:

  • (Plugin)

    the registered plugin



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

def register_plugin(plugin)
  plugin.register
  plugin
rescue => e
  @logger.error("Error registering plugin", default_logging_keys(:plugin => plugin.inspect, :error => e.message))
  raise e
end

#register_plugins(plugins) ⇒ Object

register_plugins calls #register_plugin on the plugins list and upon exception will call Plugin#do_close on all registered plugins

Parameters:

  • plugins (Array[Plugin])

    the list of plugins to register



268
269
270
271
272
273
274
# File 'lib/logstash/pipeline.rb', line 268

def register_plugins(plugins)
  registered = []
  plugins.each { |plugin| registered << register_plugin(plugin) }
rescue => e
  registered.each(&:do_close)
  raise e
end

#runObject



212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
# File 'lib/logstash/pipeline.rb', line 212

def run
  @started_at = Time.now
  @thread = Thread.current
  Util.set_thread_name("[#{pipeline_id}]-pipeline-manager")

  start_workers

  # Block until all inputs have stopped
  # Generally this happens if SIGINT is sent and `shutdown` is called from an external thread

  transition_to_running
  start_flusher # Launches a non-blocking thread for flush events
  wait_inputs
  transition_to_stopped

  shutdown_flusher
  @logger.debug("Shutting down filter/output workers", default_logging_keys)
  shutdown_workers

  close

  @logger.info("Pipeline has terminated", default_logging_keys)

  # exit code
  return 0
end

#running?Boolean

Returns:

  • (Boolean)


247
248
249
# File 'lib/logstash/pipeline.rb', line 247

def running?
  @running.true?
end

#safe_pipeline_worker_countObject



131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
# File 'lib/logstash/pipeline.rb', line 131

def safe_pipeline_worker_count
  default = settings.get_default("pipeline.workers")
  pipeline_workers = settings.get("pipeline.workers") #override from args "-w 8" or config
  safe_filters, unsafe_filters = @filters.partition(&:threadsafe?)
  plugins = unsafe_filters.collect { |f| f.config_name }

  return pipeline_workers if unsafe_filters.empty?

  if settings.set?("pipeline.workers")
    if pipeline_workers > 1
      @logger.warn("Warning: Manual override - there are filters that might not work with multiple worker threads", default_logging_keys(:worker_threads => pipeline_workers, :filters => plugins))
    end
  else
    # user did not specify a worker thread count
    # warn if the default is multiple
    if default > 1
      @logger.warn("Defaulting pipeline worker threads to 1 because there are some filters that might not work with multiple worker threads",
                   default_logging_keys(:count_was => default, :filters => plugins))
      return 1 # can't allow the default value to propagate if there are unsafe filters
    end
  end
  pipeline_workers
end

#shutdown(&before_stop) ⇒ Object

initiate the pipeline shutdown sequence this method is intended to be called from outside the pipeline thread

Parameters:

  • before_stop (Proc)

    code block called before performing stop operation on input plugins



454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
# File 'lib/logstash/pipeline.rb', line 454

def shutdown(&before_stop)
  # shutdown can only start once the pipeline has completed its startup.
  # avoid potential race condition between the startup sequence and this
  # shutdown method which can be called from another thread at any time
  sleep(0.1) while !ready?

  # TODO: should we also check against calling shutdown multiple times concurrently?

  before_stop.call if block_given?

  stop_inputs

  # We make this call blocking, so we know for sure when the method return the shtudown is
  # stopped
  wait_for_workers
  clear_pipeline_metrics
end

#shutdown_flusherObject



532
533
534
# File 'lib/logstash/pipeline.rb', line 532

def shutdown_flusher
  @flusher_thread.join
end

#shutdown_workersObject

After ‘shutdown` is called from an external thread this is called from the main thread to tell the worker threads to stop and then block until they’ve fully stopped This also stops all filter and output plugins



488
489
490
491
492
493
494
495
496
497
498
499
# File 'lib/logstash/pipeline.rb', line 488

def shutdown_workers
  @logger.debug("Setting shutdown", default_logging_keys)
  @worker_shutdown.set(true)

  @worker_threads.each do |t|
    @logger.debug("Shutdown waiting for worker thread" , default_logging_keys(:thread => t.inspect))
    t.join
  end

  @filters.each(&:do_close)
  @outputs.each(&:do_close)
end

#stalling_threads_infoObject



572
573
574
575
576
577
578
# File 'lib/logstash/pipeline.rb', line 572

def stalling_threads_info
  plugin_threads_info
    .reject {|t| t["blocked_on"] } # known benign blocking statuses
    .each {|t| t.delete("backtrace") }
    .each {|t| t.delete("blocked_on") }
    .each {|t| t.delete("status") }
end

#startObject



159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
# File 'lib/logstash/pipeline.rb', line 159

def start
  # Since we start lets assume that the metric namespace is cleared
  # this is useful in the context of pipeline reloading
  collect_stats
  collect_dlq_stats

  @logger.info("Starting pipeline", default_logging_keys(
    "pipeline.workers" => settings.get("pipeline.workers"),
    "pipeline.batch.size" => settings.get("pipeline.batch.size"),
    "pipeline.batch.delay" => settings.get("pipeline.batch.delay")))

  @finished_execution.make_false
  @finished_run.make_false

  @thread = Thread.new do
    begin
      LogStash::Util.set_thread_name("pipeline.#{pipeline_id}")
      run
      @finished_run.make_true
    rescue => e
      close
      @logger.error("Pipeline aborted due to error", default_logging_keys(:exception => e, :backtrace => e.backtrace))
    ensure
      @finished_execution.make_true
    end
  end

  status = wait_until_started

  if status
    @logger.info("Pipeline started successfully", default_logging_keys)
  end

  status
end

#start_flusherObject



520
521
522
523
524
525
526
527
528
529
530
# File 'lib/logstash/pipeline.rb', line 520

def start_flusher
  # Invariant to help detect improper initialization
  raise "Attempted to start flusher on a stopped pipeline!" if stopped?

  @flusher_thread = Thread.new do
    while Stud.stoppable_sleep(5, 0.1) { stopped? }
      flush
      break if stopped?
    end
  end
end

#start_input(plugin) ⇒ Object



419
420
421
# File 'lib/logstash/pipeline.rb', line 419

def start_input(plugin)
  @input_threads << Thread.new { inputworker(plugin) }
end

#start_inputsObject



401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
# File 'lib/logstash/pipeline.rb', line 401

def start_inputs
  moreinputs = []
  @inputs.each do |input|
    if input.threadable && input.threads > 1
      (input.threads - 1).times do |i|
        moreinputs << input.clone
      end
    end
  end
  @inputs += moreinputs

  # first make sure we can register all input plugins
  register_plugins(@inputs)

  # then after all input plugins are successfully registered, start them
  @inputs.each { |input| start_input(input) }
end

#start_workersObject



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
# File 'lib/logstash/pipeline.rb', line 276

def start_workers
  @worker_threads.clear # In case we're restarting the pipeline
  @outputs_registered.make_false
  begin
    maybe_setup_out_plugins

    pipeline_workers = safe_pipeline_worker_count
    batch_size = settings.get("pipeline.batch.size")
    batch_delay = settings.get("pipeline.batch.delay")

    max_inflight = batch_size * pipeline_workers

    config_metric = metric.namespace([:stats, :pipelines, pipeline_id.to_s.to_sym, :config])
    config_metric.gauge(:workers, pipeline_workers)
    config_metric.gauge(:batch_size, batch_size)
    config_metric.gauge(:batch_delay, batch_delay)
    config_metric.gauge(:config_reload_automatic, settings.get("config.reload.automatic"))
    config_metric.gauge(:config_reload_interval, settings.get("config.reload.interval"))
    config_metric.gauge(:dead_letter_queue_enabled, dlq_enabled?)
    config_metric.gauge(:dead_letter_queue_path, dlq_writer.get_path.to_absolute_path.to_s) if dlq_enabled?

    if max_inflight > MAX_INFLIGHT_WARN_THRESHOLD
      @logger.warn("CAUTION: Recommended inflight events max exceeded! Logstash will run with up to #{max_inflight} events in memory in your current configuration. If your message sizes are large this may cause instability with the default heap size. Please consider setting a non-standard heap size, changing the batch size (currently #{batch_size}), or changing the number of pipeline workers (currently #{pipeline_workers})", default_logging_keys)
    end

    pipeline_workers.times do |t|
      thread = Thread.new(batch_size, batch_delay, self) do |_b_size, _b_delay, _pipeline|
        Util.set_thread_name("[#{pipeline_id}]>worker#{t}")
        _pipeline.worker_loop(_b_size, _b_delay)
      end
      @worker_threads << thread
    end

    # inputs should be started last, after all workers
    begin
      start_inputs
    rescue => e
      # if there is any exception in starting inputs, make sure we shutdown workers.
      # exception will already by logged in start_inputs
      shutdown_workers
      raise e
    end
  ensure
    # it is important to guarantee @ready to be true after the startup sequence has been completed
    # to potentially unblock the shutdown method which may be waiting on @ready to proceed
    @ready.make_true
  end
end

#stop_inputsObject



479
480
481
482
483
# File 'lib/logstash/pipeline.rb', line 479

def stop_inputs
  @logger.debug("Stopping inputs", default_logging_keys)
  @inputs.each(&:do_stop)
  @logger.debug("Stopped inputs", default_logging_keys)
end

#stopped?Boolean

Returns:

  • (Boolean)


251
252
253
# File 'lib/logstash/pipeline.rb', line 251

def stopped?
  @running.false?
end

#transition_to_runningObject

def run



239
240
241
# File 'lib/logstash/pipeline.rb', line 239

def transition_to_running
  @running.make_true
end

#transition_to_stoppedObject



243
244
245
# File 'lib/logstash/pipeline.rb', line 243

def transition_to_stopped
  @running.make_false
end

#uptimeInteger

Calculate the uptime in milliseconds

Returns:

  • (Integer)

    Uptime in milliseconds, 0 if the pipeline is not started



546
547
548
549
# File 'lib/logstash/pipeline.rb', line 546

def uptime
  return 0 if started_at.nil?
  ((Time.now.to_f - started_at.to_f) * 1000.0).to_i
end

#wait_for_workersObject

def shutdown



472
473
474
475
476
477
# File 'lib/logstash/pipeline.rb', line 472

def wait_for_workers
  @worker_threads.each do |t|
    t.join
    @logger.debug("Worker terminated", default_logging_keys(:thread => t.inspect))
  end
end

#wait_inputsObject



397
398
399
# File 'lib/logstash/pipeline.rb', line 397

def wait_inputs
  @input_threads.each(&:join)
end

#wait_until_startedObject



195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
# File 'lib/logstash/pipeline.rb', line 195

def wait_until_started
  while true do
    if @finished_run.true?
      # it completed run without exception
      return true
    elsif thread.nil? || !thread.alive?
      # some exception occured and the thread is dead
      return false
    elsif running?
      # fully initialized and running
      return true
    else
      sleep 0.01
    end
  end
end

#worker_loop(batch_size, batch_delay) ⇒ Object

Main body of what a worker thread does Repeatedly takes batches off the queue, filters, then outputs them



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
# File 'lib/logstash/pipeline.rb', line 327

def worker_loop(batch_size, batch_delay)
  filter_queue_client.set_batch_dimensions(batch_size, batch_delay)
  output_events_map = Hash.new { |h, k| h[k] = [] }
  while true
    signal = @signal_queue.poll || NO_SIGNAL

    batch = filter_queue_client.read_batch.to_java # metrics are started in read_batch
    batch_size = batch.filteredSize
    if batch_size > 0
      @events_consumed.add(batch_size)
      filter_batch(batch)
    end
    flush_filters_to_batch(batch, :final => false) if signal.flush?
    if batch.filteredSize > 0
      output_batch(batch, output_events_map)
      filter_queue_client.close_batch(batch)
    end
    # keep break at end of loop, after the read_batch operation, some pipeline specs rely on this "final read_batch" before shutdown.
    break if (@worker_shutdown.get && !draining_queue?)
  end

  # we are shutting down, queue is drained if it was required, now  perform a final flush.
  # for this we need to create a new empty batch to contain the final flushed events
  batch = filter_queue_client.to_java.newBatch
  filter_queue_client.start_metrics(batch) # explicitly call start_metrics since we dont do a read_batch here
  flush_filters_to_batch(batch, :final => true)
  output_batch(batch, output_events_map)
  filter_queue_client.close_batch(batch)
end