Class: LogStash::Pipeline
- Inherits:
-
Object
- Object
- LogStash::Pipeline
- Defined in:
- lib/logstash/pipeline.rb
Constant Summary collapse
- DEFAULT_SETTINGS =
{ :default_pipeline_workers => LogStash::Config::CpuCoreStrategy.maximum, :pipeline_batch_size => 125, :pipeline_batch_delay => 5, # in milliseconds :flush_interval => 5, # in seconds :flush_timeout_interval => 60 # in seconds }
- MAX_INFLIGHT_WARN_THRESHOLD =
in seconds
10_000
Instance Attribute Summary collapse
-
#events_consumed ⇒ Object
readonly
Returns the value of attribute events_consumed.
-
#events_filtered ⇒ Object
readonly
Returns the value of attribute events_filtered.
-
#filters ⇒ Object
readonly
Returns the value of attribute filters.
-
#inputs ⇒ Object
readonly
Returns the value of attribute inputs.
-
#logger ⇒ Object
readonly
Returns the value of attribute logger.
-
#outputs ⇒ Object
readonly
Returns the value of attribute outputs.
-
#pipeline_id ⇒ Object
readonly
Returns the value of attribute pipeline_id.
-
#reporter ⇒ Object
readonly
Returns the value of attribute reporter.
-
#worker_threads ⇒ Object
readonly
Returns the value of attribute worker_threads.
Instance Method Summary collapse
- #configure(setting, value) ⇒ Object
- #default_output_workers ⇒ Object
-
#filter(event, &block) ⇒ Object
for backward compatibility in devutils for the rspec helpers, this method is not used in the pipeline anymore.
- #filter_batch(batch) ⇒ Object
- #filters? ⇒ Boolean
- #flush ⇒ Object
-
#flush_filters(options = {}, &block) ⇒ Object
perform filters flush and yeild flushed event to the passed block.
-
#flush_filters_to_batch(batch, options = {}) ⇒ Object
perform filters flush into the output queue.
- #inflight_batches_synchronize ⇒ Object
-
#initialize(config_str, settings = {}) ⇒ Pipeline
constructor
A new instance of Pipeline.
- #inputworker(plugin) ⇒ Object
-
#output_batch(batch) ⇒ Object
Take an array of events and send them to the correct output.
- #plugin(plugin_type, name, *args) ⇒ Object
-
#plugin_threads_info ⇒ Object
flush_filters_to_output!.
-
#ready? ⇒ Boolean
def initialize.
- #run ⇒ Object
- #running? ⇒ Boolean
- #safe_pipeline_worker_count ⇒ Object
- #set_current_thread_inflight_batch(batch) ⇒ Object
-
#shutdown(&before_stop) ⇒ Object
initiate the pipeline shutdown sequence this method is intended to be called from outside the pipeline thread.
- #shutdown_flusher ⇒ Object
-
#shutdown_workers ⇒ Object
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.
- #stalling_threads_info ⇒ Object
- #start_flusher ⇒ Object
- #start_input(plugin) ⇒ Object
- #start_inputs ⇒ Object
- #start_workers ⇒ Object
- #stopped? ⇒ Boolean
- #take_batch(batch_size, batch_delay) ⇒ Object
-
#transition_to_running ⇒ Object
def run.
- #transition_to_stopped ⇒ Object
- #wait_inputs ⇒ Object
-
#worker_loop(batch_size, batch_delay) ⇒ Object
Main body of what a worker thread does Repeatedly takes batches off the queu, filters, then outputs them.
Constructor Details
#initialize(config_str, settings = {}) ⇒ Pipeline
Returns a new instance of Pipeline.
31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 |
# File 'lib/logstash/pipeline.rb', line 31 def initialize(config_str, settings = {}) @pipeline_id = settings[:pipeline_id] || self.object_id @logger = Cabin::Channel.get(LogStash) @settings = DEFAULT_SETTINGS.clone settings.each {|setting, value| configure(setting, value) } @reporter = LogStash::PipelineReporter.new(@logger, self) @inputs = nil @filters = nil @outputs = nil @worker_threads = [] grammar = LogStashConfigParser.new @config = grammar.parse(config_str) if @config.nil? raise LogStash::ConfigurationError, grammar.failure_reason end # This will compile the config to ruby and evaluate the resulting code. # The code will initialize all the plugins and define the # filter and output methods. code = @config.compile # The config code is hard to represent as a log message... # So just print it. @logger.debug? && @logger.debug("Compiled pipeline code:\n#{code}") begin eval(code) rescue => e raise end @input_queue = LogStash::Util::WrappedSynchronousQueue.new @events_filtered = Concurrent::AtomicFixnum.new(0) @events_consumed = Concurrent::AtomicFixnum.new(0) # We generally only want one thread at a time able to access pop/take/poll operations # from this queue. We also depend on this to be able to block consumers while we snapshot # in-flight buffers @input_queue_pop_mutex = Mutex.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) end |
Instance Attribute Details
#events_consumed ⇒ Object (readonly)
Returns the value of attribute events_consumed.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def events_consumed @events_consumed end |
#events_filtered ⇒ Object (readonly)
Returns the value of attribute events_filtered.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def events_filtered @events_filtered end |
#filters ⇒ Object (readonly)
Returns the value of attribute filters.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def filters @filters end |
#inputs ⇒ Object (readonly)
Returns the value of attribute inputs.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def inputs @inputs end |
#logger ⇒ Object (readonly)
Returns the value of attribute logger.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def logger @logger end |
#outputs ⇒ Object (readonly)
Returns the value of attribute outputs.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def outputs @outputs end |
#pipeline_id ⇒ Object (readonly)
Returns the value of attribute pipeline_id.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def pipeline_id @pipeline_id end |
#reporter ⇒ Object (readonly)
Returns the value of attribute reporter.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def reporter @reporter end |
#worker_threads ⇒ Object (readonly)
Returns the value of attribute worker_threads.
20 21 22 |
# File 'lib/logstash/pipeline.rb', line 20 def worker_threads @worker_threads end |
Instance Method Details
#configure(setting, value) ⇒ Object
81 82 83 |
# File 'lib/logstash/pipeline.rb', line 81 def configure(setting, value) @settings[setting] = value end |
#default_output_workers ⇒ Object
409 410 411 |
# File 'lib/logstash/pipeline.rb', line 409 def default_output_workers @settings[:pipeline_workers] || @settings[:default_pipeline_workers] end |
#filter(event, &block) ⇒ Object
for backward compatibility in devutils for the rspec helpers, this method is not used in the pipeline anymore.
415 416 417 418 |
# File 'lib/logstash/pipeline.rb', line 415 def filter(event, &block) # filter_func returns all filtered events, including cancelled ones filter_func(event).each { |e| block.call(e) } end |
#filter_batch(batch) ⇒ Object
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 |
# File 'lib/logstash/pipeline.rb', line 253 def filter_batch(batch) batch.reduce([]) do |acc,e| if e.is_a?(LogStash::Event) filtered = filter_func(e) filtered.each {|fe| acc << fe unless fe.cancelled?} end acc end 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.", "exception" => e, "backtrace" => e.backtrace) raise end |
#filters? ⇒ Boolean
115 116 117 |
# File 'lib/logstash/pipeline.rb', line 115 def filters? return @filters.any? end |
#flush ⇒ Object
448 449 450 451 452 453 |
# File 'lib/logstash/pipeline.rb', line 448 def flush if @flushing.compare_and_set(false, true) @logger.debug? && @logger.debug("Pushing flush onto pipeline") @input_queue.push(LogStash::FLUSH) end end |
#flush_filters(options = {}, &block) ⇒ Object
perform filters flush and yeild flushed event to the passed block
424 425 426 427 428 429 430 |
# File 'lib/logstash/pipeline.rb', line 424 def flush_filters( = {}, &block) flushers = [:final] ? @shutdown_flushers : @periodic_flushers flushers.each do |flusher| flusher.call(, &block) end end |
#flush_filters_to_batch(batch, options = {}) ⇒ Object
perform filters flush into the output queue
458 459 460 461 462 463 464 465 466 467 |
# File 'lib/logstash/pipeline.rb', line 458 def flush_filters_to_batch(batch, = {}) flush_filters() do |event| unless event.cancelled? @logger.debug? and @logger.debug("Pushing flushed events", :event => event) batch << event end end @flushing.set(false) end |
#inflight_batches_synchronize ⇒ Object
297 298 299 300 301 |
# File 'lib/logstash/pipeline.rb', line 297 def inflight_batches_synchronize @input_queue_pop_mutex.synchronize do yield(@inflight_batches) end end |
#inputworker(plugin) ⇒ Object
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 |
# File 'lib/logstash/pipeline.rb', line 328 def inputworker(plugin) LogStash::Util::set_thread_name("[#{pipeline_id}]<#{plugin.class.config_name}") begin plugin.run(@input_queue) rescue => e if plugin.stop? @logger.debug("Input plugin raised exception during shutdown, ignoring it.", :plugin => plugin.class.config_name, :exception => e, :backtrace => e.backtrace) return end # otherwise, report error and restart if @logger.debug? @logger.error(I18n.t("logstash.pipeline.worker-error-debug", :plugin => plugin.inspect, :error => e.to_s, :exception => e.class, :stacktrace => e.backtrace.join("\n"))) else @logger.error(I18n.t("logstash.pipeline.worker-error", :plugin => plugin.inspect, :error => e)) end # 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 |
#output_batch(batch) ⇒ Object
Take an array of events and send them to the correct output
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 |
# File 'lib/logstash/pipeline.rb', line 274 def output_batch(batch) # Build a mapping of { output_plugin => [events...]} outputs_events = batch.reduce(Hash.new { |h, k| h[k] = [] }) do |acc, 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 should never return anything other than an Array but we have lots of legacy specs # that monkeypatch it and return nil. We can deprecate "|| []" after fixing these specs outputs_for_event = output_func(event) || [] outputs_for_event.each { |output| acc[output] << event } acc end # Now that we have our output to event mapping we can just invoke each output # once with its list of events outputs_events.each { |output, events| output.multi_receive(events) } end |
#plugin(plugin_type, name, *args) ⇒ Object
397 398 399 400 401 402 403 404 405 406 407 |
# File 'lib/logstash/pipeline.rb', line 397 def plugin(plugin_type, name, *args) args << {} if args.empty? klass = LogStash::Plugin.lookup(plugin_type, name) if plugin_type == "output" LogStash::OutputDelegator.new(@logger, klass, default_output_workers, *args) else klass.new(*args) end end |
#plugin_threads_info ⇒ Object
flush_filters_to_output!
469 470 471 472 473 |
# File 'lib/logstash/pipeline.rb', line 469 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| LogStash::Util.thread_info(t) } end |
#ready? ⇒ Boolean
def initialize
77 78 79 |
# File 'lib/logstash/pipeline.rb', line 77 def ready? @ready.value end |
#run ⇒ Object
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 |
# File 'lib/logstash/pipeline.rb', line 119 def run LogStash::Util.set_thread_name("[#{pipeline_id}]-pipeline-manager") @logger.terminal(LogStash::Util::DefaultsPrinter.print(@settings)) start_workers @logger.info("Pipeline started") @logger.terminal("Logstash startup completed") # 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 @logger.info("Input plugins stopped! Will shutdown filter/output workers.") shutdown_flusher shutdown_workers @logger.info("Pipeline shutdown complete.") @logger.terminal("Logstash shutdown completed") # exit code return 0 end |
#running? ⇒ Boolean
156 157 158 |
# File 'lib/logstash/pipeline.rb', line 156 def running? @running.true? end |
#safe_pipeline_worker_count ⇒ Object
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 |
# File 'lib/logstash/pipeline.rb', line 85 def safe_pipeline_worker_count default = DEFAULT_SETTINGS[:default_pipeline_workers] thread_count = @settings[:pipeline_workers] #override from args "-w 8" or config safe_filters, unsafe_filters = @filters.partition(&:threadsafe?) if unsafe_filters.any? plugins = unsafe_filters.collect { |f| f.class.config_name } case thread_count when nil # 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", :count_was => default, :filters => plugins) end 1 # can't allow the default value to propagate if there are unsafe filters when 0, 1 1 else @logger.warn("Warning: Manual override - there are filters that might not work with multiple worker threads", :worker_threads => thread_count, :filters => plugins) thread_count # allow user to force this even if there are unsafe filters end else thread_count || default end end |
#set_current_thread_inflight_batch(batch) ⇒ Object
293 294 295 |
# File 'lib/logstash/pipeline.rb', line 293 def set_current_thread_inflight_batch(batch) @inflight_batches[Thread.current] = batch end |
#shutdown(&before_stop) ⇒ Object
initiate the pipeline shutdown sequence this method is intended to be called from outside the pipeline thread
363 364 365 366 367 368 369 370 371 372 373 374 375 376 |
# File 'lib/logstash/pipeline.rb', line 363 def shutdown(&before_stop) # shutdown can only start once the pipeline has completed its startup. # avoid potential race conditoon 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 concurently? before_stop.call if block_given? @logger.info "Closing inputs" @inputs.each(&:do_stop) @logger.info "Closed inputs" end |
#shutdown_flusher ⇒ Object
444 445 446 |
# File 'lib/logstash/pipeline.rb', line 444 def shutdown_flusher @flusher_thread.join end |
#shutdown_workers ⇒ Object
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
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 |
# File 'lib/logstash/pipeline.rb', line 381 def shutdown_workers # Each worker thread will receive this exactly once! @worker_threads.each do |t| @logger.debug("Pushing shutdown", :thread => t) @input_queue.push(LogStash::SHUTDOWN) end @worker_threads.each do |t| @logger.debug("Shutdown waiting for worker thread #{t}") t.join end @filters.each(&:do_close) @outputs.each(&:do_close) end |
#stalling_threads_info ⇒ Object
475 476 477 478 479 480 481 |
# File 'lib/logstash/pipeline.rb', line 475 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 |
#start_flusher ⇒ Object
432 433 434 435 436 437 438 439 440 441 442 |
# File 'lib/logstash/pipeline.rb', line 432 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
324 325 326 |
# File 'lib/logstash/pipeline.rb', line 324 def start_input(plugin) @input_threads << Thread.new { inputworker(plugin) } end |
#start_inputs ⇒ Object
307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 |
# File 'lib/logstash/pipeline.rb', line 307 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 @inputs.each do |input| input.register start_input(input) end end |
#start_workers ⇒ Object
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 |
# File 'lib/logstash/pipeline.rb', line 164 def start_workers @inflight_batches = {} @worker_threads.clear # In case we're restarting the pipeline begin start_inputs @outputs.each {|o| o.register } @filters.each {|f| f.register} pipeline_workers = safe_pipeline_worker_count batch_size = @settings[:pipeline_batch_size] batch_delay = @settings[:pipeline_batch_delay] max_inflight = batch_size * pipeline_workers @logger.info("Starting pipeline", :id => self.pipeline_id, :pipeline_workers => pipeline_workers, :batch_size => batch_size, :batch_delay => batch_delay, :max_inflight => max_inflight) 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})" end pipeline_workers.times do |t| @worker_threads << Thread.new do LogStash::Util.set_thread_name("[#{pipeline_id}]>worker#{t}") worker_loop(batch_size, batch_delay) end end ensure # it is important to garantee @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 |
#stopped? ⇒ Boolean
160 161 162 |
# File 'lib/logstash/pipeline.rb', line 160 def stopped? @running.false? end |
#take_batch(batch_size, batch_delay) ⇒ Object
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 |
# File 'lib/logstash/pipeline.rb', line 227 def take_batch(batch_size, batch_delay) batch = [] # Since this is externally synchronized in `worker_look` wec can guarantee that the visibility of an insight batch # guaranteed to be a full batch not a partial batch set_current_thread_inflight_batch(batch) signal = false batch_size.times do |t| event = (t == 0) ? @input_queue.take : @input_queue.poll(batch_delay) if event.nil? next elsif event == LogStash::SHUTDOWN || event == LogStash::FLUSH # We MUST break here. If a batch consumes two SHUTDOWN events # then another worker may have its SHUTDOWN 'stolen', thus blocking # the pipeline. We should stop doing work after flush as well. signal = event break else batch << event end end [batch, signal] end |
#transition_to_running ⇒ Object
def run
148 149 150 |
# File 'lib/logstash/pipeline.rb', line 148 def transition_to_running @running.make_true end |
#transition_to_stopped ⇒ Object
152 153 154 |
# File 'lib/logstash/pipeline.rb', line 152 def transition_to_stopped @running.make_false end |
#wait_inputs ⇒ Object
303 304 305 |
# File 'lib/logstash/pipeline.rb', line 303 def wait_inputs @input_threads.each(&:join) end |
#worker_loop(batch_size, batch_delay) ⇒ Object
Main body of what a worker thread does Repeatedly takes batches off the queu, filters, then outputs them
202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 |
# File 'lib/logstash/pipeline.rb', line 202 def worker_loop(batch_size, batch_delay) running = true while running # To understand the purpose behind this synchronize please read the body of take_batch input_batch, signal = @input_queue_pop_mutex.synchronize { take_batch(batch_size, batch_delay) } running = false if signal == LogStash::SHUTDOWN @events_consumed.increment(input_batch.size) filtered_batch = filter_batch(input_batch) if signal # Flush on SHUTDOWN or FLUSH = (signal == LogStash::SHUTDOWN) ? {:final => true} : {} flush_filters_to_batch(filtered_batch, ) end @events_filtered.increment(filtered_batch.size) output_batch(filtered_batch) inflight_batches_synchronize { set_current_thread_inflight_batch(nil) } end end |