Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 1 | package sarama |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "fmt" |
| 6 | "sync" |
| 7 | "sync/atomic" |
| 8 | "time" |
| 9 | ) |
| 10 | |
| 11 | // ConsumerMessage encapsulates a Kafka message returned by the consumer. |
| 12 | type ConsumerMessage struct { |
| 13 | Key, Value []byte |
| 14 | Topic string |
| 15 | Partition int32 |
| 16 | Offset int64 |
| 17 | Timestamp time.Time // only set if kafka is version 0.10+, inner message timestamp |
| 18 | BlockTimestamp time.Time // only set if kafka is version 0.10+, outer (compressed) block timestamp |
| 19 | Headers []*RecordHeader // only set if kafka is version 0.11+ |
| 20 | } |
| 21 | |
| 22 | // ConsumerError is what is provided to the user when an error occurs. |
| 23 | // It wraps an error and includes the topic and partition. |
| 24 | type ConsumerError struct { |
| 25 | Topic string |
| 26 | Partition int32 |
| 27 | Err error |
| 28 | } |
| 29 | |
| 30 | func (ce ConsumerError) Error() string { |
| 31 | return fmt.Sprintf("kafka: error while consuming %s/%d: %s", ce.Topic, ce.Partition, ce.Err) |
| 32 | } |
| 33 | |
| 34 | // ConsumerErrors is a type that wraps a batch of errors and implements the Error interface. |
| 35 | // It can be returned from the PartitionConsumer's Close methods to avoid the need to manually drain errors |
| 36 | // when stopping. |
| 37 | type ConsumerErrors []*ConsumerError |
| 38 | |
| 39 | func (ce ConsumerErrors) Error() string { |
| 40 | return fmt.Sprintf("kafka: %d errors while consuming", len(ce)) |
| 41 | } |
| 42 | |
| 43 | // Consumer manages PartitionConsumers which process Kafka messages from brokers. You MUST call Close() |
| 44 | // on a consumer to avoid leaks, it will not be garbage-collected automatically when it passes out of |
| 45 | // scope. |
| 46 | // |
| 47 | // Sarama's Consumer type does not currently support automatic consumer-group rebalancing and offset tracking. |
| 48 | // For Zookeeper-based tracking (Kafka 0.8.2 and earlier), the https://github.com/wvanbergen/kafka library |
| 49 | // builds on Sarama to add this support. For Kafka-based tracking (Kafka 0.9 and later), the |
| 50 | // https://github.com/bsm/sarama-cluster library builds on Sarama to add this support. |
| 51 | type Consumer interface { |
| 52 | |
| 53 | // Topics returns the set of available topics as retrieved from the cluster |
| 54 | // metadata. This method is the same as Client.Topics(), and is provided for |
| 55 | // convenience. |
| 56 | Topics() ([]string, error) |
| 57 | |
| 58 | // Partitions returns the sorted list of all partition IDs for the given topic. |
| 59 | // This method is the same as Client.Partitions(), and is provided for convenience. |
| 60 | Partitions(topic string) ([]int32, error) |
| 61 | |
| 62 | // ConsumePartition creates a PartitionConsumer on the given topic/partition with |
| 63 | // the given offset. It will return an error if this Consumer is already consuming |
| 64 | // on the given topic/partition. Offset can be a literal offset, or OffsetNewest |
| 65 | // or OffsetOldest |
| 66 | ConsumePartition(topic string, partition int32, offset int64) (PartitionConsumer, error) |
| 67 | |
| 68 | // HighWaterMarks returns the current high water marks for each topic and partition. |
| 69 | // Consistency between partitions is not guaranteed since high water marks are updated separately. |
| 70 | HighWaterMarks() map[string]map[int32]int64 |
| 71 | |
| 72 | // Close shuts down the consumer. It must be called after all child |
| 73 | // PartitionConsumers have already been closed. |
| 74 | Close() error |
| 75 | } |
| 76 | |
| 77 | type consumer struct { |
| 78 | client Client |
| 79 | conf *Config |
| 80 | ownClient bool |
| 81 | |
| 82 | lock sync.Mutex |
| 83 | children map[string]map[int32]*partitionConsumer |
| 84 | brokerConsumers map[*Broker]*brokerConsumer |
| 85 | } |
| 86 | |
| 87 | // NewConsumer creates a new consumer using the given broker addresses and configuration. |
| 88 | func NewConsumer(addrs []string, config *Config) (Consumer, error) { |
| 89 | client, err := NewClient(addrs, config) |
| 90 | if err != nil { |
| 91 | return nil, err |
| 92 | } |
| 93 | |
| 94 | c, err := NewConsumerFromClient(client) |
| 95 | if err != nil { |
| 96 | return nil, err |
| 97 | } |
| 98 | c.(*consumer).ownClient = true |
| 99 | return c, nil |
| 100 | } |
| 101 | |
| 102 | // NewConsumerFromClient creates a new consumer using the given client. It is still |
| 103 | // necessary to call Close() on the underlying client when shutting down this consumer. |
| 104 | func NewConsumerFromClient(client Client) (Consumer, error) { |
| 105 | // Check that we are not dealing with a closed Client before processing any other arguments |
| 106 | if client.Closed() { |
| 107 | return nil, ErrClosedClient |
| 108 | } |
| 109 | |
| 110 | c := &consumer{ |
| 111 | client: client, |
| 112 | conf: client.Config(), |
| 113 | children: make(map[string]map[int32]*partitionConsumer), |
| 114 | brokerConsumers: make(map[*Broker]*brokerConsumer), |
| 115 | } |
| 116 | |
| 117 | return c, nil |
| 118 | } |
| 119 | |
| 120 | func (c *consumer) Close() error { |
| 121 | if c.ownClient { |
| 122 | return c.client.Close() |
| 123 | } |
| 124 | return nil |
| 125 | } |
| 126 | |
| 127 | func (c *consumer) Topics() ([]string, error) { |
| 128 | return c.client.Topics() |
| 129 | } |
| 130 | |
| 131 | func (c *consumer) Partitions(topic string) ([]int32, error) { |
| 132 | return c.client.Partitions(topic) |
| 133 | } |
| 134 | |
| 135 | func (c *consumer) ConsumePartition(topic string, partition int32, offset int64) (PartitionConsumer, error) { |
| 136 | child := &partitionConsumer{ |
| 137 | consumer: c, |
| 138 | conf: c.conf, |
| 139 | topic: topic, |
| 140 | partition: partition, |
| 141 | messages: make(chan *ConsumerMessage, c.conf.ChannelBufferSize), |
| 142 | errors: make(chan *ConsumerError, c.conf.ChannelBufferSize), |
| 143 | feeder: make(chan *FetchResponse, 1), |
| 144 | trigger: make(chan none, 1), |
| 145 | dying: make(chan none), |
| 146 | fetchSize: c.conf.Consumer.Fetch.Default, |
| 147 | } |
| 148 | |
| 149 | if err := child.chooseStartingOffset(offset); err != nil { |
| 150 | return nil, err |
| 151 | } |
| 152 | |
| 153 | var leader *Broker |
| 154 | var err error |
| 155 | if leader, err = c.client.Leader(child.topic, child.partition); err != nil { |
| 156 | return nil, err |
| 157 | } |
| 158 | |
| 159 | if err := c.addChild(child); err != nil { |
| 160 | return nil, err |
| 161 | } |
| 162 | |
| 163 | go withRecover(child.dispatcher) |
| 164 | go withRecover(child.responseFeeder) |
| 165 | |
| 166 | child.broker = c.refBrokerConsumer(leader) |
| 167 | child.broker.input <- child |
| 168 | |
| 169 | return child, nil |
| 170 | } |
| 171 | |
| 172 | func (c *consumer) HighWaterMarks() map[string]map[int32]int64 { |
| 173 | c.lock.Lock() |
| 174 | defer c.lock.Unlock() |
| 175 | |
| 176 | hwms := make(map[string]map[int32]int64) |
| 177 | for topic, p := range c.children { |
| 178 | hwm := make(map[int32]int64, len(p)) |
| 179 | for partition, pc := range p { |
| 180 | hwm[partition] = pc.HighWaterMarkOffset() |
| 181 | } |
| 182 | hwms[topic] = hwm |
| 183 | } |
| 184 | |
| 185 | return hwms |
| 186 | } |
| 187 | |
| 188 | func (c *consumer) addChild(child *partitionConsumer) error { |
| 189 | c.lock.Lock() |
| 190 | defer c.lock.Unlock() |
| 191 | |
| 192 | topicChildren := c.children[child.topic] |
| 193 | if topicChildren == nil { |
| 194 | topicChildren = make(map[int32]*partitionConsumer) |
| 195 | c.children[child.topic] = topicChildren |
| 196 | } |
| 197 | |
| 198 | if topicChildren[child.partition] != nil { |
| 199 | return ConfigurationError("That topic/partition is already being consumed") |
| 200 | } |
| 201 | |
| 202 | topicChildren[child.partition] = child |
| 203 | return nil |
| 204 | } |
| 205 | |
| 206 | func (c *consumer) removeChild(child *partitionConsumer) { |
| 207 | c.lock.Lock() |
| 208 | defer c.lock.Unlock() |
| 209 | |
| 210 | delete(c.children[child.topic], child.partition) |
| 211 | } |
| 212 | |
| 213 | func (c *consumer) refBrokerConsumer(broker *Broker) *brokerConsumer { |
| 214 | c.lock.Lock() |
| 215 | defer c.lock.Unlock() |
| 216 | |
| 217 | bc := c.brokerConsumers[broker] |
| 218 | if bc == nil { |
| 219 | bc = c.newBrokerConsumer(broker) |
| 220 | c.brokerConsumers[broker] = bc |
| 221 | } |
| 222 | |
| 223 | bc.refs++ |
| 224 | |
| 225 | return bc |
| 226 | } |
| 227 | |
| 228 | func (c *consumer) unrefBrokerConsumer(brokerWorker *brokerConsumer) { |
| 229 | c.lock.Lock() |
| 230 | defer c.lock.Unlock() |
| 231 | |
| 232 | brokerWorker.refs-- |
| 233 | |
| 234 | if brokerWorker.refs == 0 { |
| 235 | close(brokerWorker.input) |
| 236 | if c.brokerConsumers[brokerWorker.broker] == brokerWorker { |
| 237 | delete(c.brokerConsumers, brokerWorker.broker) |
| 238 | } |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | func (c *consumer) abandonBrokerConsumer(brokerWorker *brokerConsumer) { |
| 243 | c.lock.Lock() |
| 244 | defer c.lock.Unlock() |
| 245 | |
| 246 | delete(c.brokerConsumers, brokerWorker.broker) |
| 247 | } |
| 248 | |
| 249 | // PartitionConsumer |
| 250 | |
| 251 | // PartitionConsumer processes Kafka messages from a given topic and partition. You MUST call one of Close() or |
| 252 | // AsyncClose() on a PartitionConsumer to avoid leaks; it will not be garbage-collected automatically when it passes out |
| 253 | // of scope. |
| 254 | // |
| 255 | // The simplest way of using a PartitionConsumer is to loop over its Messages channel using a for/range |
| 256 | // loop. The PartitionConsumer will only stop itself in one case: when the offset being consumed is reported |
| 257 | // as out of range by the brokers. In this case you should decide what you want to do (try a different offset, |
| 258 | // notify a human, etc) and handle it appropriately. For all other error cases, it will just keep retrying. |
| 259 | // By default, it logs these errors to sarama.Logger; if you want to be notified directly of all errors, set |
| 260 | // your config's Consumer.Return.Errors to true and read from the Errors channel, using a select statement |
| 261 | // or a separate goroutine. Check out the Consumer examples to see implementations of these different approaches. |
| 262 | // |
| 263 | // To terminate such a for/range loop while the loop is executing, call AsyncClose. This will kick off the process of |
| 264 | // consumer tear-down & return imediately. Continue to loop, servicing the Messages channel until the teardown process |
| 265 | // AsyncClose initiated closes it (thus terminating the for/range loop). If you've already ceased reading Messages, call |
| 266 | // Close; this will signal the PartitionConsumer's goroutines to begin shutting down (just like AsyncClose), but will |
| 267 | // also drain the Messages channel, harvest all errors & return them once cleanup has completed. |
| 268 | type PartitionConsumer interface { |
| 269 | |
| 270 | // AsyncClose initiates a shutdown of the PartitionConsumer. This method will return immediately, after which you |
| 271 | // should continue to service the 'Messages' and 'Errors' channels until they are empty. It is required to call this |
| 272 | // function, or Close before a consumer object passes out of scope, as it will otherwise leak memory. You must call |
| 273 | // this before calling Close on the underlying client. |
| 274 | AsyncClose() |
| 275 | |
| 276 | // Close stops the PartitionConsumer from fetching messages. It will initiate a shutdown just like AsyncClose, drain |
| 277 | // the Messages channel, harvest any errors & return them to the caller. Note that if you are continuing to service |
| 278 | // the Messages channel when this function is called, you will be competing with Close for messages; consider |
| 279 | // calling AsyncClose, instead. It is required to call this function (or AsyncClose) before a consumer object passes |
| 280 | // out of scope, as it will otherwise leak memory. You must call this before calling Close on the underlying client. |
| 281 | Close() error |
| 282 | |
| 283 | // Messages returns the read channel for the messages that are returned by |
| 284 | // the broker. |
| 285 | Messages() <-chan *ConsumerMessage |
| 286 | |
| 287 | // Errors returns a read channel of errors that occurred during consuming, if |
| 288 | // enabled. By default, errors are logged and not returned over this channel. |
| 289 | // If you want to implement any custom error handling, set your config's |
| 290 | // Consumer.Return.Errors setting to true, and read from this channel. |
| 291 | Errors() <-chan *ConsumerError |
| 292 | |
| 293 | // HighWaterMarkOffset returns the high water mark offset of the partition, |
| 294 | // i.e. the offset that will be used for the next message that will be produced. |
| 295 | // You can use this to determine how far behind the processing is. |
| 296 | HighWaterMarkOffset() int64 |
| 297 | } |
| 298 | |
| 299 | type partitionConsumer struct { |
| 300 | highWaterMarkOffset int64 // must be at the top of the struct because https://golang.org/pkg/sync/atomic/#pkg-note-BUG |
| 301 | consumer *consumer |
| 302 | conf *Config |
| 303 | topic string |
| 304 | partition int32 |
| 305 | |
| 306 | broker *brokerConsumer |
| 307 | messages chan *ConsumerMessage |
| 308 | errors chan *ConsumerError |
| 309 | feeder chan *FetchResponse |
| 310 | |
| 311 | trigger, dying chan none |
| 312 | responseResult error |
| 313 | closeOnce sync.Once |
| 314 | |
| 315 | fetchSize int32 |
| 316 | offset int64 |
| 317 | |
| 318 | retries int32 |
| 319 | } |
| 320 | |
| 321 | var errTimedOut = errors.New("timed out feeding messages to the user") // not user-facing |
| 322 | |
| 323 | func (child *partitionConsumer) sendError(err error) { |
| 324 | cErr := &ConsumerError{ |
| 325 | Topic: child.topic, |
| 326 | Partition: child.partition, |
| 327 | Err: err, |
| 328 | } |
| 329 | |
| 330 | if child.conf.Consumer.Return.Errors { |
| 331 | child.errors <- cErr |
| 332 | } else { |
| 333 | Logger.Println(cErr) |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | func (child *partitionConsumer) computeBackoff() time.Duration { |
| 338 | if child.conf.Consumer.Retry.BackoffFunc != nil { |
| 339 | retries := atomic.AddInt32(&child.retries, 1) |
| 340 | return child.conf.Consumer.Retry.BackoffFunc(int(retries)) |
| 341 | } else { |
| 342 | return child.conf.Consumer.Retry.Backoff |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | func (child *partitionConsumer) dispatcher() { |
| 347 | for range child.trigger { |
| 348 | select { |
| 349 | case <-child.dying: |
| 350 | close(child.trigger) |
| 351 | case <-time.After(child.computeBackoff()): |
| 352 | if child.broker != nil { |
| 353 | child.consumer.unrefBrokerConsumer(child.broker) |
| 354 | child.broker = nil |
| 355 | } |
| 356 | |
| 357 | Logger.Printf("consumer/%s/%d finding new broker\n", child.topic, child.partition) |
| 358 | if err := child.dispatch(); err != nil { |
| 359 | child.sendError(err) |
| 360 | child.trigger <- none{} |
| 361 | } |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | if child.broker != nil { |
| 366 | child.consumer.unrefBrokerConsumer(child.broker) |
| 367 | } |
| 368 | child.consumer.removeChild(child) |
| 369 | close(child.feeder) |
| 370 | } |
| 371 | |
| 372 | func (child *partitionConsumer) dispatch() error { |
| 373 | if err := child.consumer.client.RefreshMetadata(child.topic); err != nil { |
| 374 | return err |
| 375 | } |
| 376 | |
| 377 | var leader *Broker |
| 378 | var err error |
| 379 | if leader, err = child.consumer.client.Leader(child.topic, child.partition); err != nil { |
| 380 | return err |
| 381 | } |
| 382 | |
| 383 | child.broker = child.consumer.refBrokerConsumer(leader) |
| 384 | |
| 385 | child.broker.input <- child |
| 386 | |
| 387 | return nil |
| 388 | } |
| 389 | |
| 390 | func (child *partitionConsumer) chooseStartingOffset(offset int64) error { |
| 391 | newestOffset, err := child.consumer.client.GetOffset(child.topic, child.partition, OffsetNewest) |
| 392 | if err != nil { |
| 393 | return err |
| 394 | } |
| 395 | oldestOffset, err := child.consumer.client.GetOffset(child.topic, child.partition, OffsetOldest) |
| 396 | if err != nil { |
| 397 | return err |
| 398 | } |
| 399 | |
| 400 | switch { |
| 401 | case offset == OffsetNewest: |
| 402 | child.offset = newestOffset |
| 403 | case offset == OffsetOldest: |
| 404 | child.offset = oldestOffset |
| 405 | case offset >= oldestOffset && offset <= newestOffset: |
| 406 | child.offset = offset |
| 407 | default: |
| 408 | return ErrOffsetOutOfRange |
| 409 | } |
| 410 | |
| 411 | return nil |
| 412 | } |
| 413 | |
| 414 | func (child *partitionConsumer) Messages() <-chan *ConsumerMessage { |
| 415 | return child.messages |
| 416 | } |
| 417 | |
| 418 | func (child *partitionConsumer) Errors() <-chan *ConsumerError { |
| 419 | return child.errors |
| 420 | } |
| 421 | |
| 422 | func (child *partitionConsumer) AsyncClose() { |
| 423 | // this triggers whatever broker owns this child to abandon it and close its trigger channel, which causes |
| 424 | // the dispatcher to exit its loop, which removes it from the consumer then closes its 'messages' and |
| 425 | // 'errors' channel (alternatively, if the child is already at the dispatcher for some reason, that will |
| 426 | // also just close itself) |
| 427 | child.closeOnce.Do(func() { |
| 428 | close(child.dying) |
| 429 | }) |
| 430 | } |
| 431 | |
| 432 | func (child *partitionConsumer) Close() error { |
| 433 | child.AsyncClose() |
| 434 | |
| 435 | go withRecover(func() { |
| 436 | for range child.messages { |
| 437 | // drain |
| 438 | } |
| 439 | }) |
| 440 | |
| 441 | var errors ConsumerErrors |
| 442 | for err := range child.errors { |
| 443 | errors = append(errors, err) |
| 444 | } |
| 445 | |
| 446 | if len(errors) > 0 { |
| 447 | return errors |
| 448 | } |
| 449 | return nil |
| 450 | } |
| 451 | |
| 452 | func (child *partitionConsumer) HighWaterMarkOffset() int64 { |
| 453 | return atomic.LoadInt64(&child.highWaterMarkOffset) |
| 454 | } |
| 455 | |
| 456 | func (child *partitionConsumer) responseFeeder() { |
| 457 | var msgs []*ConsumerMessage |
| 458 | expiryTicker := time.NewTicker(child.conf.Consumer.MaxProcessingTime) |
| 459 | firstAttempt := true |
| 460 | |
| 461 | feederLoop: |
| 462 | for response := range child.feeder { |
| 463 | msgs, child.responseResult = child.parseResponse(response) |
| 464 | |
| 465 | if child.responseResult == nil { |
| 466 | atomic.StoreInt32(&child.retries, 0) |
| 467 | } |
| 468 | |
| 469 | for i, msg := range msgs { |
| 470 | messageSelect: |
| 471 | select { |
| 472 | case child.messages <- msg: |
| 473 | firstAttempt = true |
| 474 | case <-expiryTicker.C: |
| 475 | if !firstAttempt { |
| 476 | child.responseResult = errTimedOut |
| 477 | child.broker.acks.Done() |
| 478 | for _, msg = range msgs[i:] { |
| 479 | child.messages <- msg |
| 480 | } |
| 481 | child.broker.input <- child |
| 482 | continue feederLoop |
| 483 | } else { |
| 484 | // current message has not been sent, return to select |
| 485 | // statement |
| 486 | firstAttempt = false |
| 487 | goto messageSelect |
| 488 | } |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | child.broker.acks.Done() |
| 493 | } |
| 494 | |
| 495 | expiryTicker.Stop() |
| 496 | close(child.messages) |
| 497 | close(child.errors) |
| 498 | } |
| 499 | |
| 500 | func (child *partitionConsumer) parseMessages(msgSet *MessageSet) ([]*ConsumerMessage, error) { |
| 501 | var messages []*ConsumerMessage |
| 502 | for _, msgBlock := range msgSet.Messages { |
| 503 | for _, msg := range msgBlock.Messages() { |
| 504 | offset := msg.Offset |
| 505 | timestamp := msg.Msg.Timestamp |
| 506 | if msg.Msg.Version >= 1 { |
| 507 | baseOffset := msgBlock.Offset - msgBlock.Messages()[len(msgBlock.Messages())-1].Offset |
| 508 | offset += baseOffset |
| 509 | if msg.Msg.LogAppendTime { |
| 510 | timestamp = msgBlock.Msg.Timestamp |
| 511 | } |
| 512 | } |
| 513 | if offset < child.offset { |
| 514 | continue |
| 515 | } |
| 516 | messages = append(messages, &ConsumerMessage{ |
| 517 | Topic: child.topic, |
| 518 | Partition: child.partition, |
| 519 | Key: msg.Msg.Key, |
| 520 | Value: msg.Msg.Value, |
| 521 | Offset: offset, |
| 522 | Timestamp: timestamp, |
| 523 | BlockTimestamp: msgBlock.Msg.Timestamp, |
| 524 | }) |
| 525 | child.offset = offset + 1 |
| 526 | } |
| 527 | } |
| 528 | if len(messages) == 0 { |
| 529 | child.offset++ |
| 530 | } |
| 531 | return messages, nil |
| 532 | } |
| 533 | |
| 534 | func (child *partitionConsumer) parseRecords(batch *RecordBatch) ([]*ConsumerMessage, error) { |
| 535 | var messages []*ConsumerMessage |
| 536 | for _, rec := range batch.Records { |
| 537 | offset := batch.FirstOffset + rec.OffsetDelta |
| 538 | if offset < child.offset { |
| 539 | continue |
| 540 | } |
| 541 | timestamp := batch.FirstTimestamp.Add(rec.TimestampDelta) |
| 542 | if batch.LogAppendTime { |
| 543 | timestamp = batch.MaxTimestamp |
| 544 | } |
| 545 | messages = append(messages, &ConsumerMessage{ |
| 546 | Topic: child.topic, |
| 547 | Partition: child.partition, |
| 548 | Key: rec.Key, |
| 549 | Value: rec.Value, |
| 550 | Offset: offset, |
| 551 | Timestamp: timestamp, |
| 552 | Headers: rec.Headers, |
| 553 | }) |
| 554 | child.offset = offset + 1 |
| 555 | } |
| 556 | if len(messages) == 0 { |
| 557 | child.offset++ |
| 558 | } |
| 559 | return messages, nil |
| 560 | } |
| 561 | |
| 562 | func (child *partitionConsumer) parseResponse(response *FetchResponse) ([]*ConsumerMessage, error) { |
| 563 | block := response.GetBlock(child.topic, child.partition) |
| 564 | if block == nil { |
| 565 | return nil, ErrIncompleteResponse |
| 566 | } |
| 567 | |
| 568 | if block.Err != ErrNoError { |
| 569 | return nil, block.Err |
| 570 | } |
| 571 | |
| 572 | nRecs, err := block.numRecords() |
| 573 | if err != nil { |
| 574 | return nil, err |
| 575 | } |
| 576 | if nRecs == 0 { |
| 577 | partialTrailingMessage, err := block.isPartial() |
| 578 | if err != nil { |
| 579 | return nil, err |
| 580 | } |
| 581 | // We got no messages. If we got a trailing one then we need to ask for more data. |
| 582 | // Otherwise we just poll again and wait for one to be produced... |
| 583 | if partialTrailingMessage { |
| 584 | if child.conf.Consumer.Fetch.Max > 0 && child.fetchSize == child.conf.Consumer.Fetch.Max { |
| 585 | // we can't ask for more data, we've hit the configured limit |
| 586 | child.sendError(ErrMessageTooLarge) |
| 587 | child.offset++ // skip this one so we can keep processing future messages |
| 588 | } else { |
| 589 | child.fetchSize *= 2 |
| 590 | if child.conf.Consumer.Fetch.Max > 0 && child.fetchSize > child.conf.Consumer.Fetch.Max { |
| 591 | child.fetchSize = child.conf.Consumer.Fetch.Max |
| 592 | } |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | return nil, nil |
| 597 | } |
| 598 | |
| 599 | // we got messages, reset our fetch size in case it was increased for a previous request |
| 600 | child.fetchSize = child.conf.Consumer.Fetch.Default |
| 601 | atomic.StoreInt64(&child.highWaterMarkOffset, block.HighWaterMarkOffset) |
| 602 | |
| 603 | messages := []*ConsumerMessage{} |
| 604 | for _, records := range block.RecordsSet { |
| 605 | switch records.recordsType { |
| 606 | case legacyRecords: |
| 607 | messageSetMessages, err := child.parseMessages(records.MsgSet) |
| 608 | if err != nil { |
| 609 | return nil, err |
| 610 | } |
| 611 | |
| 612 | messages = append(messages, messageSetMessages...) |
| 613 | case defaultRecords: |
| 614 | recordBatchMessages, err := child.parseRecords(records.RecordBatch) |
| 615 | if err != nil { |
| 616 | return nil, err |
| 617 | } |
| 618 | if control, err := records.isControl(); err != nil || control { |
| 619 | continue |
| 620 | } |
| 621 | |
| 622 | messages = append(messages, recordBatchMessages...) |
| 623 | default: |
| 624 | return nil, fmt.Errorf("unknown records type: %v", records.recordsType) |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | return messages, nil |
| 629 | } |
| 630 | |
| 631 | // brokerConsumer |
| 632 | |
| 633 | type brokerConsumer struct { |
| 634 | consumer *consumer |
| 635 | broker *Broker |
| 636 | input chan *partitionConsumer |
| 637 | newSubscriptions chan []*partitionConsumer |
| 638 | wait chan none |
| 639 | subscriptions map[*partitionConsumer]none |
| 640 | acks sync.WaitGroup |
| 641 | refs int |
| 642 | } |
| 643 | |
| 644 | func (c *consumer) newBrokerConsumer(broker *Broker) *brokerConsumer { |
| 645 | bc := &brokerConsumer{ |
| 646 | consumer: c, |
| 647 | broker: broker, |
| 648 | input: make(chan *partitionConsumer), |
| 649 | newSubscriptions: make(chan []*partitionConsumer), |
| 650 | wait: make(chan none), |
| 651 | subscriptions: make(map[*partitionConsumer]none), |
| 652 | refs: 0, |
| 653 | } |
| 654 | |
| 655 | go withRecover(bc.subscriptionManager) |
| 656 | go withRecover(bc.subscriptionConsumer) |
| 657 | |
| 658 | return bc |
| 659 | } |
| 660 | |
| 661 | func (bc *brokerConsumer) subscriptionManager() { |
| 662 | var buffer []*partitionConsumer |
| 663 | |
| 664 | // The subscriptionManager constantly accepts new subscriptions on `input` (even when the main subscriptionConsumer |
| 665 | // goroutine is in the middle of a network request) and batches it up. The main worker goroutine picks |
| 666 | // up a batch of new subscriptions between every network request by reading from `newSubscriptions`, so we give |
| 667 | // it nil if no new subscriptions are available. We also write to `wait` only when new subscriptions is available, |
| 668 | // so the main goroutine can block waiting for work if it has none. |
| 669 | for { |
| 670 | if len(buffer) > 0 { |
| 671 | select { |
| 672 | case event, ok := <-bc.input: |
| 673 | if !ok { |
| 674 | goto done |
| 675 | } |
| 676 | buffer = append(buffer, event) |
| 677 | case bc.newSubscriptions <- buffer: |
| 678 | buffer = nil |
| 679 | case bc.wait <- none{}: |
| 680 | } |
| 681 | } else { |
| 682 | select { |
| 683 | case event, ok := <-bc.input: |
| 684 | if !ok { |
| 685 | goto done |
| 686 | } |
| 687 | buffer = append(buffer, event) |
| 688 | case bc.newSubscriptions <- nil: |
| 689 | } |
| 690 | } |
| 691 | } |
| 692 | |
| 693 | done: |
| 694 | close(bc.wait) |
| 695 | if len(buffer) > 0 { |
| 696 | bc.newSubscriptions <- buffer |
| 697 | } |
| 698 | close(bc.newSubscriptions) |
| 699 | } |
| 700 | |
| 701 | func (bc *brokerConsumer) subscriptionConsumer() { |
| 702 | <-bc.wait // wait for our first piece of work |
| 703 | |
| 704 | // the subscriptionConsumer ensures we will get nil right away if no new subscriptions is available |
| 705 | for newSubscriptions := range bc.newSubscriptions { |
| 706 | bc.updateSubscriptions(newSubscriptions) |
| 707 | |
| 708 | if len(bc.subscriptions) == 0 { |
| 709 | // We're about to be shut down or we're about to receive more subscriptions. |
| 710 | // Either way, the signal just hasn't propagated to our goroutine yet. |
| 711 | <-bc.wait |
| 712 | continue |
| 713 | } |
| 714 | |
| 715 | response, err := bc.fetchNewMessages() |
| 716 | |
| 717 | if err != nil { |
| 718 | Logger.Printf("consumer/broker/%d disconnecting due to error processing FetchRequest: %s\n", bc.broker.ID(), err) |
| 719 | bc.abort(err) |
| 720 | return |
| 721 | } |
| 722 | |
| 723 | bc.acks.Add(len(bc.subscriptions)) |
| 724 | for child := range bc.subscriptions { |
| 725 | child.feeder <- response |
| 726 | } |
| 727 | bc.acks.Wait() |
| 728 | bc.handleResponses() |
| 729 | } |
| 730 | } |
| 731 | |
| 732 | func (bc *brokerConsumer) updateSubscriptions(newSubscriptions []*partitionConsumer) { |
| 733 | for _, child := range newSubscriptions { |
| 734 | bc.subscriptions[child] = none{} |
| 735 | Logger.Printf("consumer/broker/%d added subscription to %s/%d\n", bc.broker.ID(), child.topic, child.partition) |
| 736 | } |
| 737 | |
| 738 | for child := range bc.subscriptions { |
| 739 | select { |
| 740 | case <-child.dying: |
| 741 | Logger.Printf("consumer/broker/%d closed dead subscription to %s/%d\n", bc.broker.ID(), child.topic, child.partition) |
| 742 | close(child.trigger) |
| 743 | delete(bc.subscriptions, child) |
| 744 | default: |
| 745 | break |
| 746 | } |
| 747 | } |
| 748 | } |
| 749 | |
| 750 | func (bc *brokerConsumer) handleResponses() { |
| 751 | // handles the response codes left for us by our subscriptions, and abandons ones that have been closed |
| 752 | for child := range bc.subscriptions { |
| 753 | result := child.responseResult |
| 754 | child.responseResult = nil |
| 755 | |
| 756 | switch result { |
| 757 | case nil: |
| 758 | break |
| 759 | case errTimedOut: |
| 760 | Logger.Printf("consumer/broker/%d abandoned subscription to %s/%d because consuming was taking too long\n", |
| 761 | bc.broker.ID(), child.topic, child.partition) |
| 762 | delete(bc.subscriptions, child) |
| 763 | case ErrOffsetOutOfRange: |
| 764 | // there's no point in retrying this it will just fail the same way again |
| 765 | // shut it down and force the user to choose what to do |
| 766 | child.sendError(result) |
| 767 | Logger.Printf("consumer/%s/%d shutting down because %s\n", child.topic, child.partition, result) |
| 768 | close(child.trigger) |
| 769 | delete(bc.subscriptions, child) |
| 770 | case ErrUnknownTopicOrPartition, ErrNotLeaderForPartition, ErrLeaderNotAvailable, ErrReplicaNotAvailable: |
| 771 | // not an error, but does need redispatching |
| 772 | Logger.Printf("consumer/broker/%d abandoned subscription to %s/%d because %s\n", |
| 773 | bc.broker.ID(), child.topic, child.partition, result) |
| 774 | child.trigger <- none{} |
| 775 | delete(bc.subscriptions, child) |
| 776 | default: |
| 777 | // dunno, tell the user and try redispatching |
| 778 | child.sendError(result) |
| 779 | Logger.Printf("consumer/broker/%d abandoned subscription to %s/%d because %s\n", |
| 780 | bc.broker.ID(), child.topic, child.partition, result) |
| 781 | child.trigger <- none{} |
| 782 | delete(bc.subscriptions, child) |
| 783 | } |
| 784 | } |
| 785 | } |
| 786 | |
| 787 | func (bc *brokerConsumer) abort(err error) { |
| 788 | bc.consumer.abandonBrokerConsumer(bc) |
| 789 | _ = bc.broker.Close() // we don't care about the error this might return, we already have one |
| 790 | |
| 791 | for child := range bc.subscriptions { |
| 792 | child.sendError(err) |
| 793 | child.trigger <- none{} |
| 794 | } |
| 795 | |
| 796 | for newSubscriptions := range bc.newSubscriptions { |
| 797 | if len(newSubscriptions) == 0 { |
| 798 | <-bc.wait |
| 799 | continue |
| 800 | } |
| 801 | for _, child := range newSubscriptions { |
| 802 | child.sendError(err) |
| 803 | child.trigger <- none{} |
| 804 | } |
| 805 | } |
| 806 | } |
| 807 | |
| 808 | func (bc *brokerConsumer) fetchNewMessages() (*FetchResponse, error) { |
| 809 | request := &FetchRequest{ |
| 810 | MinBytes: bc.consumer.conf.Consumer.Fetch.Min, |
| 811 | MaxWaitTime: int32(bc.consumer.conf.Consumer.MaxWaitTime / time.Millisecond), |
| 812 | } |
| 813 | if bc.consumer.conf.Version.IsAtLeast(V0_9_0_0) { |
| 814 | request.Version = 1 |
| 815 | } |
| 816 | if bc.consumer.conf.Version.IsAtLeast(V0_10_0_0) { |
| 817 | request.Version = 2 |
| 818 | } |
| 819 | if bc.consumer.conf.Version.IsAtLeast(V0_10_1_0) { |
| 820 | request.Version = 3 |
| 821 | request.MaxBytes = MaxResponseSize |
| 822 | } |
| 823 | if bc.consumer.conf.Version.IsAtLeast(V0_11_0_0) { |
| 824 | request.Version = 4 |
| 825 | request.Isolation = ReadUncommitted // We don't support yet transactions. |
| 826 | } |
| 827 | |
| 828 | for child := range bc.subscriptions { |
| 829 | request.AddBlock(child.topic, child.partition, child.offset, child.fetchSize) |
| 830 | } |
| 831 | |
| 832 | return bc.broker.Fetch(request) |
| 833 | } |