Takahiro Suzuki | 241c10e | 2020-12-17 20:17:57 +0900 | [diff] [blame] | 1 | package cluster |
| 2 | |
| 3 | import ( |
| 4 | "sort" |
| 5 | "sync" |
| 6 | "sync/atomic" |
| 7 | "time" |
| 8 | |
| 9 | "github.com/Shopify/sarama" |
| 10 | ) |
| 11 | |
| 12 | // Consumer is a cluster group consumer |
| 13 | type Consumer struct { |
| 14 | client *Client |
| 15 | ownClient bool |
| 16 | |
| 17 | consumer sarama.Consumer |
| 18 | subs *partitionMap |
| 19 | |
| 20 | consumerID string |
| 21 | groupID string |
| 22 | |
| 23 | memberID string |
| 24 | generationID int32 |
| 25 | membershipMu sync.RWMutex |
| 26 | |
| 27 | coreTopics []string |
| 28 | extraTopics []string |
| 29 | |
| 30 | dying, dead chan none |
| 31 | closeOnce sync.Once |
| 32 | |
| 33 | consuming int32 |
| 34 | messages chan *sarama.ConsumerMessage |
| 35 | errors chan error |
| 36 | partitions chan PartitionConsumer |
| 37 | notifications chan *Notification |
| 38 | |
| 39 | commitMu sync.Mutex |
| 40 | } |
| 41 | |
| 42 | // NewConsumer initializes a new consumer |
| 43 | func NewConsumer(addrs []string, groupID string, topics []string, config *Config) (*Consumer, error) { |
| 44 | client, err := NewClient(addrs, config) |
| 45 | if err != nil { |
| 46 | return nil, err |
| 47 | } |
| 48 | |
| 49 | consumer, err := NewConsumerFromClient(client, groupID, topics) |
| 50 | if err != nil { |
| 51 | return nil, err |
| 52 | } |
| 53 | consumer.ownClient = true |
| 54 | return consumer, nil |
| 55 | } |
| 56 | |
| 57 | // NewConsumerFromClient initializes a new consumer from an existing client. |
| 58 | // |
| 59 | // Please note that clients cannot be shared between consumers (due to Kafka internals), |
| 60 | // they can only be re-used which requires the user to call Close() on the first consumer |
| 61 | // before using this method again to initialize another one. Attempts to use a client with |
| 62 | // more than one consumer at a time will return errors. |
| 63 | func NewConsumerFromClient(client *Client, groupID string, topics []string) (*Consumer, error) { |
| 64 | if !client.claim() { |
| 65 | return nil, errClientInUse |
| 66 | } |
| 67 | |
| 68 | consumer, err := sarama.NewConsumerFromClient(client.Client) |
| 69 | if err != nil { |
| 70 | client.release() |
| 71 | return nil, err |
| 72 | } |
| 73 | |
| 74 | sort.Strings(topics) |
| 75 | c := &Consumer{ |
| 76 | client: client, |
| 77 | consumer: consumer, |
| 78 | subs: newPartitionMap(), |
| 79 | groupID: groupID, |
| 80 | |
| 81 | coreTopics: topics, |
| 82 | |
| 83 | dying: make(chan none), |
| 84 | dead: make(chan none), |
| 85 | |
| 86 | messages: make(chan *sarama.ConsumerMessage), |
| 87 | errors: make(chan error, client.config.ChannelBufferSize), |
| 88 | partitions: make(chan PartitionConsumer, 1), |
| 89 | notifications: make(chan *Notification), |
| 90 | } |
| 91 | if err := c.client.RefreshCoordinator(groupID); err != nil { |
| 92 | client.release() |
| 93 | return nil, err |
| 94 | } |
| 95 | |
| 96 | go c.mainLoop() |
| 97 | return c, nil |
| 98 | } |
| 99 | |
| 100 | // Messages returns the read channel for the messages that are returned by |
| 101 | // the broker. |
| 102 | // |
| 103 | // This channel will only return if Config.Group.Mode option is set to |
| 104 | // ConsumerModeMultiplex (default). |
| 105 | func (c *Consumer) Messages() <-chan *sarama.ConsumerMessage { return c.messages } |
| 106 | |
| 107 | // Partitions returns the read channels for individual partitions of this broker. |
| 108 | // |
| 109 | // This will channel will only return if Config.Group.Mode option is set to |
| 110 | // ConsumerModePartitions. |
| 111 | // |
| 112 | // The Partitions() channel must be listened to for the life of this consumer; |
| 113 | // when a rebalance happens old partitions will be closed (naturally come to |
| 114 | // completion) and new ones will be emitted. The returned channel will only close |
| 115 | // when the consumer is completely shut down. |
| 116 | func (c *Consumer) Partitions() <-chan PartitionConsumer { return c.partitions } |
| 117 | |
| 118 | // Errors returns a read channel of errors that occur during offset management, if |
| 119 | // enabled. By default, errors are logged and not returned over this channel. If |
| 120 | // you want to implement any custom error handling, set your config's |
| 121 | // Consumer.Return.Errors setting to true, and read from this channel. |
| 122 | func (c *Consumer) Errors() <-chan error { return c.errors } |
| 123 | |
| 124 | // Notifications returns a channel of Notifications that occur during consumer |
| 125 | // rebalancing. Notifications will only be emitted over this channel, if your config's |
| 126 | // Group.Return.Notifications setting to true. |
| 127 | func (c *Consumer) Notifications() <-chan *Notification { return c.notifications } |
| 128 | |
| 129 | // HighWaterMarks returns the current high water marks for each topic and partition |
| 130 | // Consistency between partitions is not guaranteed since high water marks are updated separately. |
| 131 | func (c *Consumer) HighWaterMarks() map[string]map[int32]int64 { return c.consumer.HighWaterMarks() } |
| 132 | |
| 133 | // MarkOffset marks the provided message as processed, alongside a metadata string |
| 134 | // that represents the state of the partition consumer at that point in time. The |
| 135 | // metadata string can be used by another consumer to restore that state, so it |
| 136 | // can resume consumption. |
| 137 | // |
| 138 | // Note: calling MarkOffset does not necessarily commit the offset to the backend |
| 139 | // store immediately for efficiency reasons, and it may never be committed if |
| 140 | // your application crashes. This means that you may end up processing the same |
| 141 | // message twice, and your processing should ideally be idempotent. |
| 142 | func (c *Consumer) MarkOffset(msg *sarama.ConsumerMessage, metadata string) { |
| 143 | if sub := c.subs.Fetch(msg.Topic, msg.Partition); sub != nil { |
| 144 | sub.MarkOffset(msg.Offset, metadata) |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | // MarkPartitionOffset marks an offset of the provided topic/partition as processed. |
| 149 | // See MarkOffset for additional explanation. |
| 150 | func (c *Consumer) MarkPartitionOffset(topic string, partition int32, offset int64, metadata string) { |
| 151 | if sub := c.subs.Fetch(topic, partition); sub != nil { |
| 152 | sub.MarkOffset(offset, metadata) |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | // MarkOffsets marks stashed offsets as processed. |
| 157 | // See MarkOffset for additional explanation. |
| 158 | func (c *Consumer) MarkOffsets(s *OffsetStash) { |
| 159 | s.mu.Lock() |
| 160 | defer s.mu.Unlock() |
| 161 | |
| 162 | for tp, info := range s.offsets { |
| 163 | if sub := c.subs.Fetch(tp.Topic, tp.Partition); sub != nil { |
| 164 | sub.MarkOffset(info.Offset, info.Metadata) |
| 165 | } |
| 166 | delete(s.offsets, tp) |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | // ResetOffsets marks the provided message as processed, alongside a metadata string |
| 171 | // that represents the state of the partition consumer at that point in time. The |
| 172 | // metadata string can be used by another consumer to restore that state, so it |
| 173 | // can resume consumption. |
| 174 | // |
| 175 | // Difference between ResetOffset and MarkOffset is that it allows to rewind to an earlier offset |
| 176 | func (c *Consumer) ResetOffset(msg *sarama.ConsumerMessage, metadata string) { |
| 177 | if sub := c.subs.Fetch(msg.Topic, msg.Partition); sub != nil { |
| 178 | sub.ResetOffset(msg.Offset, metadata) |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | // ResetPartitionOffset marks an offset of the provided topic/partition as processed. |
| 183 | // See ResetOffset for additional explanation. |
| 184 | func (c *Consumer) ResetPartitionOffset(topic string, partition int32, offset int64, metadata string) { |
| 185 | sub := c.subs.Fetch(topic, partition) |
| 186 | if sub != nil { |
| 187 | sub.ResetOffset(offset, metadata) |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | // ResetOffsets marks stashed offsets as processed. |
| 192 | // See ResetOffset for additional explanation. |
| 193 | func (c *Consumer) ResetOffsets(s *OffsetStash) { |
| 194 | s.mu.Lock() |
| 195 | defer s.mu.Unlock() |
| 196 | |
| 197 | for tp, info := range s.offsets { |
| 198 | if sub := c.subs.Fetch(tp.Topic, tp.Partition); sub != nil { |
| 199 | sub.ResetOffset(info.Offset, info.Metadata) |
| 200 | } |
| 201 | delete(s.offsets, tp) |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | // Subscriptions returns the consumed topics and partitions |
| 206 | func (c *Consumer) Subscriptions() map[string][]int32 { |
| 207 | return c.subs.Info() |
| 208 | } |
| 209 | |
| 210 | // CommitOffsets allows to manually commit previously marked offsets. By default there is no |
| 211 | // need to call this function as the consumer will commit offsets automatically |
| 212 | // using the Config.Consumer.Offsets.CommitInterval setting. |
| 213 | // |
| 214 | // Please be aware that calling this function during an internal rebalance cycle may return |
| 215 | // broker errors (e.g. sarama.ErrUnknownMemberId or sarama.ErrIllegalGeneration). |
| 216 | func (c *Consumer) CommitOffsets() error { |
| 217 | c.commitMu.Lock() |
| 218 | defer c.commitMu.Unlock() |
| 219 | |
| 220 | memberID, generationID := c.membership() |
| 221 | req := &sarama.OffsetCommitRequest{ |
| 222 | Version: 2, |
| 223 | ConsumerGroup: c.groupID, |
| 224 | ConsumerGroupGeneration: generationID, |
| 225 | ConsumerID: memberID, |
| 226 | RetentionTime: -1, |
| 227 | } |
| 228 | |
| 229 | if ns := c.client.config.Consumer.Offsets.Retention; ns != 0 { |
| 230 | req.RetentionTime = int64(ns / time.Millisecond) |
| 231 | } |
| 232 | |
| 233 | snap := c.subs.Snapshot() |
| 234 | dirty := false |
| 235 | for tp, state := range snap { |
| 236 | if state.Dirty { |
| 237 | dirty = true |
| 238 | req.AddBlock(tp.Topic, tp.Partition, state.Info.Offset, 0, state.Info.Metadata) |
| 239 | } |
| 240 | } |
| 241 | if !dirty { |
| 242 | return nil |
| 243 | } |
| 244 | |
| 245 | broker, err := c.client.Coordinator(c.groupID) |
| 246 | if err != nil { |
| 247 | c.closeCoordinator(broker, err) |
| 248 | return err |
| 249 | } |
| 250 | |
| 251 | resp, err := broker.CommitOffset(req) |
| 252 | if err != nil { |
| 253 | c.closeCoordinator(broker, err) |
| 254 | return err |
| 255 | } |
| 256 | |
| 257 | for topic, errs := range resp.Errors { |
| 258 | for partition, kerr := range errs { |
| 259 | if kerr != sarama.ErrNoError { |
| 260 | err = kerr |
| 261 | } else if state, ok := snap[topicPartition{topic, partition}]; ok { |
| 262 | if sub := c.subs.Fetch(topic, partition); sub != nil { |
| 263 | sub.markCommitted(state.Info.Offset) |
| 264 | } |
| 265 | } |
| 266 | } |
| 267 | } |
| 268 | return err |
| 269 | } |
| 270 | |
| 271 | // Close safely closes the consumer and releases all resources |
| 272 | func (c *Consumer) Close() (err error) { |
| 273 | c.closeOnce.Do(func() { |
| 274 | close(c.dying) |
| 275 | <-c.dead |
| 276 | |
| 277 | if e := c.release(); e != nil { |
| 278 | err = e |
| 279 | } |
| 280 | if e := c.consumer.Close(); e != nil { |
| 281 | err = e |
| 282 | } |
| 283 | close(c.messages) |
| 284 | close(c.errors) |
| 285 | |
| 286 | if e := c.leaveGroup(); e != nil { |
| 287 | err = e |
| 288 | } |
| 289 | close(c.partitions) |
| 290 | close(c.notifications) |
| 291 | |
| 292 | // drain |
| 293 | for range c.messages { |
| 294 | } |
| 295 | for range c.errors { |
| 296 | } |
| 297 | for p := range c.partitions { |
| 298 | _ = p.Close() |
| 299 | } |
| 300 | for range c.notifications { |
| 301 | } |
| 302 | |
| 303 | c.client.release() |
| 304 | if c.ownClient { |
| 305 | if e := c.client.Close(); e != nil { |
| 306 | err = e |
| 307 | } |
| 308 | } |
| 309 | }) |
| 310 | return |
| 311 | } |
| 312 | |
| 313 | func (c *Consumer) mainLoop() { |
| 314 | defer close(c.dead) |
| 315 | defer atomic.StoreInt32(&c.consuming, 0) |
| 316 | |
| 317 | for { |
| 318 | atomic.StoreInt32(&c.consuming, 0) |
| 319 | |
| 320 | // Check if close was requested |
| 321 | select { |
| 322 | case <-c.dying: |
| 323 | return |
| 324 | default: |
| 325 | } |
| 326 | |
| 327 | // Start next consume cycle |
| 328 | c.nextTick() |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | func (c *Consumer) nextTick() { |
| 333 | // Remember previous subscriptions |
| 334 | var notification *Notification |
| 335 | if c.client.config.Group.Return.Notifications { |
| 336 | notification = newNotification(c.subs.Info()) |
| 337 | } |
| 338 | |
| 339 | // Refresh coordinator |
| 340 | if err := c.refreshCoordinator(); err != nil { |
| 341 | c.rebalanceError(err, nil) |
| 342 | return |
| 343 | } |
| 344 | |
| 345 | // Release subscriptions |
| 346 | if err := c.release(); err != nil { |
| 347 | c.rebalanceError(err, nil) |
| 348 | return |
| 349 | } |
| 350 | |
| 351 | // Issue rebalance start notification |
| 352 | if c.client.config.Group.Return.Notifications { |
| 353 | c.handleNotification(notification) |
| 354 | } |
| 355 | |
| 356 | // Rebalance, fetch new subscriptions |
| 357 | subs, err := c.rebalance() |
| 358 | if err != nil { |
| 359 | c.rebalanceError(err, notification) |
| 360 | return |
| 361 | } |
| 362 | |
| 363 | // Coordinate loops, make sure everything is |
| 364 | // stopped on exit |
| 365 | tomb := newLoopTomb() |
| 366 | defer tomb.Close() |
| 367 | |
| 368 | // Start the heartbeat |
| 369 | tomb.Go(c.hbLoop) |
| 370 | |
| 371 | // Subscribe to topic/partitions |
| 372 | if err := c.subscribe(tomb, subs); err != nil { |
| 373 | c.rebalanceError(err, notification) |
| 374 | return |
| 375 | } |
| 376 | |
| 377 | // Update/issue notification with new claims |
| 378 | if c.client.config.Group.Return.Notifications { |
| 379 | notification = notification.success(subs) |
| 380 | c.handleNotification(notification) |
| 381 | } |
| 382 | |
| 383 | // Start topic watcher loop |
| 384 | tomb.Go(c.twLoop) |
| 385 | |
| 386 | // Start consuming and committing offsets |
| 387 | tomb.Go(c.cmLoop) |
| 388 | atomic.StoreInt32(&c.consuming, 1) |
| 389 | |
| 390 | // Wait for signals |
| 391 | select { |
| 392 | case <-tomb.Dying(): |
| 393 | case <-c.dying: |
| 394 | } |
| 395 | } |
| 396 | |
| 397 | // heartbeat loop, triggered by the mainLoop |
| 398 | func (c *Consumer) hbLoop(stopped <-chan none) { |
| 399 | ticker := time.NewTicker(c.client.config.Group.Heartbeat.Interval) |
| 400 | defer ticker.Stop() |
| 401 | |
| 402 | for { |
| 403 | select { |
| 404 | case <-ticker.C: |
| 405 | switch err := c.heartbeat(); err { |
| 406 | case nil, sarama.ErrNoError: |
| 407 | case sarama.ErrNotCoordinatorForConsumer, sarama.ErrRebalanceInProgress: |
| 408 | return |
| 409 | default: |
| 410 | c.handleError(&Error{Ctx: "heartbeat", error: err}) |
| 411 | return |
| 412 | } |
| 413 | case <-stopped: |
| 414 | return |
| 415 | case <-c.dying: |
| 416 | return |
| 417 | } |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | // topic watcher loop, triggered by the mainLoop |
| 422 | func (c *Consumer) twLoop(stopped <-chan none) { |
| 423 | ticker := time.NewTicker(c.client.config.Metadata.RefreshFrequency / 2) |
| 424 | defer ticker.Stop() |
| 425 | |
| 426 | for { |
| 427 | select { |
| 428 | case <-ticker.C: |
| 429 | topics, err := c.client.Topics() |
| 430 | if err != nil { |
| 431 | c.handleError(&Error{Ctx: "topics", error: err}) |
| 432 | return |
| 433 | } |
| 434 | |
| 435 | for _, topic := range topics { |
| 436 | if !c.isKnownCoreTopic(topic) && |
| 437 | !c.isKnownExtraTopic(topic) && |
| 438 | c.isPotentialExtraTopic(topic) { |
| 439 | return |
| 440 | } |
| 441 | } |
| 442 | case <-stopped: |
| 443 | return |
| 444 | case <-c.dying: |
| 445 | return |
| 446 | } |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | // commit loop, triggered by the mainLoop |
| 451 | func (c *Consumer) cmLoop(stopped <-chan none) { |
| 452 | ticker := time.NewTicker(c.client.config.Consumer.Offsets.CommitInterval) |
| 453 | defer ticker.Stop() |
| 454 | |
| 455 | for { |
| 456 | select { |
| 457 | case <-ticker.C: |
| 458 | if err := c.commitOffsetsWithRetry(c.client.config.Group.Offsets.Retry.Max); err != nil { |
| 459 | c.handleError(&Error{Ctx: "commit", error: err}) |
| 460 | return |
| 461 | } |
| 462 | case <-stopped: |
| 463 | return |
| 464 | case <-c.dying: |
| 465 | return |
| 466 | } |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | func (c *Consumer) rebalanceError(err error, n *Notification) { |
| 471 | if n != nil { |
| 472 | n.Type = RebalanceError |
| 473 | c.handleNotification(n) |
| 474 | } |
| 475 | |
| 476 | switch err { |
| 477 | case sarama.ErrRebalanceInProgress: |
| 478 | default: |
| 479 | c.handleError(&Error{Ctx: "rebalance", error: err}) |
| 480 | } |
| 481 | |
| 482 | select { |
| 483 | case <-c.dying: |
| 484 | case <-time.After(c.client.config.Metadata.Retry.Backoff): |
| 485 | } |
| 486 | } |
| 487 | |
| 488 | func (c *Consumer) handleNotification(n *Notification) { |
| 489 | if c.client.config.Group.Return.Notifications { |
| 490 | select { |
| 491 | case c.notifications <- n: |
| 492 | case <-c.dying: |
| 493 | return |
| 494 | } |
| 495 | } |
| 496 | } |
| 497 | |
| 498 | func (c *Consumer) handleError(e *Error) { |
| 499 | if c.client.config.Consumer.Return.Errors { |
| 500 | select { |
| 501 | case c.errors <- e: |
| 502 | case <-c.dying: |
| 503 | return |
| 504 | } |
| 505 | } else { |
| 506 | sarama.Logger.Printf("%s error: %s\n", e.Ctx, e.Error()) |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | // Releases the consumer and commits offsets, called from rebalance() and Close() |
| 511 | func (c *Consumer) release() (err error) { |
| 512 | // Stop all consumers |
| 513 | c.subs.Stop() |
| 514 | |
| 515 | // Clear subscriptions on exit |
| 516 | defer c.subs.Clear() |
| 517 | |
| 518 | // Wait for messages to be processed |
| 519 | timeout := time.NewTimer(c.client.config.Group.Offsets.Synchronization.DwellTime) |
| 520 | defer timeout.Stop() |
| 521 | |
| 522 | select { |
| 523 | case <-c.dying: |
| 524 | case <-timeout.C: |
| 525 | } |
| 526 | |
| 527 | // Commit offsets, continue on errors |
| 528 | if e := c.commitOffsetsWithRetry(c.client.config.Group.Offsets.Retry.Max); e != nil { |
| 529 | err = e |
| 530 | } |
| 531 | |
| 532 | return |
| 533 | } |
| 534 | |
| 535 | // -------------------------------------------------------------------- |
| 536 | |
| 537 | // Performs a heartbeat, part of the mainLoop() |
| 538 | func (c *Consumer) heartbeat() error { |
| 539 | broker, err := c.client.Coordinator(c.groupID) |
| 540 | if err != nil { |
| 541 | c.closeCoordinator(broker, err) |
| 542 | return err |
| 543 | } |
| 544 | |
| 545 | memberID, generationID := c.membership() |
| 546 | resp, err := broker.Heartbeat(&sarama.HeartbeatRequest{ |
| 547 | GroupId: c.groupID, |
| 548 | MemberId: memberID, |
| 549 | GenerationId: generationID, |
| 550 | }) |
| 551 | if err != nil { |
| 552 | c.closeCoordinator(broker, err) |
| 553 | return err |
| 554 | } |
| 555 | return resp.Err |
| 556 | } |
| 557 | |
| 558 | // Performs a rebalance, part of the mainLoop() |
| 559 | func (c *Consumer) rebalance() (map[string][]int32, error) { |
| 560 | memberID, _ := c.membership() |
| 561 | sarama.Logger.Printf("cluster/consumer %s rebalance\n", memberID) |
| 562 | |
| 563 | allTopics, err := c.client.Topics() |
| 564 | if err != nil { |
| 565 | return nil, err |
| 566 | } |
| 567 | c.extraTopics = c.selectExtraTopics(allTopics) |
| 568 | sort.Strings(c.extraTopics) |
| 569 | |
| 570 | // Re-join consumer group |
| 571 | strategy, err := c.joinGroup() |
| 572 | switch { |
| 573 | case err == sarama.ErrUnknownMemberId: |
| 574 | c.membershipMu.Lock() |
| 575 | c.memberID = "" |
| 576 | c.membershipMu.Unlock() |
| 577 | return nil, err |
| 578 | case err != nil: |
| 579 | return nil, err |
| 580 | } |
| 581 | |
| 582 | // Sync consumer group state, fetch subscriptions |
| 583 | subs, err := c.syncGroup(strategy) |
| 584 | switch { |
| 585 | case err == sarama.ErrRebalanceInProgress: |
| 586 | return nil, err |
| 587 | case err != nil: |
| 588 | _ = c.leaveGroup() |
| 589 | return nil, err |
| 590 | } |
| 591 | return subs, nil |
| 592 | } |
| 593 | |
| 594 | // Performs the subscription, part of the mainLoop() |
| 595 | func (c *Consumer) subscribe(tomb *loopTomb, subs map[string][]int32) error { |
| 596 | // fetch offsets |
| 597 | offsets, err := c.fetchOffsets(subs) |
| 598 | if err != nil { |
| 599 | _ = c.leaveGroup() |
| 600 | return err |
| 601 | } |
| 602 | |
| 603 | // create consumers in parallel |
| 604 | var mu sync.Mutex |
| 605 | var wg sync.WaitGroup |
| 606 | |
| 607 | for topic, partitions := range subs { |
| 608 | for _, partition := range partitions { |
| 609 | wg.Add(1) |
| 610 | |
| 611 | info := offsets[topic][partition] |
| 612 | go func(topic string, partition int32) { |
| 613 | if e := c.createConsumer(tomb, topic, partition, info); e != nil { |
| 614 | mu.Lock() |
| 615 | err = e |
| 616 | mu.Unlock() |
| 617 | } |
| 618 | wg.Done() |
| 619 | }(topic, partition) |
| 620 | } |
| 621 | } |
| 622 | wg.Wait() |
| 623 | |
| 624 | if err != nil { |
| 625 | _ = c.release() |
| 626 | _ = c.leaveGroup() |
| 627 | } |
| 628 | return err |
| 629 | } |
| 630 | |
| 631 | // -------------------------------------------------------------------- |
| 632 | |
| 633 | // Send a request to the broker to join group on rebalance() |
| 634 | func (c *Consumer) joinGroup() (*balancer, error) { |
| 635 | memberID, _ := c.membership() |
| 636 | req := &sarama.JoinGroupRequest{ |
| 637 | GroupId: c.groupID, |
| 638 | MemberId: memberID, |
| 639 | SessionTimeout: int32(c.client.config.Group.Session.Timeout / time.Millisecond), |
| 640 | ProtocolType: "consumer", |
| 641 | } |
| 642 | |
| 643 | meta := &sarama.ConsumerGroupMemberMetadata{ |
| 644 | Version: 1, |
| 645 | Topics: append(c.coreTopics, c.extraTopics...), |
| 646 | UserData: c.client.config.Group.Member.UserData, |
| 647 | } |
| 648 | err := req.AddGroupProtocolMetadata(string(StrategyRange), meta) |
| 649 | if err != nil { |
| 650 | return nil, err |
| 651 | } |
| 652 | err = req.AddGroupProtocolMetadata(string(StrategyRoundRobin), meta) |
| 653 | if err != nil { |
| 654 | return nil, err |
| 655 | } |
| 656 | |
| 657 | broker, err := c.client.Coordinator(c.groupID) |
| 658 | if err != nil { |
| 659 | c.closeCoordinator(broker, err) |
| 660 | return nil, err |
| 661 | } |
| 662 | |
| 663 | resp, err := broker.JoinGroup(req) |
| 664 | if err != nil { |
| 665 | c.closeCoordinator(broker, err) |
| 666 | return nil, err |
| 667 | } else if resp.Err != sarama.ErrNoError { |
| 668 | c.closeCoordinator(broker, resp.Err) |
| 669 | return nil, resp.Err |
| 670 | } |
| 671 | |
| 672 | var strategy *balancer |
| 673 | if resp.LeaderId == resp.MemberId { |
| 674 | members, err := resp.GetMembers() |
| 675 | if err != nil { |
| 676 | return nil, err |
| 677 | } |
| 678 | |
| 679 | strategy, err = newBalancerFromMeta(c.client, members) |
| 680 | if err != nil { |
| 681 | return nil, err |
| 682 | } |
| 683 | } |
| 684 | |
| 685 | c.membershipMu.Lock() |
| 686 | c.memberID = resp.MemberId |
| 687 | c.generationID = resp.GenerationId |
| 688 | c.membershipMu.Unlock() |
| 689 | |
| 690 | return strategy, nil |
| 691 | } |
| 692 | |
| 693 | // Send a request to the broker to sync the group on rebalance(). |
| 694 | // Returns a list of topics and partitions to consume. |
| 695 | func (c *Consumer) syncGroup(strategy *balancer) (map[string][]int32, error) { |
| 696 | memberID, generationID := c.membership() |
| 697 | req := &sarama.SyncGroupRequest{ |
| 698 | GroupId: c.groupID, |
| 699 | MemberId: memberID, |
| 700 | GenerationId: generationID, |
| 701 | } |
| 702 | |
| 703 | if strategy != nil { |
| 704 | for memberID, topics := range strategy.Perform(c.client.config.Group.PartitionStrategy) { |
| 705 | if err := req.AddGroupAssignmentMember(memberID, &sarama.ConsumerGroupMemberAssignment{ |
| 706 | Topics: topics, |
| 707 | }); err != nil { |
| 708 | return nil, err |
| 709 | } |
| 710 | } |
| 711 | } |
| 712 | |
| 713 | broker, err := c.client.Coordinator(c.groupID) |
| 714 | if err != nil { |
| 715 | c.closeCoordinator(broker, err) |
| 716 | return nil, err |
| 717 | } |
| 718 | |
| 719 | resp, err := broker.SyncGroup(req) |
| 720 | if err != nil { |
| 721 | c.closeCoordinator(broker, err) |
| 722 | return nil, err |
| 723 | } else if resp.Err != sarama.ErrNoError { |
| 724 | c.closeCoordinator(broker, resp.Err) |
| 725 | return nil, resp.Err |
| 726 | } |
| 727 | |
| 728 | // Return if there is nothing to subscribe to |
| 729 | if len(resp.MemberAssignment) == 0 { |
| 730 | return nil, nil |
| 731 | } |
| 732 | |
| 733 | // Get assigned subscriptions |
| 734 | members, err := resp.GetMemberAssignment() |
| 735 | if err != nil { |
| 736 | return nil, err |
| 737 | } |
| 738 | |
| 739 | // Sort partitions, for each topic |
| 740 | for topic := range members.Topics { |
| 741 | sort.Sort(int32Slice(members.Topics[topic])) |
| 742 | } |
| 743 | return members.Topics, nil |
| 744 | } |
| 745 | |
| 746 | // Fetches latest committed offsets for all subscriptions |
| 747 | func (c *Consumer) fetchOffsets(subs map[string][]int32) (map[string]map[int32]offsetInfo, error) { |
| 748 | offsets := make(map[string]map[int32]offsetInfo, len(subs)) |
| 749 | req := &sarama.OffsetFetchRequest{ |
| 750 | Version: 1, |
| 751 | ConsumerGroup: c.groupID, |
| 752 | } |
| 753 | |
| 754 | for topic, partitions := range subs { |
| 755 | offsets[topic] = make(map[int32]offsetInfo, len(partitions)) |
| 756 | for _, partition := range partitions { |
| 757 | offsets[topic][partition] = offsetInfo{Offset: -1} |
| 758 | req.AddPartition(topic, partition) |
| 759 | } |
| 760 | } |
| 761 | |
| 762 | broker, err := c.client.Coordinator(c.groupID) |
| 763 | if err != nil { |
| 764 | c.closeCoordinator(broker, err) |
| 765 | return nil, err |
| 766 | } |
| 767 | |
| 768 | resp, err := broker.FetchOffset(req) |
| 769 | if err != nil { |
| 770 | c.closeCoordinator(broker, err) |
| 771 | return nil, err |
| 772 | } |
| 773 | |
| 774 | for topic, partitions := range subs { |
| 775 | for _, partition := range partitions { |
| 776 | block := resp.GetBlock(topic, partition) |
| 777 | if block == nil { |
| 778 | return nil, sarama.ErrIncompleteResponse |
| 779 | } |
| 780 | |
| 781 | if block.Err == sarama.ErrNoError { |
| 782 | offsets[topic][partition] = offsetInfo{Offset: block.Offset, Metadata: block.Metadata} |
| 783 | } else { |
| 784 | return nil, block.Err |
| 785 | } |
| 786 | } |
| 787 | } |
| 788 | return offsets, nil |
| 789 | } |
| 790 | |
| 791 | // Send a request to the broker to leave the group on failes rebalance() and on Close() |
| 792 | func (c *Consumer) leaveGroup() error { |
| 793 | broker, err := c.client.Coordinator(c.groupID) |
| 794 | if err != nil { |
| 795 | c.closeCoordinator(broker, err) |
| 796 | return err |
| 797 | } |
| 798 | |
| 799 | memberID, _ := c.membership() |
| 800 | if _, err = broker.LeaveGroup(&sarama.LeaveGroupRequest{ |
| 801 | GroupId: c.groupID, |
| 802 | MemberId: memberID, |
| 803 | }); err != nil { |
| 804 | c.closeCoordinator(broker, err) |
| 805 | } |
| 806 | return err |
| 807 | } |
| 808 | |
| 809 | // -------------------------------------------------------------------- |
| 810 | |
| 811 | func (c *Consumer) createConsumer(tomb *loopTomb, topic string, partition int32, info offsetInfo) error { |
| 812 | memberID, _ := c.membership() |
| 813 | sarama.Logger.Printf("cluster/consumer %s consume %s/%d from %d\n", memberID, topic, partition, info.NextOffset(c.client.config.Consumer.Offsets.Initial)) |
| 814 | |
| 815 | // Create partitionConsumer |
| 816 | pc, err := newPartitionConsumer(c.consumer, topic, partition, info, c.client.config.Consumer.Offsets.Initial) |
| 817 | if err != nil { |
| 818 | return err |
| 819 | } |
| 820 | |
| 821 | // Store in subscriptions |
| 822 | c.subs.Store(topic, partition, pc) |
| 823 | |
| 824 | // Start partition consumer goroutine |
| 825 | tomb.Go(func(stopper <-chan none) { |
| 826 | if c.client.config.Group.Mode == ConsumerModePartitions { |
| 827 | pc.waitFor(stopper, c.errors) |
| 828 | } else { |
| 829 | pc.multiplex(stopper, c.messages, c.errors) |
| 830 | } |
| 831 | }) |
| 832 | |
| 833 | if c.client.config.Group.Mode == ConsumerModePartitions { |
| 834 | c.partitions <- pc |
| 835 | } |
| 836 | return nil |
| 837 | } |
| 838 | |
| 839 | func (c *Consumer) commitOffsetsWithRetry(retries int) error { |
| 840 | err := c.CommitOffsets() |
| 841 | if err != nil && retries > 0 { |
| 842 | return c.commitOffsetsWithRetry(retries - 1) |
| 843 | } |
| 844 | return err |
| 845 | } |
| 846 | |
| 847 | func (c *Consumer) closeCoordinator(broker *sarama.Broker, err error) { |
| 848 | if broker != nil { |
| 849 | _ = broker.Close() |
| 850 | } |
| 851 | |
| 852 | switch err { |
| 853 | case sarama.ErrConsumerCoordinatorNotAvailable, sarama.ErrNotCoordinatorForConsumer: |
| 854 | _ = c.client.RefreshCoordinator(c.groupID) |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | func (c *Consumer) selectExtraTopics(allTopics []string) []string { |
| 859 | extra := allTopics[:0] |
| 860 | for _, topic := range allTopics { |
| 861 | if !c.isKnownCoreTopic(topic) && c.isPotentialExtraTopic(topic) { |
| 862 | extra = append(extra, topic) |
| 863 | } |
| 864 | } |
| 865 | return extra |
| 866 | } |
| 867 | |
| 868 | func (c *Consumer) isKnownCoreTopic(topic string) bool { |
| 869 | pos := sort.SearchStrings(c.coreTopics, topic) |
| 870 | return pos < len(c.coreTopics) && c.coreTopics[pos] == topic |
| 871 | } |
| 872 | |
| 873 | func (c *Consumer) isKnownExtraTopic(topic string) bool { |
| 874 | pos := sort.SearchStrings(c.extraTopics, topic) |
| 875 | return pos < len(c.extraTopics) && c.extraTopics[pos] == topic |
| 876 | } |
| 877 | |
| 878 | func (c *Consumer) isPotentialExtraTopic(topic string) bool { |
| 879 | rx := c.client.config.Group.Topics |
| 880 | if rx.Blacklist != nil && rx.Blacklist.MatchString(topic) { |
| 881 | return false |
| 882 | } |
| 883 | if rx.Whitelist != nil && rx.Whitelist.MatchString(topic) { |
| 884 | return true |
| 885 | } |
| 886 | return false |
| 887 | } |
| 888 | |
| 889 | func (c *Consumer) refreshCoordinator() error { |
| 890 | if err := c.refreshMetadata(); err != nil { |
| 891 | return err |
| 892 | } |
| 893 | return c.client.RefreshCoordinator(c.groupID) |
| 894 | } |
| 895 | |
| 896 | func (c *Consumer) refreshMetadata() (err error) { |
| 897 | if c.client.config.Metadata.Full { |
| 898 | err = c.client.RefreshMetadata() |
| 899 | } else { |
| 900 | var topics []string |
| 901 | if topics, err = c.client.Topics(); err == nil && len(topics) != 0 { |
| 902 | err = c.client.RefreshMetadata(topics...) |
| 903 | } |
| 904 | } |
| 905 | |
| 906 | // maybe we didn't have authorization to describe all topics |
| 907 | switch err { |
| 908 | case sarama.ErrTopicAuthorizationFailed: |
| 909 | err = c.client.RefreshMetadata(c.coreTopics...) |
| 910 | } |
| 911 | return |
| 912 | } |
| 913 | |
| 914 | func (c *Consumer) membership() (memberID string, generationID int32) { |
| 915 | c.membershipMu.RLock() |
| 916 | memberID, generationID = c.memberID, c.generationID |
| 917 | c.membershipMu.RUnlock() |
| 918 | return |
| 919 | } |