Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 1 | package sarama |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "sort" |
| 8 | "sync" |
| 9 | "time" |
| 10 | ) |
| 11 | |
| 12 | // ErrClosedConsumerGroup is the error returned when a method is called on a consumer group that has been closed. |
| 13 | var ErrClosedConsumerGroup = errors.New("kafka: tried to use a consumer group that was closed") |
| 14 | |
| 15 | // ConsumerGroup is responsible for dividing up processing of topics and partitions |
| 16 | // over a collection of processes (the members of the consumer group). |
| 17 | type ConsumerGroup interface { |
| 18 | // Consume joins a cluster of consumers for a given list of topics and |
| 19 | // starts a blocking ConsumerGroupSession through the ConsumerGroupHandler. |
| 20 | // |
| 21 | // The life-cycle of a session is represented by the following steps: |
| 22 | // |
| 23 | // 1. The consumers join the group (as explained in https://kafka.apache.org/documentation/#intro_consumers) |
| 24 | // and is assigned their "fair share" of partitions, aka 'claims'. |
| 25 | // 2. Before processing starts, the handler's Setup() hook is called to notify the user |
| 26 | // of the claims and allow any necessary preparation or alteration of state. |
| 27 | // 3. For each of the assigned claims the handler's ConsumeClaim() function is then called |
| 28 | // in a separate goroutine which requires it to be thread-safe. Any state must be carefully protected |
| 29 | // from concurrent reads/writes. |
| 30 | // 4. The session will persist until one of the ConsumeClaim() functions exits. This can be either when the |
| 31 | // parent context is cancelled or when a server-side rebalance cycle is initiated. |
| 32 | // 5. Once all the ConsumeClaim() loops have exited, the handler's Cleanup() hook is called |
| 33 | // to allow the user to perform any final tasks before a rebalance. |
| 34 | // 6. Finally, marked offsets are committed one last time before claims are released. |
| 35 | // |
| 36 | // Please note, that once a rebalance is triggered, sessions must be completed within |
| 37 | // Config.Consumer.Group.Rebalance.Timeout. This means that ConsumeClaim() functions must exit |
| 38 | // as quickly as possible to allow time for Cleanup() and the final offset commit. If the timeout |
| 39 | // is exceeded, the consumer will be removed from the group by Kafka, which will cause offset |
| 40 | // commit failures. |
| 41 | Consume(ctx context.Context, topics []string, handler ConsumerGroupHandler) error |
| 42 | |
| 43 | // Errors returns a read channel of errors that occurred during the consumer life-cycle. |
| 44 | // By default, errors are logged and not returned over this channel. |
| 45 | // If you want to implement any custom error handling, set your config's |
| 46 | // Consumer.Return.Errors setting to true, and read from this channel. |
| 47 | Errors() <-chan error |
| 48 | |
| 49 | // Close stops the ConsumerGroup and detaches any running sessions. It is required to call |
| 50 | // this function before the object passes out of scope, as it will otherwise leak memory. |
| 51 | Close() error |
| 52 | } |
| 53 | |
| 54 | type consumerGroup struct { |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 55 | client Client |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 56 | |
| 57 | config *Config |
| 58 | consumer Consumer |
| 59 | groupID string |
| 60 | memberID string |
| 61 | errors chan error |
| 62 | |
| 63 | lock sync.Mutex |
| 64 | closed chan none |
| 65 | closeOnce sync.Once |
| 66 | } |
| 67 | |
| 68 | // NewConsumerGroup creates a new consumer group the given broker addresses and configuration. |
| 69 | func NewConsumerGroup(addrs []string, groupID string, config *Config) (ConsumerGroup, error) { |
| 70 | client, err := NewClient(addrs, config) |
| 71 | if err != nil { |
| 72 | return nil, err |
| 73 | } |
| 74 | |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 75 | c, err := newConsumerGroup(groupID, client) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 76 | if err != nil { |
| 77 | _ = client.Close() |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 78 | } |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 79 | return c, err |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 80 | } |
| 81 | |
| 82 | // NewConsumerGroupFromClient creates a new consumer group using the given client. It is still |
| 83 | // necessary to call Close() on the underlying client when shutting down this consumer. |
| 84 | // PLEASE NOTE: consumer groups can only re-use but not share clients. |
| 85 | func NewConsumerGroupFromClient(groupID string, client Client) (ConsumerGroup, error) { |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 86 | // For clients passed in by the client, ensure we don't |
| 87 | // call Close() on it. |
| 88 | cli := &nopCloserClient{client} |
| 89 | return newConsumerGroup(groupID, cli) |
| 90 | } |
| 91 | |
| 92 | func newConsumerGroup(groupID string, client Client) (ConsumerGroup, error) { |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 93 | config := client.Config() |
| 94 | if !config.Version.IsAtLeast(V0_10_2_0) { |
| 95 | return nil, ConfigurationError("consumer groups require Version to be >= V0_10_2_0") |
| 96 | } |
| 97 | |
| 98 | consumer, err := NewConsumerFromClient(client) |
| 99 | if err != nil { |
| 100 | return nil, err |
| 101 | } |
| 102 | |
| 103 | return &consumerGroup{ |
| 104 | client: client, |
| 105 | consumer: consumer, |
| 106 | config: config, |
| 107 | groupID: groupID, |
| 108 | errors: make(chan error, config.ChannelBufferSize), |
| 109 | closed: make(chan none), |
| 110 | }, nil |
| 111 | } |
| 112 | |
| 113 | // Errors implements ConsumerGroup. |
| 114 | func (c *consumerGroup) Errors() <-chan error { return c.errors } |
| 115 | |
| 116 | // Close implements ConsumerGroup. |
| 117 | func (c *consumerGroup) Close() (err error) { |
| 118 | c.closeOnce.Do(func() { |
| 119 | close(c.closed) |
| 120 | |
| 121 | c.lock.Lock() |
| 122 | defer c.lock.Unlock() |
| 123 | |
| 124 | // leave group |
| 125 | if e := c.leave(); e != nil { |
| 126 | err = e |
| 127 | } |
| 128 | |
| 129 | // drain errors |
| 130 | go func() { |
| 131 | close(c.errors) |
| 132 | }() |
| 133 | for e := range c.errors { |
| 134 | err = e |
| 135 | } |
| 136 | |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 137 | if e := c.client.Close(); e != nil { |
| 138 | err = e |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 139 | } |
| 140 | }) |
| 141 | return |
| 142 | } |
| 143 | |
| 144 | // Consume implements ConsumerGroup. |
| 145 | func (c *consumerGroup) Consume(ctx context.Context, topics []string, handler ConsumerGroupHandler) error { |
| 146 | // Ensure group is not closed |
| 147 | select { |
| 148 | case <-c.closed: |
| 149 | return ErrClosedConsumerGroup |
| 150 | default: |
| 151 | } |
| 152 | |
| 153 | c.lock.Lock() |
| 154 | defer c.lock.Unlock() |
| 155 | |
| 156 | // Quick exit when no topics are provided |
| 157 | if len(topics) == 0 { |
| 158 | return fmt.Errorf("no topics provided") |
| 159 | } |
| 160 | |
| 161 | // Refresh metadata for requested topics |
| 162 | if err := c.client.RefreshMetadata(topics...); err != nil { |
| 163 | return err |
| 164 | } |
| 165 | |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 166 | // Init session |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 167 | sess, err := c.newSession(ctx, topics, handler, c.config.Consumer.Group.Rebalance.Retry.Max) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 168 | if err == ErrClosedClient { |
| 169 | return ErrClosedConsumerGroup |
| 170 | } else if err != nil { |
| 171 | return err |
| 172 | } |
| 173 | |
| 174 | // Wait for session exit signal |
| 175 | <-sess.ctx.Done() |
| 176 | |
| 177 | // Gracefully release session claims |
| 178 | return sess.release(true) |
| 179 | } |
| 180 | |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 181 | func (c *consumerGroup) retryNewSession(ctx context.Context, topics []string, handler ConsumerGroupHandler, retries int, refreshCoordinator bool) (*consumerGroupSession, error) { |
| 182 | select { |
| 183 | case <-c.closed: |
| 184 | return nil, ErrClosedConsumerGroup |
| 185 | case <-time.After(c.config.Consumer.Group.Rebalance.Retry.Backoff): |
| 186 | } |
| 187 | |
| 188 | if refreshCoordinator { |
| 189 | err := c.client.RefreshCoordinator(c.groupID) |
| 190 | if err != nil { |
| 191 | return c.retryNewSession(ctx, topics, handler, retries, true) |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | return c.newSession(ctx, topics, handler, retries-1) |
| 196 | } |
| 197 | |
| 198 | func (c *consumerGroup) newSession(ctx context.Context, topics []string, handler ConsumerGroupHandler, retries int) (*consumerGroupSession, error) { |
| 199 | coordinator, err := c.client.Coordinator(c.groupID) |
| 200 | if err != nil { |
| 201 | if retries <= 0 { |
| 202 | return nil, err |
| 203 | } |
| 204 | |
| 205 | return c.retryNewSession(ctx, topics, handler, retries, true) |
| 206 | } |
| 207 | |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 208 | // Join consumer group |
| 209 | join, err := c.joinGroupRequest(coordinator, topics) |
| 210 | if err != nil { |
| 211 | _ = coordinator.Close() |
| 212 | return nil, err |
| 213 | } |
| 214 | switch join.Err { |
| 215 | case ErrNoError: |
| 216 | c.memberID = join.MemberId |
| 217 | case ErrUnknownMemberId, ErrIllegalGeneration: // reset member ID and retry immediately |
| 218 | c.memberID = "" |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 219 | return c.newSession(ctx, topics, handler, retries) |
| 220 | case ErrNotCoordinatorForConsumer: // retry after backoff with coordinator refresh |
| 221 | if retries <= 0 { |
| 222 | return nil, join.Err |
| 223 | } |
| 224 | |
| 225 | return c.retryNewSession(ctx, topics, handler, retries, true) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 226 | case ErrRebalanceInProgress: // retry after backoff |
| 227 | if retries <= 0 { |
| 228 | return nil, join.Err |
| 229 | } |
| 230 | |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 231 | return c.retryNewSession(ctx, topics, handler, retries, false) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 232 | default: |
| 233 | return nil, join.Err |
| 234 | } |
| 235 | |
| 236 | // Prepare distribution plan if we joined as the leader |
| 237 | var plan BalanceStrategyPlan |
| 238 | if join.LeaderId == join.MemberId { |
| 239 | members, err := join.GetMembers() |
| 240 | if err != nil { |
| 241 | return nil, err |
| 242 | } |
| 243 | |
| 244 | plan, err = c.balance(members) |
| 245 | if err != nil { |
| 246 | return nil, err |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | // Sync consumer group |
| 251 | sync, err := c.syncGroupRequest(coordinator, plan, join.GenerationId) |
| 252 | if err != nil { |
| 253 | _ = coordinator.Close() |
| 254 | return nil, err |
| 255 | } |
| 256 | switch sync.Err { |
| 257 | case ErrNoError: |
| 258 | case ErrUnknownMemberId, ErrIllegalGeneration: // reset member ID and retry immediately |
| 259 | c.memberID = "" |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 260 | return c.newSession(ctx, topics, handler, retries) |
| 261 | case ErrNotCoordinatorForConsumer: // retry after backoff with coordinator refresh |
| 262 | if retries <= 0 { |
| 263 | return nil, sync.Err |
| 264 | } |
| 265 | |
| 266 | return c.retryNewSession(ctx, topics, handler, retries, true) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 267 | case ErrRebalanceInProgress: // retry after backoff |
| 268 | if retries <= 0 { |
| 269 | return nil, sync.Err |
| 270 | } |
| 271 | |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 272 | return c.retryNewSession(ctx, topics, handler, retries, false) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 273 | default: |
| 274 | return nil, sync.Err |
| 275 | } |
| 276 | |
| 277 | // Retrieve and sort claims |
| 278 | var claims map[string][]int32 |
| 279 | if len(sync.MemberAssignment) > 0 { |
| 280 | members, err := sync.GetMemberAssignment() |
| 281 | if err != nil { |
| 282 | return nil, err |
| 283 | } |
| 284 | claims = members.Topics |
| 285 | |
| 286 | for _, partitions := range claims { |
| 287 | sort.Sort(int32Slice(partitions)) |
| 288 | } |
| 289 | } |
| 290 | |
| 291 | return newConsumerGroupSession(ctx, c, claims, join.MemberId, join.GenerationId, handler) |
| 292 | } |
| 293 | |
| 294 | func (c *consumerGroup) joinGroupRequest(coordinator *Broker, topics []string) (*JoinGroupResponse, error) { |
| 295 | req := &JoinGroupRequest{ |
| 296 | GroupId: c.groupID, |
| 297 | MemberId: c.memberID, |
| 298 | SessionTimeout: int32(c.config.Consumer.Group.Session.Timeout / time.Millisecond), |
| 299 | ProtocolType: "consumer", |
| 300 | } |
| 301 | if c.config.Version.IsAtLeast(V0_10_1_0) { |
| 302 | req.Version = 1 |
| 303 | req.RebalanceTimeout = int32(c.config.Consumer.Group.Rebalance.Timeout / time.Millisecond) |
| 304 | } |
| 305 | |
| 306 | meta := &ConsumerGroupMemberMetadata{ |
| 307 | Topics: topics, |
| 308 | UserData: c.config.Consumer.Group.Member.UserData, |
| 309 | } |
| 310 | strategy := c.config.Consumer.Group.Rebalance.Strategy |
| 311 | if err := req.AddGroupProtocolMetadata(strategy.Name(), meta); err != nil { |
| 312 | return nil, err |
| 313 | } |
| 314 | |
| 315 | return coordinator.JoinGroup(req) |
| 316 | } |
| 317 | |
| 318 | func (c *consumerGroup) syncGroupRequest(coordinator *Broker, plan BalanceStrategyPlan, generationID int32) (*SyncGroupResponse, error) { |
| 319 | req := &SyncGroupRequest{ |
| 320 | GroupId: c.groupID, |
| 321 | MemberId: c.memberID, |
| 322 | GenerationId: generationID, |
| 323 | } |
| 324 | for memberID, topics := range plan { |
| 325 | err := req.AddGroupAssignmentMember(memberID, &ConsumerGroupMemberAssignment{ |
| 326 | Topics: topics, |
| 327 | }) |
| 328 | if err != nil { |
| 329 | return nil, err |
| 330 | } |
| 331 | } |
| 332 | return coordinator.SyncGroup(req) |
| 333 | } |
| 334 | |
| 335 | func (c *consumerGroup) heartbeatRequest(coordinator *Broker, memberID string, generationID int32) (*HeartbeatResponse, error) { |
| 336 | req := &HeartbeatRequest{ |
| 337 | GroupId: c.groupID, |
| 338 | MemberId: memberID, |
| 339 | GenerationId: generationID, |
| 340 | } |
| 341 | |
| 342 | return coordinator.Heartbeat(req) |
| 343 | } |
| 344 | |
| 345 | func (c *consumerGroup) balance(members map[string]ConsumerGroupMemberMetadata) (BalanceStrategyPlan, error) { |
| 346 | topics := make(map[string][]int32) |
| 347 | for _, meta := range members { |
| 348 | for _, topic := range meta.Topics { |
| 349 | topics[topic] = nil |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | for topic := range topics { |
| 354 | partitions, err := c.client.Partitions(topic) |
| 355 | if err != nil { |
| 356 | return nil, err |
| 357 | } |
| 358 | topics[topic] = partitions |
| 359 | } |
| 360 | |
| 361 | strategy := c.config.Consumer.Group.Rebalance.Strategy |
| 362 | return strategy.Plan(members, topics) |
| 363 | } |
| 364 | |
| 365 | // Leaves the cluster, called by Close, protected by lock. |
| 366 | func (c *consumerGroup) leave() error { |
| 367 | if c.memberID == "" { |
| 368 | return nil |
| 369 | } |
| 370 | |
| 371 | coordinator, err := c.client.Coordinator(c.groupID) |
| 372 | if err != nil { |
| 373 | return err |
| 374 | } |
| 375 | |
| 376 | resp, err := coordinator.LeaveGroup(&LeaveGroupRequest{ |
| 377 | GroupId: c.groupID, |
| 378 | MemberId: c.memberID, |
| 379 | }) |
| 380 | if err != nil { |
| 381 | _ = coordinator.Close() |
| 382 | return err |
| 383 | } |
| 384 | |
| 385 | // Unset memberID |
| 386 | c.memberID = "" |
| 387 | |
| 388 | // Check response |
| 389 | switch resp.Err { |
| 390 | case ErrRebalanceInProgress, ErrUnknownMemberId, ErrNoError: |
| 391 | return nil |
| 392 | default: |
| 393 | return resp.Err |
| 394 | } |
| 395 | } |
| 396 | |
| 397 | func (c *consumerGroup) handleError(err error, topic string, partition int32) { |
| 398 | select { |
| 399 | case <-c.closed: |
| 400 | return |
| 401 | default: |
| 402 | } |
| 403 | |
| 404 | if _, ok := err.(*ConsumerError); !ok && topic != "" && partition > -1 { |
| 405 | err = &ConsumerError{ |
| 406 | Topic: topic, |
| 407 | Partition: partition, |
| 408 | Err: err, |
| 409 | } |
| 410 | } |
| 411 | |
| 412 | if c.config.Consumer.Return.Errors { |
| 413 | select { |
| 414 | case c.errors <- err: |
| 415 | default: |
| 416 | } |
| 417 | } else { |
| 418 | Logger.Println(err) |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | // -------------------------------------------------------------------- |
| 423 | |
| 424 | // ConsumerGroupSession represents a consumer group member session. |
| 425 | type ConsumerGroupSession interface { |
| 426 | // Claims returns information about the claimed partitions by topic. |
| 427 | Claims() map[string][]int32 |
| 428 | |
| 429 | // MemberID returns the cluster member ID. |
| 430 | MemberID() string |
| 431 | |
| 432 | // GenerationID returns the current generation ID. |
| 433 | GenerationID() int32 |
| 434 | |
| 435 | // MarkOffset marks the provided offset, alongside a metadata string |
| 436 | // that represents the state of the partition consumer at that point in time. The |
| 437 | // metadata string can be used by another consumer to restore that state, so it |
| 438 | // can resume consumption. |
| 439 | // |
| 440 | // To follow upstream conventions, you are expected to mark the offset of the |
| 441 | // next message to read, not the last message read. Thus, when calling `MarkOffset` |
| 442 | // you should typically add one to the offset of the last consumed message. |
| 443 | // |
| 444 | // Note: calling MarkOffset does not necessarily commit the offset to the backend |
| 445 | // store immediately for efficiency reasons, and it may never be committed if |
| 446 | // your application crashes. This means that you may end up processing the same |
| 447 | // message twice, and your processing should ideally be idempotent. |
| 448 | MarkOffset(topic string, partition int32, offset int64, metadata string) |
| 449 | |
| 450 | // ResetOffset resets to the provided offset, alongside a metadata string that |
| 451 | // represents the state of the partition consumer at that point in time. Reset |
| 452 | // acts as a counterpart to MarkOffset, the difference being that it allows to |
| 453 | // reset an offset to an earlier or smaller value, where MarkOffset only |
| 454 | // allows incrementing the offset. cf MarkOffset for more details. |
| 455 | ResetOffset(topic string, partition int32, offset int64, metadata string) |
| 456 | |
| 457 | // MarkMessage marks a message as consumed. |
| 458 | MarkMessage(msg *ConsumerMessage, metadata string) |
| 459 | |
| 460 | // Context returns the session context. |
| 461 | Context() context.Context |
| 462 | } |
| 463 | |
| 464 | type consumerGroupSession struct { |
| 465 | parent *consumerGroup |
| 466 | memberID string |
| 467 | generationID int32 |
| 468 | handler ConsumerGroupHandler |
| 469 | |
| 470 | claims map[string][]int32 |
| 471 | offsets *offsetManager |
| 472 | ctx context.Context |
| 473 | cancel func() |
| 474 | |
| 475 | waitGroup sync.WaitGroup |
| 476 | releaseOnce sync.Once |
| 477 | hbDying, hbDead chan none |
| 478 | } |
| 479 | |
| 480 | func newConsumerGroupSession(ctx context.Context, parent *consumerGroup, claims map[string][]int32, memberID string, generationID int32, handler ConsumerGroupHandler) (*consumerGroupSession, error) { |
| 481 | // init offset manager |
| 482 | offsets, err := newOffsetManagerFromClient(parent.groupID, memberID, generationID, parent.client) |
| 483 | if err != nil { |
| 484 | return nil, err |
| 485 | } |
| 486 | |
| 487 | // init context |
| 488 | ctx, cancel := context.WithCancel(ctx) |
| 489 | |
| 490 | // init session |
| 491 | sess := &consumerGroupSession{ |
| 492 | parent: parent, |
| 493 | memberID: memberID, |
| 494 | generationID: generationID, |
| 495 | handler: handler, |
| 496 | offsets: offsets, |
| 497 | claims: claims, |
| 498 | ctx: ctx, |
| 499 | cancel: cancel, |
| 500 | hbDying: make(chan none), |
| 501 | hbDead: make(chan none), |
| 502 | } |
| 503 | |
| 504 | // start heartbeat loop |
| 505 | go sess.heartbeatLoop() |
| 506 | |
| 507 | // create a POM for each claim |
| 508 | for topic, partitions := range claims { |
| 509 | for _, partition := range partitions { |
| 510 | pom, err := offsets.ManagePartition(topic, partition) |
| 511 | if err != nil { |
| 512 | _ = sess.release(false) |
| 513 | return nil, err |
| 514 | } |
| 515 | |
| 516 | // handle POM errors |
| 517 | go func(topic string, partition int32) { |
| 518 | for err := range pom.Errors() { |
| 519 | sess.parent.handleError(err, topic, partition) |
| 520 | } |
| 521 | }(topic, partition) |
| 522 | } |
| 523 | } |
| 524 | |
| 525 | // perform setup |
| 526 | if err := handler.Setup(sess); err != nil { |
| 527 | _ = sess.release(true) |
| 528 | return nil, err |
| 529 | } |
| 530 | |
| 531 | // start consuming |
| 532 | for topic, partitions := range claims { |
| 533 | for _, partition := range partitions { |
| 534 | sess.waitGroup.Add(1) |
| 535 | |
| 536 | go func(topic string, partition int32) { |
| 537 | defer sess.waitGroup.Done() |
| 538 | |
| 539 | // cancel the as session as soon as the first |
| 540 | // goroutine exits |
| 541 | defer sess.cancel() |
| 542 | |
| 543 | // consume a single topic/partition, blocking |
| 544 | sess.consume(topic, partition) |
| 545 | }(topic, partition) |
| 546 | } |
| 547 | } |
| 548 | return sess, nil |
| 549 | } |
| 550 | |
| 551 | func (s *consumerGroupSession) Claims() map[string][]int32 { return s.claims } |
| 552 | func (s *consumerGroupSession) MemberID() string { return s.memberID } |
| 553 | func (s *consumerGroupSession) GenerationID() int32 { return s.generationID } |
| 554 | |
| 555 | func (s *consumerGroupSession) MarkOffset(topic string, partition int32, offset int64, metadata string) { |
| 556 | if pom := s.offsets.findPOM(topic, partition); pom != nil { |
| 557 | pom.MarkOffset(offset, metadata) |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | func (s *consumerGroupSession) ResetOffset(topic string, partition int32, offset int64, metadata string) { |
| 562 | if pom := s.offsets.findPOM(topic, partition); pom != nil { |
| 563 | pom.ResetOffset(offset, metadata) |
| 564 | } |
| 565 | } |
| 566 | |
| 567 | func (s *consumerGroupSession) MarkMessage(msg *ConsumerMessage, metadata string) { |
| 568 | s.MarkOffset(msg.Topic, msg.Partition, msg.Offset+1, metadata) |
| 569 | } |
| 570 | |
| 571 | func (s *consumerGroupSession) Context() context.Context { |
| 572 | return s.ctx |
| 573 | } |
| 574 | |
| 575 | func (s *consumerGroupSession) consume(topic string, partition int32) { |
| 576 | // quick exit if rebalance is due |
| 577 | select { |
| 578 | case <-s.ctx.Done(): |
| 579 | return |
| 580 | case <-s.parent.closed: |
| 581 | return |
| 582 | default: |
| 583 | } |
| 584 | |
| 585 | // get next offset |
| 586 | offset := s.parent.config.Consumer.Offsets.Initial |
| 587 | if pom := s.offsets.findPOM(topic, partition); pom != nil { |
| 588 | offset, _ = pom.NextOffset() |
| 589 | } |
| 590 | |
| 591 | // create new claim |
| 592 | claim, err := newConsumerGroupClaim(s, topic, partition, offset) |
| 593 | if err != nil { |
| 594 | s.parent.handleError(err, topic, partition) |
| 595 | return |
| 596 | } |
| 597 | |
| 598 | // handle errors |
| 599 | go func() { |
| 600 | for err := range claim.Errors() { |
| 601 | s.parent.handleError(err, topic, partition) |
| 602 | } |
| 603 | }() |
| 604 | |
| 605 | // trigger close when session is done |
| 606 | go func() { |
| 607 | select { |
| 608 | case <-s.ctx.Done(): |
| 609 | case <-s.parent.closed: |
| 610 | } |
| 611 | claim.AsyncClose() |
| 612 | }() |
| 613 | |
| 614 | // start processing |
| 615 | if err := s.handler.ConsumeClaim(s, claim); err != nil { |
| 616 | s.parent.handleError(err, topic, partition) |
| 617 | } |
| 618 | |
| 619 | // ensure consumer is closed & drained |
| 620 | claim.AsyncClose() |
| 621 | for _, err := range claim.waitClosed() { |
| 622 | s.parent.handleError(err, topic, partition) |
| 623 | } |
| 624 | } |
| 625 | |
| 626 | func (s *consumerGroupSession) release(withCleanup bool) (err error) { |
| 627 | // signal release, stop heartbeat |
| 628 | s.cancel() |
| 629 | |
| 630 | // wait for consumers to exit |
| 631 | s.waitGroup.Wait() |
| 632 | |
| 633 | // perform release |
| 634 | s.releaseOnce.Do(func() { |
| 635 | if withCleanup { |
| 636 | if e := s.handler.Cleanup(s); e != nil { |
Scott Baker | f579f13 | 2019-10-24 14:31:41 -0700 | [diff] [blame] | 637 | s.parent.handleError(e, "", -1) |
Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame] | 638 | err = e |
| 639 | } |
| 640 | } |
| 641 | |
| 642 | if e := s.offsets.Close(); e != nil { |
| 643 | err = e |
| 644 | } |
| 645 | |
| 646 | close(s.hbDying) |
| 647 | <-s.hbDead |
| 648 | }) |
| 649 | |
| 650 | return |
| 651 | } |
| 652 | |
| 653 | func (s *consumerGroupSession) heartbeatLoop() { |
| 654 | defer close(s.hbDead) |
| 655 | defer s.cancel() // trigger the end of the session on exit |
| 656 | |
| 657 | pause := time.NewTicker(s.parent.config.Consumer.Group.Heartbeat.Interval) |
| 658 | defer pause.Stop() |
| 659 | |
| 660 | retries := s.parent.config.Metadata.Retry.Max |
| 661 | for { |
| 662 | coordinator, err := s.parent.client.Coordinator(s.parent.groupID) |
| 663 | if err != nil { |
| 664 | if retries <= 0 { |
| 665 | s.parent.handleError(err, "", -1) |
| 666 | return |
| 667 | } |
| 668 | |
| 669 | select { |
| 670 | case <-s.hbDying: |
| 671 | return |
| 672 | case <-time.After(s.parent.config.Metadata.Retry.Backoff): |
| 673 | retries-- |
| 674 | } |
| 675 | continue |
| 676 | } |
| 677 | |
| 678 | resp, err := s.parent.heartbeatRequest(coordinator, s.memberID, s.generationID) |
| 679 | if err != nil { |
| 680 | _ = coordinator.Close() |
| 681 | |
| 682 | if retries <= 0 { |
| 683 | s.parent.handleError(err, "", -1) |
| 684 | return |
| 685 | } |
| 686 | |
| 687 | retries-- |
| 688 | continue |
| 689 | } |
| 690 | |
| 691 | switch resp.Err { |
| 692 | case ErrNoError: |
| 693 | retries = s.parent.config.Metadata.Retry.Max |
| 694 | case ErrRebalanceInProgress, ErrUnknownMemberId, ErrIllegalGeneration: |
| 695 | return |
| 696 | default: |
| 697 | s.parent.handleError(err, "", -1) |
| 698 | return |
| 699 | } |
| 700 | |
| 701 | select { |
| 702 | case <-pause.C: |
| 703 | case <-s.hbDying: |
| 704 | return |
| 705 | } |
| 706 | } |
| 707 | } |
| 708 | |
| 709 | // -------------------------------------------------------------------- |
| 710 | |
| 711 | // ConsumerGroupHandler instances are used to handle individual topic/partition claims. |
| 712 | // It also provides hooks for your consumer group session life-cycle and allow you to |
| 713 | // trigger logic before or after the consume loop(s). |
| 714 | // |
| 715 | // PLEASE NOTE that handlers are likely be called from several goroutines concurrently, |
| 716 | // ensure that all state is safely protected against race conditions. |
| 717 | type ConsumerGroupHandler interface { |
| 718 | // Setup is run at the beginning of a new session, before ConsumeClaim. |
| 719 | Setup(ConsumerGroupSession) error |
| 720 | |
| 721 | // Cleanup is run at the end of a session, once all ConsumeClaim goroutines have exited |
| 722 | // but before the offsets are committed for the very last time. |
| 723 | Cleanup(ConsumerGroupSession) error |
| 724 | |
| 725 | // ConsumeClaim must start a consumer loop of ConsumerGroupClaim's Messages(). |
| 726 | // Once the Messages() channel is closed, the Handler must finish its processing |
| 727 | // loop and exit. |
| 728 | ConsumeClaim(ConsumerGroupSession, ConsumerGroupClaim) error |
| 729 | } |
| 730 | |
| 731 | // ConsumerGroupClaim processes Kafka messages from a given topic and partition within a consumer group. |
| 732 | type ConsumerGroupClaim interface { |
| 733 | // Topic returns the consumed topic name. |
| 734 | Topic() string |
| 735 | |
| 736 | // Partition returns the consumed partition. |
| 737 | Partition() int32 |
| 738 | |
| 739 | // InitialOffset returns the initial offset that was used as a starting point for this claim. |
| 740 | InitialOffset() int64 |
| 741 | |
| 742 | // HighWaterMarkOffset returns the high water mark offset of the partition, |
| 743 | // i.e. the offset that will be used for the next message that will be produced. |
| 744 | // You can use this to determine how far behind the processing is. |
| 745 | HighWaterMarkOffset() int64 |
| 746 | |
| 747 | // Messages returns the read channel for the messages that are returned by |
| 748 | // the broker. The messages channel will be closed when a new rebalance cycle |
| 749 | // is due. You must finish processing and mark offsets within |
| 750 | // Config.Consumer.Group.Session.Timeout before the topic/partition is eventually |
| 751 | // re-assigned to another group member. |
| 752 | Messages() <-chan *ConsumerMessage |
| 753 | } |
| 754 | |
| 755 | type consumerGroupClaim struct { |
| 756 | topic string |
| 757 | partition int32 |
| 758 | offset int64 |
| 759 | PartitionConsumer |
| 760 | } |
| 761 | |
| 762 | func newConsumerGroupClaim(sess *consumerGroupSession, topic string, partition int32, offset int64) (*consumerGroupClaim, error) { |
| 763 | pcm, err := sess.parent.consumer.ConsumePartition(topic, partition, offset) |
| 764 | if err == ErrOffsetOutOfRange { |
| 765 | offset = sess.parent.config.Consumer.Offsets.Initial |
| 766 | pcm, err = sess.parent.consumer.ConsumePartition(topic, partition, offset) |
| 767 | } |
| 768 | if err != nil { |
| 769 | return nil, err |
| 770 | } |
| 771 | |
| 772 | go func() { |
| 773 | for err := range pcm.Errors() { |
| 774 | sess.parent.handleError(err, topic, partition) |
| 775 | } |
| 776 | }() |
| 777 | |
| 778 | return &consumerGroupClaim{ |
| 779 | topic: topic, |
| 780 | partition: partition, |
| 781 | offset: offset, |
| 782 | PartitionConsumer: pcm, |
| 783 | }, nil |
| 784 | } |
| 785 | |
| 786 | func (c *consumerGroupClaim) Topic() string { return c.topic } |
| 787 | func (c *consumerGroupClaim) Partition() int32 { return c.partition } |
| 788 | func (c *consumerGroupClaim) InitialOffset() int64 { return c.offset } |
| 789 | |
| 790 | // Drains messages and errors, ensures the claim is fully closed. |
| 791 | func (c *consumerGroupClaim) waitClosed() (errs ConsumerErrors) { |
| 792 | go func() { |
| 793 | for range c.Messages() { |
| 794 | } |
| 795 | }() |
| 796 | |
| 797 | for err := range c.Errors() { |
| 798 | errs = append(errs, err) |
| 799 | } |
| 800 | return |
| 801 | } |