apache-kafka - Spring Cloud Stream Kafka Stream 应用程序显示在每次重新启动时将分区 event-x 的偏移量重置为偏移量 0

Question

我有一个 Spring Cloud Stream Kafka Stream 应用程序，它从一个主题（事件）中读取并执行一个简单的处理：

@Configuration
class EventKStreamConfiguration {

    private val logger = LoggerFactory.getLogger(javaClass)

    @StreamListener
    fun process(@Input("event") eventStream: KStream<String, EventReceived>) {

        eventStream.foreach { key, value ->
            logger.info("--------> Processing Event {}", value)
            // Save in DB
        }
    }
}

此应用程序使用来自 Confluent Cloud 的 Kafka 环境，其事件主题具有 6 个分区。完整的配置是：

spring:
  application:
    name: events-processor
  cloud:
    stream:
      schema-registry-client:
        endpoint: ${schema-registry-url:http://localhost:8081}
      kafka:
        streams:
          binder:
            brokers: ${kafka-brokers:localhost}
            configuration:
              application:
                id: ${spring.application.name}
              default:
                key:
                  serde: org.apache.kafka.common.serialization.Serdes$StringSerde
              schema:
                registry:
                  url: ${spring.cloud.stream.schema-registry-client.endpoint}
              value:
                subject:
                  name:
                    strategy: io.confluent.kafka.serializers.subject.RecordNameStrategy
              processing:
                guarantee: exactly_once
          bindings:
            event:
              consumer:
                valueSerde: io.confluent.kafka.streams.serdes.avro.SpecificAvroSerde
      bindings:
        event:
          destination: event

  data:
    mongodb:
      uri: ${mongodb-uri:mongodb://localhost/test}

server:
  port: 8085

logging:
  level:
    org.springframework.kafka.config: debug

---

spring:
  profiles: confluent-cloud
  cloud:
    stream:
      kafka:
        streams:
          binder:
            autoCreateTopics: false
            configuration:
              retry:
                backoff:
                  ms: 500
              security:
                protocol: SASL_SSL
              sasl:
                mechanism: PLAIN
                jaas:
                  config: xxx
              basic:
                auth:
                  credentials:
                    source: USER_INFO
              schema:
                registry:
                  basic:
                    auth:
                      user:
                        info: yyy

KStream 正在正确处理消息。如果我重新启动应用程序，它们不会被重新处理。注意：我不希望它们被重新处理，所以这种行为是可以的。

但是启动日志显示了一些奇怪的位：

首先，它显示了恢复消费者客户端的创建。自动偏移重置无：

2019-07-19 10:20:17.120  INFO 82473 --- [           main] o.a.k.s.p.internals.StreamThread         : stream-thread [events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1] Creating restore consumer client
2019-07-19 10:20:17.123  INFO 82473 --- [           main] o.a.k.clients.consumer.ConsumerConfig    : ConsumerConfig values: 
    auto.commit.interval.ms = 5000
    auto.offset.reset = none

然后它最早创建一个具有自动偏移重置的消费者客户端。

2019-07-19 10:20:17.235  INFO 82473 --- [           main] o.a.k.s.p.internals.StreamThread         : stream-thread [events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1] Creating consumer client
2019-07-19 10:20:17.241  INFO 82473 --- [           main] o.a.k.clients.consumer.ConsumerConfig    : ConsumerConfig values: 
    auto.commit.interval.ms = 5000
    auto.offset.reset = earliest

启动日志的最终跟踪显示偏移量重置为 0。每次重新启动应用程序时都会发生这种情况：

2019-07-19 10:20:31.577  INFO 82473 --- [-StreamThread-1] o.a.k.s.p.internals.StreamThread         : stream-thread [events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1] State transition from PARTITIONS_ASSIGNED to RUNNING
2019-07-19 10:20:31.578  INFO 82473 --- [-StreamThread-1] org.apache.kafka.streams.KafkaStreams    : stream-client [events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f] State transition from REBALANCING to RUNNING
2019-07-19 10:20:31.669  INFO 82473 --- [events-processor] o.a.k.c.consumer.internals.Fetcher       : [Consumer clientId=events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1-consumer, groupId=events-processor] Resetting offset for partition event-3 to offset 0.
2019-07-19 10:20:31.669  INFO 82473 --- [events-processor] o.a.k.c.consumer.internals.Fetcher       : [Consumer clientId=events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1-consumer, groupId=events-processor] Resetting offset for partition event-0 to offset 0.
2019-07-19 10:20:31.669  INFO 82473 --- [events-processor] o.a.k.c.consumer.internals.Fetcher       : [Consumer clientId=events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1-consumer, groupId=events-processor] Resetting offset for partition event-1 to offset 0.
2019-07-19 10:20:31.669  INFO 82473 --- [events-processor] o.a.k.c.consumer.internals.Fetcher       : [Consumer clientId=events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1-consumer, groupId=events-processor] Resetting offset for partition event-5 to offset 0.
2019-07-19 10:20:31.670  INFO 82473 --- [events-processor] o.a.k.c.consumer.internals.Fetcher       : [Consumer clientId=events-processor-9a8069c4-3fb6-4d76-a207-efbbadd52b8f-StreamThread-1-consumer, groupId=events-processor] Resetting offset for partition event-4 to offset 0.

配置了两个消费者的原因是什么？
auto.offset.reset = earliest当我没有明确配置它并且Kafka默认值是最新的时，为什么第二个有？
我想要默认的（auto.offset.reset = latest）行为，它似乎工作正常。但是，这与我在日志中看到的内容不矛盾吗？

更新：

我会这样改写第三个问题：为什么日志显示分区在每次重新启动时都被重置为 0，尽管如此，没有消息被重新传递到 KStream？

更新 2：

我已经简化了这个场景，这次使用的是原生 Kafka Streams 应用程序。该行为与使用 Spring Cloud Stream 观察到的行为完全相同。但是，检查消费者组和我发现的分区是有道理的。

流媒体：

fun main() {

    val props = Properties()
    props[StreamsConfig.APPLICATION_ID_CONFIG] = "streams-wordcount"
    props[StreamsConfig.BOOTSTRAP_SERVERS_CONFIG] = "localhost:9092"
    props[StreamsConfig.CACHE_MAX_BYTES_BUFFERING_CONFIG] = 0
    props[StreamsConfig.DEFAULT_KEY_SERDE_CLASS_CONFIG] = Serdes.String().javaClass.name
    props[StreamsConfig.DEFAULT_VALUE_SERDE_CLASS_CONFIG] = Serdes.String().javaClass.name

    val builder = StreamsBuilder()

    val source = builder.stream<String, String>("streams-plaintext-input")

    source.foreach { key, value -> println("$key $value") }

    val streams = KafkaStreams(builder.build(), props)
    val latch = CountDownLatch(1)

    // attach shutdown handler to catch control-c
    Runtime.getRuntime().addShutdownHook(object : Thread("streams-wordcount-shutdown-hook") {
        override fun run() {
            streams.close()
            latch.countDown()
        }
    })

    try {
        streams.start()
        latch.await()
    } catch (e: Throwable) {
        exitProcess(1)
    }

    exitProcess(0)
}

这是我所看到的：

1）对于一个空主题，启动显示所有分区重置为偏移量0：

07:55:03.885 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-2 to offset 0.
07:55:03.886 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-3 to offset 0.
07:55:03.886 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-0 to offset 0.
07:55:03.886 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-1 to offset 0.
07:55:03.886 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-4 to offset 0.
07:55:03.886 [streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-3549a54e-49db-4490-bd9f-7156e972021a-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-5 to offset 0

2）我在主题中放入一条消息并检查消费者组，看到记录在分区 4 中：

TOPIC                   PARTITION  CURRENT-OFFSET  LOG-END-OFFSET  LAG             CONSUMER-ID                                                                                                         HOST            CLIENT-ID
streams-plaintext-input 0          -               0               -               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer
streams-plaintext-input 5          -               0               -               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer
streams-plaintext-input 1          -               0               -               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer
streams-plaintext-input 2          -               0               -               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer
streams-plaintext-input 3          -               0               -               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer
streams-plaintext-input 4          1               1               0               streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer-905a307a-4c49-4d8b-ac2e-5525ba2e8a8e /127.0.0.1      streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer

3）我重新启动应用程序。现在重置只影响空分区（0、1、2、3、5）：

07:57:39.477 [streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-2 to offset 0.
07:57:39.478 [streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-3 to offset 0.
07:57:39.478 [streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-0 to offset 0.
07:57:39.479 [streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-1 to offset 0.
07:57:39.479 [streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-b1565eca-7d80-4550-97d2-e78ead62a840-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-5 to offset 0.

4）我插入另一条消息，检查消费者组状态，同样的事情发生了：记录在分区 2 中，当重新启动应用程序时，它只重置空分区（0、1、3、5）：

TOPIC                   PARTITION  CURRENT-OFFSET  LOG-END-OFFSET  LAG             CONSUMER-ID                                                                                                         HOST            CLIENT-ID
streams-plaintext-input 0          -               0               -               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer
streams-plaintext-input 5          -               0               -               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer
streams-plaintext-input 1          -               0               -               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer
streams-plaintext-input 2          1               1               0               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer
streams-plaintext-input 3          -               0               -               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer
streams-plaintext-input 4          1               1               0               streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer-cb04e2bd-598f-455f-b913-1370b4144dd6 /127.0.0.1      streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer

08:00:42.313 [streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-3 to offset 0.
08:00:42.314 [streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-0 to offset 0.
08:00:42.314 [streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-1 to offset 0.
08:00:42.314 [streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1] INFO org.apache.kafka.clients.consumer.internals.Fetcher - [Consumer clientId=streams-wordcount-addb08ed-62ce-47f9-a446-f2ee0592c53d-StreamThread-1-consumer, groupId=streams-wordcount] Resetting offset for partition streams-plaintext-input-5 to offset 0.

score 2 · Accepted Answer

配置了两个消费者的原因是什么？

Restore Consumer Client是容错和状态管理的专用消费者。它负责从变更日志主题中恢复状态。它与应用程序消费者客户端分开显示。您可以在此处找到更多信息： https ://docs.confluent.io/current/streams/monitoring.html#kafka-restore-consumer-client-id

为什么当我没有明确配置第二个并且 Kafka 默认是最新的时，第二个有 auto.offset.reset = early？

你是对的，auto.offset.reset默认值latest在 Kafka Consumer 中。但在 Spring Cloud Stream 中，消费者的默认startOffset值为earliest. 因此它显示earliest在第二个消费者中。它也取决于spring.cloud.stream.bindings.<channelName>.group绑定。如果明确设置，则startOffset设置为earliest，否则设置latest为anonymous消费者。

参考：Spring Cloud Stream Kafka 消费者属性

我想要默认的（auto.offset.reset = latest）行为，它似乎工作正常。但是，这与我在日志中看到的内容不矛盾吗？

在消费者组的情况下anonymous，默认值为.startOffsetlatest

apache-kafka - Spring Cloud Stream Kafka Stream 应用程序显示在每次重新启动时将分区 event-x 的偏移量重置为偏移量 0

1 回答 1

Related

Reference