Apache spark 卡夫卡的火花流总是从一开始就开始

查看我对解决方案的公认答案的最后一条评论

我像这样配置了一个

DStream

：

  val kafkaParams = Map[String, Object](
    "bootstrap.servers" -> "kafka1.example.com:9092",
    "key.deserializer" -> classOf[StringDeserializer],
    "value.deserializer" -> classOf[KafkaAvroDeserializer],
    "group.id" -> "mygroup",
    "specific.avro.reader" -> true,
    "schema.registry.url" -> "http://schema.example.com:8081"
  )

  val stream = KafkaUtils.createDirectStream(
    ssc,
    PreferConsistent,
    Subscribe[String, DataFile](topics, kafkaParams)
  )

虽然这样做很有效，我得到了预期的

数据文件

s，但当我停止并重新运行作业时，它总是从主题的开头开始。我如何才能实现它在上次爆炸的地方继续

后续行动1 正如Bhima Rao Goginini的回答，我改变了我的配置，如下所示：

val consumerParams =
  Map("bootstrap.servers" -> bootstrapServerString,
      "schema.registry.url" -> schemaRegistryUri.toString,
      "specific.avro.reader" -> "true",
      "group.id" -> "measuring-data-files",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[KafkaAvroDeserializer],
      "enable.auto.commit" -> (false: JavaBool),
      "auto.offset.reset" -> "earliest")

    // Commit the offsets
    log.info("Committing the offsets")
    val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges
    if(offsetRanges.isEmpty) {
      log.info("Offset ranges is empty...")
    } else {
      log.info("# offset ranges: %d" format offsetRanges.length)
    }
    object cb extends OffsetCommitCallback {

      def onComplete(offsets: util.Map[TopicPartition, OffsetAndMetadata],
                     exception: Exception): Unit =
        if(exception != null) {
          log.info("Commit FAILED")
          log.error(exception.getMessage, exception)
        } else {
          log.info("Commit SUCCEEDED - count: %d" format offsets.size())
          offsets.
            asScala.
            foreach {
              case (p, omd) =>
                log.info("partition = %d; topic = %s; offset = %d; metadata = %s".
                  format(p.partition(), p.topic(), omd.offset(), omd.metadata()))
            }
        }
    }
    stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges, cb)

我设置了一条流：

val stream = KafkaUtils.
  createDirectStream(ssc,
                     LocationStrategies.PreferConsistent,
                     ConsumerStrategies.Subscribe[String, DataFile](List(inTopic), consumerParams))

然后我处理它：

stream.
  foreachRDD { rdd =>
    ... // Do stuff with the RDD - transform, produce to other topic etc.
    // Commit the offsets
    log.info("Committing the offsets")
    val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges
    stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges)        
  }

但在重新运行时，它仍然总是从头开始

以下是我的卡夫卡日志摘录：

跑步：

[2018-07-04 07:47:31,593] INFO [GroupCoordinator 0]: Preparing to rebalance group measuring-data-files with old generation 22 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:47:31,594] INFO [GroupCoordinator 0]: Stabilized group measuring-data-files generation 23 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:47:31,599] INFO [GroupCoordinator 0]: Assignment received from leader for group measuring-data-files for generation 23 (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:48:06,690] INFO [ProducerStateManager partition=data-0] Writing producer snapshot at offset 131488999 (kafka.log.ProducerStateManager)
[2018-07-04 07:48:06,690] INFO [Log partition=data-0, dir=E:\confluent-4.1.1\data\kafka] Rolled new log segment at offset 131488999 in 1 ms. (kafka.log.Log)
[2018-07-04 07:48:10,788] INFO [GroupMetadataManager brokerId=0] Removed 0 expired offsets in 0 milliseconds. (kafka.coordinator.group.GroupMetadataManager)
[2018-07-04 07:48:30,074] INFO [GroupCoordinator 0]: Member consumer-1-262ece09-93c4-483e-b488-87057578dabc in group measuring-data-files has failed, removing it from the group (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:48:30,074] INFO [GroupCoordinator 0]: Preparing to rebalance group measuring-data-files with old generation 23 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:48:30,074] INFO [GroupCoordinator 0]: Group measuring-data-files with generation 24 is now empty (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:48:45,761] INFO [ProducerStateManager partition=data-0] Writing producer snapshot at offset 153680971 (kafka.log.ProducerStateManager)
[2018-07-04 07:48:45,763] INFO [Log partition=data-0, dir=E:\confluent-4.1.1\data\kafka] Rolled new log segment at offset 153680971 in 3 ms. (kafka.log.Log)
[2018-07-04 07:49:24,819] INFO [ProducerStateManager partition=data-0] Writing producer snapshot at offset 175872864 (kafka.log.ProducerStateManager)
[2018-07-04 07:49:24,820] INFO [Log partition=data-0, dir=E:\confluent-4.1.1\data\kafka] Rolled new log segment at offset 175872864 in 1 ms. (kafka.log.Log)

下次运行：

[2018-07-04 07:50:13,748] INFO [GroupCoordinator 0]: Preparing to rebalance group measuring-data-files with old generation 24 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:50:13,749] INFO [GroupCoordinator 0]: Stabilized group measuring-data-files generation 25 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:50:13,754] INFO [GroupCoordinator 0]: Assignment received from leader for group measuring-data-files for generation 25 (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:50:43,758] INFO [GroupCoordinator 0]: Member consumer-1-906c2eaa-f012-4283-96fc-c34582de33fb in group measuring-data-files has failed, removing it from the group (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:50:43,758] INFO [GroupCoordinator 0]: Preparing to rebalance group measuring-data-files with old generation 25 (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)
[2018-07-04 07:50:43,758] INFO [GroupCoordinator 0]: Group measuring-data-files with generation 26 is now empty (__consumer_offsets-8) (kafka.coordinator.group.GroupCoordinator)

后续行动2 我使保存偏移量变得更加冗长，如下所示：

val consumerParams =
  Map("bootstrap.servers" -> bootstrapServerString,
      "schema.registry.url" -> schemaRegistryUri.toString,
      "specific.avro.reader" -> "true",
      "group.id" -> "measuring-data-files",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[KafkaAvroDeserializer],
      "enable.auto.commit" -> (false: JavaBool),
      "auto.offset.reset" -> "earliest")

    // Commit the offsets
    log.info("Committing the offsets")
    val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges
    if(offsetRanges.isEmpty) {
      log.info("Offset ranges is empty...")
    } else {
      log.info("# offset ranges: %d" format offsetRanges.length)
    }
    object cb extends OffsetCommitCallback {

      def onComplete(offsets: util.Map[TopicPartition, OffsetAndMetadata],
                     exception: Exception): Unit =
        if(exception != null) {
          log.info("Commit FAILED")
          log.error(exception.getMessage, exception)
        } else {
          log.info("Commit SUCCEEDED - count: %d" format offsets.size())
          offsets.
            asScala.
            foreach {
              case (p, omd) =>
                log.info("partition = %d; topic = %s; offset = %d; metadata = %s".
                  format(p.partition(), p.topic(), omd.offset(), omd.metadata()))
            }
        }
    }
    stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges, cb)

我得到一个例外：

2018-07-04 10:14:00 ERROR DumpTask$:136 - Commit cannot be completed since the group has already rebalanced and assigned the partitions to another member. This means that the time between subsequent calls to poll() was longer than the configured session.timeout.ms, which typically implies that the poll loop is spending too much time message processing. You can address this either by increasing the session timeout or by reducing the maximum size of batches returned in poll() with max.poll.records.
org.apache.kafka.clients.consumer.CommitFailedException: Commit cannot be completed since the group has already rebalanced and assigned the partitions to another member. This means that the time between subsequent calls to poll() was longer than the configured session.timeout.ms, which typically implies that the poll loop is spending too much time message processing. You can address this either by increasing the session timeout or by reducing the maximum size of batches returned in poll() with max.poll.records.
        at org.apache.kafka.clients.consumer.internals.ConsumerCoordinator$OffsetCommitResponseHandler.handle(ConsumerCoordinator.java:600)
        at org.apache.kafka.clients.consumer.internals.ConsumerCoordinator$OffsetCommitResponseHandler.handle(ConsumerCoordinator.java:541)
        at org.apache.kafka.clients.consumer.internals.AbstractCoordinator$CoordinatorResponseHandler.onSuccess(AbstractCoordinator.java:679)
        at org.apache.kafka.clients.consumer.internals.AbstractCoordinator$CoordinatorResponseHandler.onSuccess(AbstractCoordinator.java:658)
        at org.apache.kafka.clients.consumer.internals.RequestFuture$1.onSuccess(RequestFuture.java:167)
        at org.apache.kafka.clients.consumer.internals.RequestFuture.fireSuccess(RequestFuture.java:133)
        at org.apache.kafka.clients.consumer.internals.RequestFuture.complete(RequestFuture.java:107)
        at org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient$RequestFutureCompletionHandler.onComplete(ConsumerNetworkClient.java:426)
        at org.apache.kafka.clients.NetworkClient.poll(NetworkClient.java:278)
        at org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.clientPoll(ConsumerNetworkClient.java:360)
        at org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.poll(ConsumerNetworkClient.java:224)
        at org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.poll(ConsumerNetworkClient.java:192)
        at org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.awaitPendingRequests(ConsumerNetworkClient.java:260)
        at org.apache.kafka.clients.consumer.internals.AbstractCoordinator.ensureActiveGroup(AbstractCoordinator.java:222)
        at org.apache.kafka.clients.consumer.internals.ConsumerCoordinator.ensurePartitionAssignment(ConsumerCoordinator.java:366)
        at org.apache.kafka.clients.consumer.KafkaConsumer.pollOnce(KafkaConsumer.java:978)
        at org.apache.kafka.clients.consumer.KafkaConsumer.poll(KafkaConsumer.java:938)
        at org.apache.spark.streaming.kafka010.DirectKafkaInputDStream.paranoidPoll(DirectKafkaInputDStream.scala:163)
        at org.apache.spark.streaming.kafka010.DirectKafkaInputDStream.latestOffsets(DirectKafkaInputDStream.scala:182)
        at org.apache.spark.streaming.kafka010.DirectKafkaInputDStream.compute(DirectKafkaInputDStream.scala:209)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1$$anonfun$1$$anonfun$apply$7.apply(DStream.scala:342)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1$$anonfun$1$$anonfun$apply$7.apply(DStream.scala:342)
        at scala.util.DynamicVariable.withValue(DynamicVariable.scala:58)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1$$anonfun$1.apply(DStream.scala:341)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1$$anonfun$1.apply(DStream.scala:341)
        at org.apache.spark.streaming.dstream.DStream.createRDDWithLocalProperties(DStream.scala:416)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1.apply(DStream.scala:336)
        at org.apache.spark.streaming.dstream.DStream$$anonfun$getOrCompute$1.apply(DStream.scala:334)
        at scala.Option.orElse(Option.scala:289)
        at org.apache.spark.streaming.dstream.DStream.getOrCompute(DStream.scala:331)
        at org.apache.spark.streaming.dstream.ForEachDStream.generateJob(ForEachDStream.scala:48)
        at org.apache.spark.streaming.DStreamGraph$$anonfun$1.apply(DStreamGraph.scala:122)
        at org.apache.spark.streaming.DStreamGraph$$anonfun$1.apply(DStreamGraph.scala:121)
        at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:241)
        at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:241)
        at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
        at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
        at scala.collection.TraversableLike$class.flatMap(TraversableLike.scala:241)
        at scala.collection.AbstractTraversable.flatMap(Traversable.scala:104)
        at org.apache.spark.streaming.DStreamGraph.generateJobs(DStreamGraph.scala:121)
        at org.apache.spark.streaming.scheduler.JobGenerator$$anonfun$3.apply(JobGenerator.scala:249)
        at org.apache.spark.streaming.scheduler.JobGenerator$$anonfun$3.apply(JobGenerator.scala:247)
        at scala.util.Try$.apply(Try.scala:192)
        at org.apache.spark.streaming.scheduler.JobGenerator.generateJobs(JobGenerator.scala:247)
        at org.apache.spark.streaming.scheduler.JobGenerator.org$apache$spark$streaming$scheduler$JobGenerator$$processEvent(JobGenerator.scala:183)
        at org.apache.spark.streaming.scheduler.JobGenerator$$anon$1.onReceive(JobGenerator.scala:89)
        at org.apache.spark.streaming.scheduler.JobGenerator$$anon$1.onReceive(JobGenerator.scala:88)
        at org.apache.spark.util.EventLoop$$anon$1.run(EventLoop.scala:48)

---

我应该如何解决这个问题？

Spark提供了两个API来读取来自卡夫卡的消息

来自Spark文档

方法1：基于接收机的方法 这种方法使用接收器来接收数据。接收器是使用卡夫卡实现的 高级消费API。与所有接收器一样，从 卡夫卡通过一个接收器被存储在火花执行器中，然后是作业 由Spark Streaming启动处理数据。

进近2：直接进近（无接收机） Spark中引入了这种新的无接收机“直接”方法 1.3确保更强大的端到端保障。这种方法不使用接收器来接收数据，而是定期查询卡夫卡 每个主题+分区中的最新偏移量，以及相应的 定义每个批次中要处理的偏移范围。工作什么时候开始 处理数据时，Kafka的简单消费者API用于 从Kafka读取定义的偏移范围（类似于读取文件 从文件系统）。
请注意，这种方法的一个缺点是 它不会更新Zookeeper中的偏移量，因此基于Zookeeper 卡夫卡监控工具不会显示进度。但是，您可以访问 此方法在每个批处理和更新中处理的偏移量 动物园管理员自己

在您的情况下，您使用的是直接方法，因此您需要自己处理邮件偏移量，并指定要读取邮件的偏移量范围。或者，如果您希望zookeeper处理您的消息偏移量，那么您可以使用基于接收器的方法，使用

KafkaUtils.createStream（）

API

---

您可以在中找到有关如何处理卡夫卡偏移的更多信息。

Spark提供了两个API来读取卡夫卡的消息

来自Spark文档

方法1：基于接收机的方法 这种方法使用接收器来接收数据。接收器是使用卡夫卡实现的 高级消费API。与所有接收器一样，从 卡夫卡通过一个接收器被存储在火花执行器中，然后是作业 由Spark Streaming启动处理数据。

进近2：直接进近（无接收机） Spark中引入了这种新的无接收机“直接”方法 1.3确保更强大的端到端保障。这种方法不使用接收器来接收数据，而是定期查询卡夫卡 每个主题+分区中的最新偏移量，以及相应的 定义每个批次中要处理的偏移范围。工作什么时候开始 处理数据时，Kafka的简单消费者API用于 从Kafka读取定义的偏移范围（类似于读取文件 从文件系统）。
请注意，这种方法的一个缺点是 它不会更新Zookeeper中的偏移量，因此基于Zookeeper 卡夫卡监控工具不会显示进度。但是，您可以访问 此方法在每个批处理和更新中处理的偏移量 动物园管理员自己

在您的情况下，您使用的是直接方法，因此您需要自己处理邮件偏移量，并指定要读取邮件的偏移量范围。或者，如果您希望zookeeper处理您的消息偏移量，那么您可以使用基于接收器的方法，使用

KafkaUtils.createStream（）

API

---

您可以在中找到有关如何处理kafka偏移量的更多信息。

使用新的Spark kafka Connect API，我们可以尝试异步提交

读取偏移量并在流程完成后提交

卡夫卡的配置相同：

enable.auto.commit=false

auto.offset.reset=earlime

或

auto.offset.reset=latest

-->如果卡夫卡主题中没有最后提交的偏移量可用，则此配置将根据此配置从开始或结束读取偏移量

 stream.foreachRDD { rdd =>
   val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges

   // some time later, after outputs have completed
   stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges)
}

---

来源如下：

使用新的Spark Kafka Connect API，我们可以尝试异步提交

读取偏移量并在流程完成后提交

卡夫卡的配置相同：

enable.auto.commit=false

auto.offset.reset=earlime

或

auto.offset.reset=latest

-->如果卡夫卡主题中没有最后提交的偏移量可用，则此配置将根据此配置从开始或结束读取偏移量

 stream.foreachRDD { rdd =>
   val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges

   // some time later, after outputs have completed
   stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges)
}

---

来源如下：

这对于0.8卡夫卡集成库是正确的。我使用的是0.10版本，它会定期存储偏移量。也许我等的时间不够长。我尝试禁用自动提交和手动提交，如0.8 Kafka集成库中所述。我使用的是0.10版本，它会定期存储偏移量。也许我等的时间不够长。我尝试禁用自动提交和手动提交，如中所述。不幸的是，这并没有改变任何事情。它仍然从一开始就开始了。我编辑了我的原始问题，以包括我的调查。stream.foreachRDD{rdd=>val offsetRanges=rdd.asInstanceOf[HasOffsetRanges].offsetRanges//一段时间后，在输出完成stream.asInstanceOf[CanCommitOffsets].commitAsync（offsetRa