我们都知道,使用Kafka生产者往Kafka的经纪人发送消息的时候,Kafka会根据消息的密钥计算出这条消息应该发送到分区。最小的分区计算类是HashPartitioner,其实现如下:

class HashPartitioner(props: VerifiableProperties = null) extends Partitioner {
  def partition(data: Any, numPartitions: Int): Int = {
    (data.hashCode % numPartitions)
  }
}
其中numPartitions就是Tpoic的分区总数。partition函数会在kafka的getPartition函数中被调用,计算消息的分区ID。而numPartitions的数量是通过

val topicPartitionsList = getPartitionListForTopic(message)
val numPartitions = topicPartitionList.size
计算得到的,getPartitionListForTopic实现如下:

private def getPartitionListForTopic(m: KeyedMessage[K,Message]):
            Seq[PartitionAndLeader] = {
  val topicPartitionsList =
brokerPartitionInfo.getBrokerPartitionInfo(m.topic, correlationId.getAndIncrement)
  debug("Broker partitions registered for topic: %s are %s"
    .format(m.topic, topicPartitionsList.map(p => p.partitionId).mkString(",")))
  val totalNumPartitions = topicPartitionsList.length
  if(totalNumPartitions == 0)
    throw new NoBrokersForPartitionException("Partition key = " + m.key)
  topicPartitionsList
}

那么问题是,如果在Kafka Producer往Kafka的Broker发送消息的时候用户通过命令修改了改主题的分区数,Kafka Producer能动态感知吗?答案是可以的。那是立刻就感知到吗?不是,是过一定的时间(topic.metadata.refresh.interval.ms参数决定)才知道分区数改变的,我们来看看代码实现。

上面代码中的topicPartitionsList是通过getBrokerPartitionInfo函数获取的,其实现如下:

def getBrokerPartitionInfo(topic: String, correlationId: Int): Seq[PartitionAndLeader] = {
  debug("Getting broker partition info for topic %s".format(topic))
  // check if the cache has metadata for this topic
  val topicMetadata = topicPartitionInfo.get(topic)
  val metadata: TopicMetadata =
    topicMetadata match {
      case Some(m) => m
      case None =>
        // refresh the topic metadata cache
        updateInfo(Set(topic), correlationId)
        val topicMetadata = topicPartitionInfo.get(topic)
        topicMetadata match {
          case Some(m) => m
          case None => 
throw new KafkaException("Failed to fetch topic metadata for topic: " + topic)
        }
    }
  val partitionMetadata = metadata.partitionsMetadata
  if(partitionMetadata.size == 0) {
    if(metadata.errorCode != ErrorMapping.NoError) {
      throw
  new KafkaException(ErrorMapping.exceptionFor(metadata.errorCode))
    } else {
      throw new KafkaException("Topic metadata %s has empty 
      partition metadata and no error code".format(metadata))
    }
  }
  partitionMetadata.map { m =>
    m.leader match {
      case Some(leader) =>
        debug("Partition [%s,%d] has leader %d".format(topic, m.partitionId, leader.id))
        new PartitionAndLeader(topic, m.partitionId, Some(leader.id))
      case None =>
        debug("Partition [%s,%d] does not have a leader yet".format(topic, m.partitionId))
        new PartitionAndLeader(topic, m.partitionId, None)
    }
  }.sortWith((s, t) => s.partitionId < t.partitionId)
}

topicPartitionInfo的数据类型是HashMap[String, TopicMetadata],程序先通过话题从名topicPartitionInfo中查找是否有这个话题的元数据,如果有则直接返回;如果没有呢则调用?updateInfo(Set(topic), correlationId)函数获取该话题的元数据:

def updateInfo(topics: Set[String], correlationId: Int) {
  var topicsMetadata: Seq[TopicMetadata] = Nil
  val topicMetadataResponse =
  ClientUtils.fetchTopicMetadata(topics, brokers, producerConfig, correlationId)
  topicsMetadata = topicMetadataResponse.topicsMetadata
  // throw partition specific exception
  topicsMetadata.foreach(tmd =>{
    trace("Metadata for topic %s is %s".format(tmd.topic, tmd))
    if(tmd.errorCode == ErrorMapping.NoError) {
      topicPartitionInfo.put(tmd.topic, tmd)
    } else
      warn("Error while fetching metadata [%s] for topic [%s]: %s 
  ".format(tmd, tmd.topic, ErrorMapping.exceptionFor(tmd.errorCode).getClass))
    tmd.partitionsMetadata.foreach(pmd =>{
      if (pmd.errorCode != ErrorMapping.NoError 
  && pmd.errorCode == ErrorMapping.LeaderNotAvailableCode) {
        warn("Error while fetching metadata %s for topic partition
     [%s,%d]: [%s]".format(pmd, tmd.topic, pmd.partitionId,
          ErrorMapping.exceptionFor(pmd.errorCode).getClass))
      } // any other error code (e.g. ReplicaNotAvailable) 
  //can be ignored since the producer does not need to 
  //access the replica and isr metadata
    })
  })
  producerPool.updateProducer(topicsMetadata)
}

上面的程序通过调用ClientUtils.fetchTopicMetadata函数获取topicsMetadata。如果没有遇到错误,那么就将获取到的topicsMetadata放置到topicPartitionInfo中。那么程序是如何感知当前主题的元数据发生变化,而前面说的分区数增加了吗?其实主要逻辑在这里:

if (topicMetadataRefreshInterval >= 0 &&
          SystemTime.milliseconds - lastTopicMetadataRefreshTime > 
    topicMetadataRefreshInterval) {
        Utils.swallowError(brokerPartitionInfo.updateInfo(topicMetadataToRefresh.toSet, 
    correlationId.getAndIncrement))
 
        ......
}

这个逻辑在处理消息的时候就会被调用,如果和topicMetadataRefreshInterval>=0当前时间初始化上一次元数据更新的时间间隙大于topicMetadataRefreshInterval,则重新再一次更新Tpoic的元数据,而其topicMetadataRefreshInterval值就是通过topic.metadata.refresh.interval.ms配置的。

结论:在启动Kafka Producer往Kafka的Brok发送消息的时候,用户修改了该主题的分区数,Producer可以在最多topic.metadata.refresh.interval.ms的时间之后感知到,此感知同时适用于async和sync模式,并且可以将数据发送到新添加的分区中。下面附上topic.metadata.refresh.interval.ms参数的解释:

发生故障(分区丢失,领导者不可用…)时,生产者通常会从代理刷新主题元数据。它还将定期轮询(默认值:每10分钟一次,即600000ms)。如果将此值设置为负值,则仅在失败时刷新元数据。如果将其设置为零,则每条消息发送后元数据都会刷新(不推荐)重要说明:仅在消息发送后刷新才发生,因此,如果生产者从不发送消息,则元数据也不会刷新