Kafka详解六：Kafka如何通过源码实现监控

问题导读：

1.kafka的消费者组的消费偏移存储，kafka支持两个版本?
2.ConsumerOffsetChecker类的作用是什么？
3.Kafka如何通过源码实现监控？

一，基本思路介绍

Kafka作为一个好用的且应用很广泛的消息队列，在大数据处理系统中基本是必不可少的。当然，作为缓存消息的消息队列，我们对其进行流量监控及消费滞后告警就显得异常重要了。

读过前面的文章，<Kafka源码系列之源码解析SimpleConsumer的消费过程>和<Kafka源码系列之Consumer高级API性能分析>这两篇文章的兄弟姐妹应该看本篇文章会很简单。实际就是利用SimpleConsumer获取Partition最新的offset，用Zookeeper的工具获取消费者组的各个分区的消费偏移，两者做差就是lagSize。

但是实际kafka的消费者组的消费偏移存储，kafka支持两个版本的：

1，基于Zookeeper。OffsetFetchRequest.CurrentVersion为0。

2，基于kafka自身。OffsetFetchRequest.CurrentVersion为1(默认)。

那么要实现一个消费者消费滞后预警，就要兼容两种方式，那么我们就详细的来介绍这两种方式的实现。

二，重要工具类

1，ConsumerOffsetChecker

Kafka提供的检查消费者消费偏移，LogEndSize和lagsize的工具。我们实现自己的监控均可以模仿该类实现。本文也仅限于基于该类将实现过程。

2，ZkUtils

Kafka提供的操作Zookeeper的工具类。

3，SimpleConsumer

Kafka消费者实现类。Kafka的副本同步，低级消费者，高级消费者都是基于该类实现从kafka消费消息的。

4，OffsetRequest

消费者去获取分区数据偏移的请求类，对应的请求key是：RequestKeys.OffsetsKey。在kafka的服务端kafkaApis的处理函数是：handleOffsetRequest(request)

5，OffsetFetchRequest

这个是请求某个topic的某个消费组的消费偏移，对应的请求key:RequestKeys.OffsetFetchKey。在kafka的服务端kafkaApis的处理函数是：handleOffsetFetchRequest(request)

6，OffsetManager

偏移管理器。内部维护了一个Scheduler，会定时执行compact，进行偏移的合并。

三，源代码实现

1，首先是获得消费者的消费偏移

ConsumerOffsetChecker当main方法中首先是获得topic列表

[Bash shell] 纯文本查看复制代码

val topicList = topics match {

case Some(x) => x.split(",").view.toList

case None => ZkUtils.getChildren(zkClient, groupDirs.consumerGroupDir + "/owners").toList

}

接着是建立到Broker链接，然后从kafka获取消费者偏移

[Bash shell] 纯文本查看复制代码

val topicPartitions = topicPidMap.flatMap { case(topic, partitionSeq) => partitionSeq.map(TopicAndPartition(topic, _)) }.toSeq
val channel = ClientUtils.channelToOffsetManager(group, zkClient, channelSocketTimeoutMs, channelRetryBackoffMs)
 
debug("Sending offset fetch request to coordinator %s:%d.".format(channel.host, channel.port))
channel.send(OffsetFetchRequest(group, topicPartitions))
val offsetFetchResponse = OffsetFetchResponse.readFrom(channel.receive().buffer)
debug("Received offset fetch response %s.".format(offsetFetchResponse))
 
offsetFetchResponse.requestInfo.foreach { case (topicAndPartition, offsetAndMetadata) =>
  if (offsetAndMetadata == OffsetMetadataAndError.NoOffset) {
    val topicDirs = new ZKGroupTopicDirs(group, topicAndPartition.topic)
    // this group may not have migrated off zookeeper for offsets storage (we don't expose the dual-commit option in this tool
    // (meaning the lag may be off until all the consumers in the group have the same setting for offsets storage)
    try {
      val offset = ZkUtils.readData(zkClient, topicDirs.consumerOffsetDir + "/%d".format(topicAndPartition.partition))._1.toLong
      offsetMap.put(topicAndPartition, offset)
    } catch {
      case z: ZkNoNodeException =>
        if(ZkUtils.pathExists(zkClient,topicDirs.consumerOffsetDir))
          offsetMap.put(topicAndPartition,-1)
        else
          throw z
    }
  }
  else if (offsetAndMetadata.error == ErrorMapping.NoError)
    offsetMap.put(topicAndPartition, offsetAndMetadata.offset)
  else {
    println("Could not fetch offset for %s due to %s.".format(topicAndPartition, ErrorMapping.exceptionFor(offsetAndMetadata.error)))
  }
}

假如，获得的偏移信息为空，那么就从Zookeeper获取消费者偏移。
解决获取topic的分区的最大偏移，实际思路是构建simpleConsumer，然后由其去请求偏移，再跟获取的消费者偏移做差就得到消费者最大偏移。

[Bash shell] 纯文本查看复制代码

topicList.sorted.foreach {
  topic => processTopic(zkClient, group, topic)
}
topicPidMap.get(topic) match {
  case Some(pids) =>
    pids.sorted.foreach {
      pid => processPartition(zkClient, group, topic, pid)
    }
  case None => // ignore
}

在processPartition中

[Bash shell] 纯文本查看复制代码

val offsetOpt = offsetMap.get(topicPartition)
val groupDirs = new ZKGroupTopicDirs(group, topic)
val owner = ZkUtils.readDataMaybeNull(zkClient, groupDirs.consumerOwnerDir + "/%s".format(pid))._1
ZkUtils.getLeaderForPartition(zkClient, topic, pid) match {
  case Some(bid) =>
    val consumerOpt = consumerMap.getOrElseUpdate(bid, getConsumer(zkClient, bid))
    consumerOpt match {
      case Some(consumer) =>
        val topicAndPartition = TopicAndPartition(topic, pid)
        val request =
          OffsetRequest(immutable.Map(topicAndPartition -> PartitionOffsetRequestInfo(OffsetRequest.LatestTime, 1)))
        val logSize = consumer.getOffsetsBefore(request).partitionErrorAndOffsets(topicAndPartition).offsets.head

然后做差得到LagSize

[Bash shell] 纯文本查看复制代码

val lagString = offsetOpt.map(o => if (o == -1) "unknown" else (logSize - o).toString)
println("%-15s %-30s %-3s %-15s %-15s %-15s %s".format(group, topic, pid, offsetOpt.getOrElse("unknown"), logSize, lagString.getOrElse("unknown"),
                                                       owner match {case Some(ownerStr) => ownerStr case None => "none"}))

getConsumer方法中

[Bash shell] 纯文本查看复制代码

private def getConsumer(zkClient: ZkClient, bid: Int): Option[SimpleConsumer] = {
  try {
    ZkUtils.readDataMaybeNull(zkClient, ZkUtils.BrokerIdsPath + "/" + bid)._1 match {
      case Some(brokerInfoString) =>
        Json.parseFull(brokerInfoString) match {
          case Some(m) =>
            val brokerInfo = m.asInstanceOf[Map[String, Any]]
            val host = brokerInfo.get("host").get.asInstanceOf[String]
            val port = brokerInfo.get("port").get.asInstanceOf[Int]
            Some(new SimpleConsumer(host, port, 10000, 100000, "ConsumerOffsetChecker"))
          case None =>
            throw new BrokerNotAvailableException("Broker id %d does not exist".format(bid))
        }
      case None =>
        throw new BrokerNotAvailableException("Broker id %d does not exist".format(bid))
    }
  } catch {
    case t: Throwable =>
      println("Could not parse broker info due to " + t.getCause)
      None
  }
}

四，总结
该工具类的使用

[Bash shell] 纯文本查看复制代码

bin/kafka-consumer-offset-checker.sh --group yourgroup -topic yourtopic --zookeeper  localhost:2181

输出结果

Offset是消费者消费到的偏移，logsize是kafka数据的最大偏移，Lag是二者的差。也即
LagSize = LogSize - Offset
得到我们消费组的滞后情况后，我们就可以根据需求(比如，设定滞后多少消息后给出告警)，给出相应的告警。

转自：http://www.aboutyun.com/forum.php?mod=viewthread&tid=22215&extra=page%3D1&page=1&