序

本文主要讨论一下kafka consumer offset lag的监控

方案

• 利用官方的类库

ConsumerOffsetChecker
ConsumerGroupCommand

• 利用官方的JMX

ConsumerOffsetChecker

在0.8.2.2版本如下
kafka_2.10-0.8.2.2-sources.jar!/kafka/tools/ConsumerOffsetChecker.scala

object ConsumerOffsetChecker extends Logging {private val consumerMap: mutable.Map[Int, Option[SimpleConsumer]] = mutable.Map()private val offsetMap: mutable.Map[TopicAndPartition, Long] = mutable.Map()private var topicPidMap: immutable.Map[String, Seq[Int]] = immutable.Map()private def getConsumer(zkClient: ZkClient, bid: Int): Option[SimpleConsumer] = {//...}private def processPartition(zkClient: ZkClient,group: String, topic: String, pid: Int) {//...}private def processTopic(zkClient: ZkClient, group: String, topic: String) {topicPidMap.get(topic) match {case Some(pids) =>pids.sorted.foreach {pid => processPartition(zkClient, group, topic, pid)}case None => // ignore}}private def printBrokerInfo() {println("BROKER INFO")for ((bid, consumerOpt) <- consumerMap)consumerOpt match {case Some(consumer) =>println("%s -> %s:%d".format(bid, consumer.host, consumer.port))case None => // ignore}}def main(args: Array[String]) {//...try {zkClient = new ZkClient(zkConnect, 30000, 30000, ZKStringSerializer)val topicList = topics match {case Some(x) => x.split(",").view.toListcase None => ZkUtils.getChildren(zkClient, groupDirs.consumerGroupDir +  "/owners").toList}topicPidMap = immutable.Map(ZkUtils.getPartitionsForTopics(zkClient, topicList).toSeq:_*)val topicPartitions = topicPidMap.flatMap { case(topic, partitionSeq) => partitionSeq.map(TopicAndPartition(topic, _)) }.toSeqval 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.toLongoffsetMap.put(topicAndPartition, offset)} catch {case z: ZkNoNodeException =>if(ZkUtils.pathExists(zkClient,topicDirs.consumerOffsetDir))offsetMap.put(topicAndPartition,-1)elsethrow 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)))}}channel.disconnect()println("%-15s %-30s %-3s %-15s %-15s %-15s %s".format("Group", "Topic", "Pid", "Offset", "logSize", "Lag", "Owner"))topicList.sorted.foreach {topic => processTopic(zkClient, group, topic)}if (options.has("broker-info"))printBrokerInfo()for ((_, consumerOpt) <- consumerMap)consumerOpt match {case Some(consumer) => consumer.close()case None => // ignore}}catch {case t: Throwable =>println("Exiting due to: %s.".format(t.getMessage))}finally {for (consumerOpt <- consumerMap.values) {consumerOpt match {case Some(consumer) => consumer.close()case None => // ignore}}if (zkClient != null)zkClient.close()if (channel != null)channel.disconnect()}}
}

缺点就是该类是给命令行调用的，每调用一次，就new一次zkClient，对于监控用来说，不是太合适，需要改造一下，抽取zkClient出来

ConsumerGroupCommand

0.8.2.2以上版本使用ConsumerGroupCommand替代了ConsumerOffsetChecker
kafka_2.11-0.10.2.1-sources.jar!/kafka/admin/ConsumerGroupCommand.scala

object ConsumerGroupCommand extends Logging {//...def main(args: Array[String]) {val opts = new ConsumerGroupCommandOptions(args)if (args.length == 0)CommandLineUtils.printUsageAndDie(opts.parser, "List all consumer groups, describe a consumer group, or delete consumer group info.")// should have exactly one actionval actions = Seq(opts.listOpt, opts.describeOpt, opts.deleteOpt).count(opts.options.has _)if (actions != 1)CommandLineUtils.printUsageAndDie(opts.parser, "Command must include exactly one action: --list, --describe, --delete")opts.checkArgs()val consumerGroupService = {if (opts.useOldConsumer) {System.err.println("Note: This will only show information about consumers that use ZooKeeper (not those using the Java consumer API).\n")new ZkConsumerGroupService(opts)} else {System.err.println("Note: This will only show information about consumers that use the Java consumer API (non-ZooKeeper-based consumers).\n")new KafkaConsumerGroupService(opts)}}try {if (opts.options.has(opts.listOpt))consumerGroupService.listGroups().foreach(println(_))else if (opts.options.has(opts.describeOpt)) {val (state, assignments) = consumerGroupService.describeGroup()val groupId = opts.options.valuesOf(opts.groupOpt).asScala.headassignments match {case None =>// applies to both old and new consumerprintError(s"The consumer group '$groupId' does not exist.")case Some(assignments) =>if (opts.useOldConsumer)printAssignment(assignments, false)elsestate match {case Some("Dead") =>printError(s"Consumer group '$groupId' does not exist.")case Some("Empty") =>System.err.println(s"Consumer group '$groupId' has no active members.")printAssignment(assignments, true)case Some("PreparingRebalance") | Some("AwaitingSync") =>System.err.println(s"Warning: Consumer group '$groupId' is rebalancing.")printAssignment(assignments, true)case Some("Stable") =>printAssignment(assignments, true)case other =>// the control should never reach herethrow new KafkaException(s"Expected a valid consumer group state, but found '${other.getOrElse("NONE")}'.")}}}else if (opts.options.has(opts.deleteOpt)) {consumerGroupService match {case service: ZkConsumerGroupService => service.deleteGroups()case _ => throw new IllegalStateException(s"delete is not supported for $consumerGroupService.")}}} catch {case e: Throwable =>printError(s"Executing consumer group command failed due to ${e.getMessage}", Some(e))} finally {consumerGroupService.close()}}
}

也是基于命令行来设计的

JMX

这个是利用kafka本身写入的JMX的数据，就不用额外在去像ConsumerOffsetChecker去自己连接再去获取。比如

            ObjectName oName = new ObjectName("kafka.producer:*");Set<ObjectName> metricsBeans = mBeanServer.queryNames(oName, null);for (ObjectName mBeanName : metricsBeans) {MBeanInfo metricsBean = mBeanServer.getMBeanInfo(mBeanName);MBeanAttributeInfo[] metricsAttrs = metricsBean.getAttributes();for (MBeanAttributeInfo metricsAttr : metricsAttrs) {//get valueObject value = mBeanServer.getAttribute(mBeanName, metricsAttr.getName());//process ...}}

小结

可以自己改造ConsumerOffsetChecker或者ConsumerGroupCommand，然后上报到statsd或者Prometheus。当然能利用JMX是最省事的了。

doc

• kafka官方JMX+Reporters