Kafka: Get broker host from ZooKeeper

Question

For particular reasons I need to use both - ConsumerGroup (a.k.a. high-level consumer) and SimpleConsumer (a.k.a. low-level consumer) to read from Kafka. For ConsumerGroup I use ZooKeeper-based config and am completely satisfied with it, but SimpleConsumer requires seed brokers to be instantiated.

I don't want to keep list of both - ZooKeeper and broker hosts. Thus, I'm looking for a way to automatically discover brokers for a particular topic from ZooKeeper.

Because of some indirect information I belief that these data is stored in ZooKeeper under one of the following paths:

• /brokers/topics/<topic>/partitions/<partition-id>/state
• /brokers/ids/

However, when I try to read data from these nodes, I'm getting serialization error (I'm using com.101tec.zkclient for this):

org.I0Itec.zkclient.exception.ZkMarshallingError: java.io.StreamCorruptedException: invalid stream header: 7B226A6D at org.I0Itec.zkclient.serialize.SerializableSerializer.deserialize(SerializableSerializer.java:37) at org.I0Itec.zkclient.ZkClient.derializable(ZkClient.java:740) at org.I0Itec.zkclient.ZkClient.readData(ZkClient.java:773) at org.I0Itec.zkclient.ZkClient.readData(ZkClient.java:761) at org.I0Itec.zkclient.ZkClient.readData(ZkClient.java:750) at org.I0Itec.zkclient.ZkClient.readData(ZkClient.java:744) ... 64 elided Caused by: java.io.StreamCorruptedException: invalid stream header: 7B226A6D at java.io.ObjectInputStream.readStreamHeader(ObjectInputStream.java:804) at java.io.ObjectInputStream.(ObjectInputStream.java:299) at org.I0Itec.zkclient.serialize.TcclAwareObjectIputStream.(TcclAwareObjectIputStream.java:30) at org.I0Itec.zkclient.serialize.SerializableSerializer.deserialize(SerializableSerializer.java:31) ... 69 more

I can write and read custom Java objects (e.g. Strings) without any problem, so I believe it's not a problem of a client, but rather tricky encoding. Thus, I want to know:

1. If this is the right way to go, how to read these nodes properly?
2. If the whole approach is wrong, what is the right one?

Bloodless

One tool that I found helpful to poke around in the Zookeeper data is this one: code.google.com/p/zooviewer – Karynkaryo

Answer 1

That is the way of what one of my colleagues did to get a list of Kafka brokers. I think it's a correct way when you want to get a broker list dynamically.

Here is an example code that shows how to get the list.

public class KafkaBrokerInfoFetcher {

    public static void main(String[] args) throws Exception {
        ZooKeeper zk = new ZooKeeper("localhost:2181", 10000, null);
        List<String> ids = zk.getChildren("/brokers/ids", false);
        for (String id : ids) {
            String brokerInfo = new String(zk.getData("/brokers/ids/" + id, false, null));
            System.out.println(id + ": " + brokerInfo);
        }
    }
}

Running the code onto the cluster consisting of three brokers results in

1: {"jmx_port":-1,"timestamp":"1428512949385","host":"192.168.0.11","version":1,"port":9093}
2: {"jmx_port":-1,"timestamp":"1428512955512","host":"192.168.0.11","version":1,"port":9094}
3: {"jmx_port":-1,"timestamp":"1428512961043","host":"192.168.0.11","version":1,"port":9095}

Understate

Thanks and sorry for late reply. Could you please also post reference to the ZooKeeper client library you are using in your code? On my side, I figured out how to fix an error I described and will post alternative solution using com.101toc.ZkClient which comes with Kafka libraries. Anyway, I accept your answer as totally valid. – Bloodless

It should be import org.apache.zookeeper.ZooKeeper; – Slovak

+1 tooo simple and tooo efficient way to get the information about kafka (brokes, topics, partitions per topic ...etc) Thanx. It needs nothing but import org.apache.zookeeper.ZooKeeper – Tobi

Answer 2

It turns out that Kafka uses ZKStringSerializer to read and write data into znodes. So, to fix the error I only had to add it as a last parameter in ZkClient constructor:

val zkClient = new ZkClient(zkQuorum, Integer.MAX_VALUE, 10000, ZKStringSerializer)

Using it, I wrote several useful functions for discovering broker ids, their addresses and other stuff:

import kafka.utils.Json
import kafka.utils.ZKStringSerializer
import kafka.utils.ZkUtils
import org.I0Itec.zkclient.ZkClient
import org.apache.kafka.common.KafkaException


def listBrokers(): List[Int] = {
  zkClient.getChildren("/brokers/ids").toList.map(_.toInt)
}

def listTopics(): List[String] = {
  zkClient.getChildren("/brokers/topics").toList
}

def listPartitions(topic: String): List[Int] = {
  val path = "/brokers/topics/" + topic + "/partitions"
  if (zkClient.exists(path)) {
    zkClient.getChildren(path).toList.map(_.toInt)
  } else {
    throw new KafkaException(s"Topic ${topic} doesn't exist")
  }
}

def getBrokerAddress(brokerId: Int): (String, Int) = {
  val path = s"/brokers/ids/${brokerId}"
  if (zkClient.exists(path)) {
    val brokerInfo = readZkData(path)
    (brokerInfo.get("host").get.asInstanceOf[String], brokerInfo.get("port").get.asInstanceOf[Int])
  } else {
    throw new KafkaException("Broker with ID ${brokerId} doesn't exist")
  }
}

def getLeaderAddress(topic: String, partitionId: Int): (String, Int) = {
  val path = s"/brokers/topics/${topic}/partitions/${partitionId}/state"
  if (zkClient.exists(path)) {
    val leaderStr = zkClient.readData[String](path)
    val leaderId = Json.parseFull(leaderStr).get.asInstanceOf[Map[String, Any]].get("leader").get.asInstanceOf[Int]
    getBrokerAddress(leaderId)
  } else {
    throw new KafkaException(s"Topic (${topic}) or partition (${partitionId}) doesn't exist")
  }
}

Bloodless

what library is kafka.utils.ZkUtils? 'org.apache.kafka', name: 'kafka_2.13'? That means we have to specify the version of kafka we are using? – Ultra

Answer 3

To do this using the shell:

zookeeper-shell myzookeeper.example.com:2181
ls /brokers/ids
  => [2, 1, 0]
get /brokers/ids/2
get /brokers/ids/1
get /brokers/ids/0 

Jazzman

Answer 4

actually, there is ZkUtils from within Kafka (at least for 0.8.x line), that you can use with one small caveat: you'll need to re-implement ZkStringSerializer that would convert strings as UTF-8 encoded byte arrays. If you'd like to use Java8's streaming APIs, you can iterate over Scala collections throug scala.collection.JavaConversions. This is the thing that helped my case.

Allyn

Answer 5

 public KafkaProducer(String zookeeperAddress, String topic) throws IOException,
        KeeperException, InterruptedException {

    this.zookeeperAddress = zookeeperAddress;
    this.topic = topic;

    ZooKeeper zk = new ZooKeeper(zookeeperAddress, 10000, null);
    List<String> brokerList = new ArrayList<String>();

    List<String> ids = zk.getChildren("/brokers/ids", false);
    for (String id : ids) {
        String brokerInfoString = new String(zk.getData("/brokers/ids/" + id, false, null));
        Broker broker = Broker.createBroker(Integer.valueOf(id), brokerInfoString);
        if (broker != null) {
            brokerList.add(broker.connectionString());
        }
    }

    props.put("serializer.class", KAFKA_STRING_ENCODER);
    props.put("metadata.broker.list", String.join(",", brokerList));
    producer = new Producer<String, String>(new ProducerConfig(props));
}

Valley