Kafka 简易教程

1.初识概念

Apache Kafka是一个分布式消息发布订阅系统。

Topic
Kafka将消息种子(Feed)分门别类， 每一类的消息称之为话题(Topic).

Producer
发布消息的对象称之为话题生产者(Kafka topic producer)

Consumer
订阅消息并处理发布的消息的种子的对象称之为话题消费者(consumers)
Broker
已发布的消息保存在一组服务器中，称之为Kafka集群。集群中的每一个服务器都是一个代理(Broker). 消费者可以订阅一个或多个话题，并从Broker拉数据，从而消费这些已发布的消息。

分区

一个topic可以有一个或多个分区，每一个分区都是一个顺序的、不可变的消息队列， 并且可以持续的添加。分区中的消息都被分配了一个序列号，称之为偏移量(offset),在每个分区中此偏移量都是唯一的。

每个分区有一个leader，零或多个follower。Leader处理此分区的所有的读写请求而follower被动的复制数据。如果leader当机，其它的一个follower会被推举为新的leader。

通过分区的概念，Kafka可以在多个consumer组并发的情况下提供较好的有序性和负载均衡。将每个分区分只分发给一个consumer组，这样一个分区就只被这个组的一个consumer消费，就可以顺序的消费这个分区的消息。因为有多个分区，依然可以在多个consumer组之间进行负载均衡。注意consumer组的数量不能多于分区的数量，也就是有多少分区就允许多少并发消费。

Kafka 只能保证一个分区之内消息的有序性，在不同的分区之间是不可以的，这已经可以满足大部分应用的需求。如果需要 topic 中所有消息的有序性，那就只能让这个 topic 只有一个分区，当然也就只有一个 consumer 组消费它。

2.安装使用

1. 下载 Kafka

• 下载 wget http://apache.01link.hk/kafka/0.10.0.0/kafka_2.11-0.10.0.0.tgz 或者
• wget http://ftp.cuhk.edu.hk/pub/packages/apache.org/kafka/0.10.0.0/kafka_2.11-0.10.0.0.tgz（看哪个源比较快）
• 解压 tar -xzf kafka_2.11-0.10.0.0.tgz
• 进入文件夹 cd kafka_2.11-0.10.0.0/

2. 启动服务

• 启动 ZooKeeper bin/zookeeper-server-start.sh config/zookeeper.properties &（利用 &放到后台方便继续操作）
• 启动 Kafka bin/kafka-server-start.sh config/server.properties &

3. 创建一个叫做 dawang 的 topic，它只有一个分区，一个副本

• 创建 bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic dawang
• 查看 bin/kafka-topics.sh --list --zookeeper localhost:2181
• 还可以配置 broker 让它自动创建 topic

4. 发送消息。Kafka 使用一个简单的命令行producer，从文件中或者从标准输入中读取消息并发送到服务端。默认的每条命令将发送一条消息。

• 发送消息 bin/kafka-console-producer.sh --broker-list localhost:9092 --topic dawang（然后可以随意输入内容，回车可以发送，ctrl+c 退出）

5. 启动 consumer。可以读取消息并输出到标准输出：

• 接收消息 bin/kafka-console-consumer.sh --zookeeper localhost:2181 --topic dawang --from-beginning
• 在一个终端中运行 consumer 命令行，另一个终端中运行 producer 命令行，就可以在一个终端输入消息，另一个终端读取消息。这两个命令都有自己的可选参数，可以在运行的时候不加任何参数可以看到帮助信息。

6. 搭建一个多个 broker 的集群，启动有 3 个 broker 组成的集群，这些 broker 节点也都在本机

首先复制一下配置文件：cp config/server.properties config/server-1.properties 和 cp config/server.properties config/server-2.properties

两个文件需要改动的内容为：

config/server-1.properties:

broker.id=1

listeners=PLAINTEXT://:9093

log.dir=/tmp/kafka-logs-1

 

config/server-2.properties:

broker.id=2

listeners=PLAINTEXT://:9094

log.dir=/tmp/kafka-logs-2

这里我们把 broker id, 端口号和日志地址配置成和之前不一样，然后我们启动这两个 broker：

bin/kafka-server-start.sh config/server-1.properties &

bin/kafka-server-start.sh config/server-2.properties &

然后创建一个复制因子为 3 的 topic

bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 3 --partitions 1 --topic oh3topic

可以使用 describe 命令来显示 topic 详情

bin/kafka-topics.sh --describe --zookeeper localhost:2181 --topic oh3topic

Topic:oh3topic PartitionCount:1 ReplicationFactor:3 Configs:

Topic: oh3topic Partition: 0 Leader: 0 Replicas: 0,1,2 Isr: 0,1,2

我们也可以来看看之前的另一个 topic 的情况

bin/kafka-topics.sh --describe --zookeeper localhost:2181 --topic dawang

Topic:dawang PartitionCount:1 ReplicationFactor:1 Configs:

Topic: dawang Partition: 0 Leader: 0 Replicas: 0 Isr: 0

最后我们可以按照同样的方法来生产和消费消息，例如

#生产

bin/kafka-console-producer.sh --broker-list localhost:9092 --topic oh3topic

# 消费

bin/kafka-console-consumer.sh --zookeeper localhost:2181 --from-beginning --topic oh3topic

开俩终端就可以一边生产消息，一边消费消息了。

测试一下容错. 干掉leader,也就是Broker 1:

ps -ef | grep server-1.properties

Leader被切换到一个follower上节, 点 1 不会被列在isr中了，因为它死了:

再次使用 describe 命令来显示 topic 详情

bin/kafka-topics.sh --describe --zookeeper localhost:2181 --topic oh3topic

但是，消息没丢啊，不信你试试:

彻底删除kafka中的topic

1、删除kafka存储目录（server.properties文件log.dirs配置，默认为"/tmp/kafka-logs"）相关topic目录
2、Kafka 删除topic的命令是：

如果kafaka启动时加载的配置文件中server.properties没有配置delete.topic.enable=true，那么此时的删除并不是真正的删除，而是把topic标记为：marked for deletion

bin/kafka-topics.sh --delete --zookeeper localhost:2181 --topic oh3topic

3.代码实例

需要自行安装librdkafka库

https://github.com/edenhill/librdkafka

produce

#include <stdio.h>
#include <stdlib.h>
#include <iostream>
#include <list>
#include <memory>
#include <string>
#include <string.h>
#include "librdkafka/rdkafkacpp.h"
//#include "librdkafka/rdkafka.h"
using namespace std;

bool run = true;

class ExampleDeliveryReportCb : public RdKafka::DeliveryReportCb
{
 public:
  void dr_cb (RdKafka::Message &message) {
    std::cout << "Message delivery for (" << message.len() << " bytes): " <<
        message.errstr() << std::endl;
    if (message.key())
      std::cout << "Key: " << *(message.key()) << ";" << std::endl;
  }
};

class ExampleEventCb : public RdKafka::EventCb {
 public:
  void event_cb (RdKafka::Event &event) {
    switch (event.type())
    {
      case RdKafka::Event::EVENT_ERROR:
        std::cerr << "ERROR (" << RdKafka::err2str(event.err()) << "): " <<
            event.str() << std::endl;
        if (event.err() == RdKafka::ERR__ALL_BROKERS_DOWN)
          run = false;
        break;

      case RdKafka::Event::EVENT_STATS:
        std::cerr << "\"STATS\": " << event.str() << std::endl;
        break;

      case RdKafka::Event::EVENT_LOG:
        fprintf(stderr, "LOG-%i-%s: %s\n",
                event.severity(), event.fac().c_str(), event.str().c_str());
        break;

      default:
        std::cerr << "EVENT " << event.type() <<
            " (" << RdKafka::err2str(event.err()) << "): " <<
            event.str() << std::endl;
        break;
    }
  }
};

/* Use of this partitioner is pretty pointless since no key is provided * in the produce() call.so when you need input your key */
class MyHashPartitionerCb : public RdKafka::PartitionerCb {
    public:
        int32_t partitioner_cb (const RdKafka::Topic *topic, const std::string *key,int32_t partition_cnt, void *msg_opaque)
        {
            std::cout<<"partition_cnt="<<partition_cnt<<std::endl;
            return djb_hash(key->c_str(), key->size()) % partition_cnt;
        }
    private:
        static inline unsigned int djb_hash (const char *str, size_t len)
        {
        unsigned int hash = ;
        for (size_t i =  ; i < len ; i++)
            hash = ((hash << ) + hash) + str[i];
        std::cout<<"hash1="<<hash<<std::endl;

        return hash;
        }
};

void TestProducer()
{
    std::string brokers = "localhost";
    std::string errstr;
    std::string topic_str="helloworld_kugou_test";//自行制定主题topic
    MyHashPartitionerCb hash_partitioner;
    int32_t partition = RdKafka::Topic::PARTITION_UA;
    int64_t start_offset = RdKafka::Topic::OFFSET_BEGINNING;
    bool do_conf_dump = false;
    int opt;

    int use_ccb = ;

    //Create configuration objects
    RdKafka::Conf *conf = RdKafka::Conf::create(RdKafka::Conf::CONF_GLOBAL);
    RdKafka::Conf *tconf = RdKafka::Conf::create(RdKafka::Conf::CONF_TOPIC);

    if (tconf->set("partitioner_cb", &hash_partitioner, errstr) != RdKafka::Conf::CONF_OK)
     {
          std::cerr << errstr << std::endl;
          exit();
     }

    /* * Set configuration properties */
    conf->set("metadata.broker.list", brokers, errstr);
    ExampleEventCb ex_event_cb;
    conf->set("event_cb", &ex_event_cb, errstr);

    ExampleDeliveryReportCb ex_dr_cb;

    /* Set delivery report callback */
    conf->set("dr_cb", &ex_dr_cb, errstr);

    /* * Create producer using accumulated global configuration. */
    RdKafka::Producer *producer = RdKafka::Producer::create(conf, errstr);
    if (!producer)
    {
        std::cerr << "Failed to create producer: " << errstr << std::endl;
        exit();
    }

    std::cout << "% Created producer " << producer->name() << std::endl;

    /* * Create topic handle. */
    RdKafka::Topic *topic = RdKafka::Topic::create(producer, topic_str, tconf, errstr);
    if (!topic) {
      std::cerr << "Failed to create topic: " << errstr << std::endl;
      exit();
    }

    /* * Read messages from stdin and produce to broker. */
    for (std::string line; run && std::getline(std::cin, line);)
    {
        if (line.empty())
        {
            producer->poll();
            continue;
        }

      /* * Produce message // 1. topic // 2. partition // 3. flags // 4. payload // 5. payload len // 6. std::string key // 7. msg_opaque? NULL */
      std::string key=line.substr(,);//根据line前5个字符串作为key值
      // int a = MyHashPartitionerCb::djb_hash(key.c_str(),key.size());
      // std::cout<<"hash="<<a<<std::endl;
      RdKafka::ErrorCode resp = producer->produce(topic, partition,
          RdKafka::Producer::RK_MSG_COPY /* Copy payload */,
          const_cast<char *>(line.c_str()), line.size(),
          key.c_str(), key.size(), NULL);//这里可以设计key值,因为会根据key值放在对应的partition
        if (resp != RdKafka::ERR_NO_ERROR)
            std::cerr << "% Produce failed: " <<RdKafka::err2str(resp) << std::endl;
        else
            std::cerr << "% Produced message (" << line.size() << " bytes)" <<std::endl;
        producer->poll();//对于socket进行读写操作。poll方法才是做实际的IO操作的。return the number of events served
    }
    //
    run = true;

    while (run && producer->outq_len() > ) {
      std::cerr << "Waiting for " << producer->outq_len() << std::endl;
      producer->poll();
    }

    delete topic;
    delete producer;
}

int main(int argc, char *argv[])
{
    TestProducer();
    return EXIT_SUCCESS;
}

consumer

#include <stdio.h>
#include <stdlib.h>
#include <iostream>
#include <list>
#include <memory>
#include <string>
#include <string.h>
#include "librdkafka/rdkafkacpp.h"
using namespace std;

bool run = true;
bool exit_eof = true;
class ExampleDeliveryReportCb : public RdKafka::DeliveryReportCb
{
 public:
  void dr_cb (RdKafka::Message &message) {
    std::cout << "Message delivery for (" << message.len() << " bytes): " <<
        message.errstr() << std::endl;
    if (message.key())
      std::cout << "Key: " << *(message.key()) << ";" << std::endl;
  }
};

class ExampleEventCb : public RdKafka::EventCb {
 public:
  void event_cb (RdKafka::Event &event) {
    switch (event.type())
    {
      case RdKafka::Event::EVENT_ERROR:
        std::cerr << "ERROR (" << RdKafka::err2str(event.err()) << "): " <<
            event.str() << std::endl;
        if (event.err() == RdKafka::ERR__ALL_BROKERS_DOWN)
          run = false;
        break;

      case RdKafka::Event::EVENT_STATS:
        std::cerr << "\"STATS\": " << event.str() << std::endl;
        break;

      case RdKafka::Event::EVENT_LOG:
        fprintf(stderr, "LOG-%i-%s: %s\n",
                event.severity(), event.fac().c_str(), event.str().c_str());
        break;

      default:
        std::cerr << "EVENT " << event.type() <<
            " (" << RdKafka::err2str(event.err()) << "): " <<
            event.str() << std::endl;
        break;
    }
  }
};

/* Use of this partitioner is pretty pointless since no key is provided * in the produce() call.so when you need input your key */
class MyHashPartitionerCb : public RdKafka::PartitionerCb {
    public:
        int32_t partitioner_cb (const RdKafka::Topic *topic, const std::string *key,int32_t partition_cnt, void *msg_opaque)
        {
            std::cout<<"partition_cnt="<<partition_cnt<<std::endl;
            return djb_hash(key->c_str(), key->size()) % partition_cnt;
        }
    private:
        static inline unsigned int djb_hash (const char *str, size_t len)
        {
        unsigned int hash = ;
        for (size_t i =  ; i < len ; i++)
            hash = ((hash << ) + hash) + str[i];
        std::cout<<"hash1="<<hash<<std::endl;

        return hash;
        }
};

void msg_consume(RdKafka::Message* message, void* opaque)
{
    switch (message->err())
    {
        case RdKafka::ERR__TIMED_OUT:
            break;

        case RdKafka::ERR_NO_ERROR:
          /* Real message */
            std::cout << "Read msg at offset " << message->offset() << std::endl;
            if (message->key())
            {
                std::cout << "Key: " << *message->key() << std::endl;
            }
            printf("%.*s\n", static_cast<int>(message->len()),static_cast<const char *>(message->payload()));
            break;
        case RdKafka::ERR__PARTITION_EOF:
              /* Last message */
              if (exit_eof)
              {
                  run = false;
                  cout << "ERR__PARTITION_EOF" << endl;
              }
              break;
        case RdKafka::ERR__UNKNOWN_TOPIC:
        case RdKafka::ERR__UNKNOWN_PARTITION:
            std::cerr << "Consume failed: " << message->errstr() << std::endl;
            run = false;
            break;
    default:
        /* Errors */
        std::cerr << "Consume failed: " << message->errstr() << std::endl;
        run = false;
    }
}
class ExampleConsumeCb : public RdKafka::ConsumeCb {
    public:
        void consume_cb (RdKafka::Message &msg, void *opaque)
        {
            msg_consume(&msg, opaque);
        }
};
void TestConsumer()
{
    std::string brokers = "localhost";
    std::string errstr;
    std::string topic_str="helloworld_kugou_test";//helloworld_kugou
    MyHashPartitionerCb hash_partitioner;
    int32_t partition = RdKafka::Topic::PARTITION_UA;//为何不能用？？在Consumer这里只能写0？？？无法自动吗？？？
    partition = ;
    int64_t start_offset = RdKafka::Topic::OFFSET_BEGINNING;
    bool do_conf_dump = false;
    int opt;

    int use_ccb = ;

    //Create configuration objects
    RdKafka::Conf *conf = RdKafka::Conf::create(RdKafka::Conf::CONF_GLOBAL);
    RdKafka::Conf *tconf = RdKafka::Conf::create(RdKafka::Conf::CONF_TOPIC);

    if (tconf->set("partitioner_cb", &hash_partitioner, errstr) != RdKafka::Conf::CONF_OK)
    {
        std::cerr << errstr << std::endl;
        exit();
    }

    /* * Set configuration properties */
    conf->set("metadata.broker.list", brokers, errstr);
    ExampleEventCb ex_event_cb;
    conf->set("event_cb", &ex_event_cb, errstr);

    ExampleDeliveryReportCb ex_dr_cb;

    /* Set delivery report callback */
    conf->set("dr_cb", &ex_dr_cb, errstr);
    /* * Create consumer using accumulated global configuration. */
    RdKafka::Consumer *consumer = RdKafka::Consumer::create(conf, errstr);
    if (!consumer)
    {
      std::cerr << "Failed to create consumer: " << errstr << std::endl;
      exit();
    }

    std::cout << "% Created consumer " << consumer->name() << std::endl;

    /* * Create topic handle. */
    RdKafka::Topic *topic = RdKafka::Topic::create(consumer, topic_str, tconf, errstr);
    if (!topic)
    {
      std::cerr << "Failed to create topic: " << errstr << std::endl;
      exit();
    }

    /* * Start consumer for topic+partition at start offset */
    RdKafka::ErrorCode resp = consumer->start(topic, partition, start_offset);
    if (resp != RdKafka::ERR_NO_ERROR) {
      std::cerr << "Failed to start consumer: " << RdKafka::err2str(resp) << std::endl;
      exit();
    }

    ExampleConsumeCb ex_consume_cb;

    /* * Consume messages */
    while (run)
    {
        if (use_ccb)
        {
            consumer->consume_callback(topic, partition, , &ex_consume_cb, &use_ccb);
      }
      else
      {
          RdKafka::Message *msg = consumer->consume(topic, partition, );
          msg_consume(msg, NULL);
          delete msg;
      }
      consumer->poll();
    }

    /* * Stop consumer */
    consumer->stop(topic, partition);

    consumer->poll();

    delete topic;
    delete consumer;
}

int main(int argc, char *argv[])
{
    TestConsumer();
    return EXIT_SUCCESS;
}