C++编程中使用librdkafka库去连接kafka集群经验总结

　　1.C++编程中使用librdkafka库去连接kafka集群分为生产端，消费端两个部分。本次是在https://github.com/edenhill/librdkafka  下载源码编译，安装的。过程很简单不再这里详细说明。

一.生产端使用

在编译完之后会有一个rdkafka_example.cpp，参考他进行编写程序。主要逻辑如下：

RdKafka::Conf *m_conf;
RdKafka::Conf *m_tconf;
RdKafka::Producer *m_producer;
RdKafka::Topic *m_topic;
MyHashPartitionerCb hash_partitioner;
ExampleDeliveryReportCb ex_dr_cb;
ExampleEventCb ex_event_cb;

m_conf = RdKafka::Conf::create(RdKafka::Conf::CONF_GLOBAL);
m_tconf = RdKafka::Conf::create(RdKafka::Conf::CONF_TOPIC);

m_tconf->set("partitioner_cb", &hash_partitioner, errstr)

m_conf->set("metadata.broker.list", m_sBrokenList, errstr);

m_conf->set("queue.buffering.max.ms", "10", errstr); //批量提交

m_conf->set("queue.buffering.max.messages", sMaxMsgs, errstr); 

m_conf->set("event_cb", &ex_event_cb, errstr);

 m_conf->set("dr_cb", &ex_dr_cb, errstr);

m_producer = RdKafka::Producer::create(m_conf, errstr);

m_topic = RdKafka::Topic::create(m_producer, m_topicName,m_tconf, errstr);

RdKafka::ErrorCode resp = m_producer->produce(m_topic, m_PartionCn , RdKafka::Producer::RK_MSG_COPY, pData, iDataLen, (const void *)pKey, strlen(pKey),NULL);

m_producer->poll(1);//可以是poll 0 

注意点：

1.成产过程中的pKey值需要注意一点，不能全部一样，这会影响到各分区的负载均衡。

2.queue.buffering.max.messages" //本地队列缓冲值，应该是在10w-50w之间，太小则会处理不过来引起丢弃。

3.在此基础上加上多线程就满足需求

二.消费端的使用

　1.主要参考rdkafka_consumer_example.cpp 文件。

主要逻辑如下：

RdKafka::Conf *m_conf;
RdKafka::Conf *m_tconf;
m_conf = RdKafka::Conf::create(RdKafka::Conf::CONF_GLOBAL);
m_tconf = RdKafka::Conf::create(RdKafka::Conf::CONF_TOPIC);
ExampleRebalanceCb ex_rebalance_cb;
m_conf->set("rebalance_cb", &ex_rebalance_cb, errstr);
m_conf->set("enable.partition.eof", "true", errstr);

m_conf->set("metadata.broker.list", m_sBrokenList, errstr)  //设置服务列表

m_conf->set("auto.commit.interval.ms", g_conf.sCommitTimeVal, errstr)

m_conf->set("group.id", m_sGroupID, errstr)

MyConsumeCb ex_consume_cb;
m_conf->set("consume_cb", &ex_consume_cb, errstr);

ExampleEventCb ex_event_cb;
m_conf->set("event_cb", &ex_event_cb, errstr);

m_tconf->set("auto.offset.reset", "latest", errstr);
m_conf->set("default_topic_conf", m_tconf, errstr);

RdKafka::KafkaConsumer *consumer = RdKafka::KafkaConsumer::create(m_conf, errstr);

std::vectorstd::string topicsVec;
topicsVec.clear();
topicsVec.push_back(m_sTopicName);
RdKafka::ErrorCode err = consumer->subscribe(topicsVec);

while (m_runFlag)
{
RdKafka::Message *msg = consumer->consume(1000);//1000是超时时间单位毫秒
msg_consume(msg, NULL);
delete msg;
}
consumer->close();
delete consumer;

注意点：

1.消费者组中消费者数量不能大于分区数量。不同的消费者组可以消费相同topic

2.加上多线程即可满足多个消费者同时消费。

附上kafka参数的对照表