kafka的分区消费模型

分区消费模型是kafka的消费者编程模型。其模型如下所示：

主要是一个consumer对应一个分区。而分区消费的伪代码如下所示：

2017-05-07_111849.png

kafka的组消费模型

kafka按照组进行消费的时候一个kafka组中的消费者可以获取到kafka集群中的所有数据以供消费。
组消费模型的伪代码描述如下：

• 上面的流数代表每个consumer组里面包含的consumer实例个数。

kafka Topic的分配算法如下所示：

消费模型的对比：

• 分区消费模型：较为灵活，但需要自己处理各种异常情况；且需要自己管理offset以实现消息传递的其他语义。
• 组消费模型：更加简单，但不灵活，不需要自己处理异常，不需要自己管理offset，其只能实现kafka默认的最少一次消息传递语义（可能会发生重复）。
• 消息传递语义有三种：最少一次（消费者收到的消息可能会重复），最多一个（消费者可能收不到这条消息），有且仅有一次（不会发生重复也不会丢失）

分区消费模型的python实现

eversilver@debian:~/silverTest/kafka/kafka/projects/consumer/partition$ cat partition_consumer.py
#!/usr/bin/env python
# coding=utf-8
import threading
from kafka.client import KafkaClient
from kafka.consumer import SimpleConsumer

class Consumer(threading.Thread):
    daemon=True
    def __init__(self, partition_index):
        threading.Thread.__init__(self)
        self.part = [partition_index]
        self.__offset = 0

    def run(self):
        client = KafkaClient("192.168.128.128:19092,192.168.128.129:19092")
        consumer=SimpleConsumer(client,"test-group","myTest",auto_commit=False,partitions=self.part)

        consumer.seek(0,0)

        while True:
            message = consumer.get_message(True, 60)
            self.__offset = message.offset
            print (message.message.value)
eversilver@debian:~/silverTest/kafka/kafka/projects/consumer/partition$ cat main.py 
#!/usr/bin/env python
# coding=utf-8
import logging, time
import partition_consumer

def main():
    threads = []
    partition = 3
    for index in range(partition):
        threads.append(partition_consumer.Consumer(index))

    for t in threads:
        t.start()

    time.sleep(50000)

if __name__ == '__main__':
    main()

组消费模型的python实现

eversilver@debian:~/silverTest/kafka/kafka/projects/consumer/group$ cat group_consumer.py 
#!/usr/bin/env python
# coding=utf-8
import threading
from kafka.client import KafkaClient
from kafka.consumer import SimpleConsumer

class Consumer(threading.Thread):
    daemon = True

    def run(self):
        client = KafkaClient("192.168.128.128:19092,192.168.128.129:19092,192.168.128.130:19092")
        consumer = SimpleConsumer(client, "test-group", "mytest")

        for message in consumer:
            print(message.message.value)
eversilver@debian:~/silverTest/kafka/kafka/projects/consumer/group$ cat main.py 
#!/usr/bin/env python
# coding=utf-8
import group_consumer
import time

def main():
    consumer_thread = group_consumer.Consumer()
    consumer_thread.start()

    time.sleep(500000)

if __name__ == '__main__':
    main()

python客户端参数调优

• fetch_size_bytes:从服务器获取得到的单个包的大小
• buffer_size:kafka客户端缓冲区大小（一次最多可以从服务器获取的数据大小）
• Group:分组消费的分组名
• auto_commit:offset是否自动进行提交（一般用于分区消费模型）

分组消费模式java实现

package kafka.consumer.group;

import kafka.consumer.ConsumerIterator;
import kafka.consumer.KafkaStream;
 
public class ConsumerTest implements Runnable {
    private KafkaStream m_stream;
    private int m_threadNumber;
 
    public ConsumerTest(KafkaStream a_stream, int a_threadNumber) {
        m_threadNumber = a_threadNumber;
        m_stream = a_stream;
    }
 
    public void run() {
        ConsumerIterator<byte[], byte[]> it = m_stream.iterator();
        while (it.hasNext()){
            System.out.println("Thread " + m_threadNumber + ": " + new String(it.next().message()));
            
        }
        System.out.println("Shutting down Thread: " + m_threadNumber);
    }
}

package kafka.consumer.group;

import kafka.consumer.ConsumerConfig;
import kafka.consumer.KafkaStream;
import kafka.javaapi.consumer.ConsumerConnector;
 
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;

public class GroupConsumerTest extends Thread {
    private final ConsumerConnector consumer;
    private final String topic;
    private  ExecutorService executor;
    
    public GroupConsumerTest(String a_zookeeper, String a_groupId, String a_topic){
        consumer = kafka.consumer.Consumer.createJavaConsumerConnector(
                createConsumerConfig(a_zookeeper, a_groupId));
        this.topic = a_topic;
    }
    
    public void shutdown() {
        if (consumer != null) consumer.shutdown();
        if (executor != null) executor.shutdown();
        try {
            if (!executor.awaitTermination(Long.MAX_VALUE, TimeUnit.MILLISECONDS)) {
                System.out.println("Timed out waiting for consumer threads to shut down, exiting uncleanly");
            }
        } catch (InterruptedException e) {
            System.out.println("Interrupted during shutdown, exiting uncleanly");
        }
   }
 
    public void run(int a_numThreads) {
        Map<String, Integer> topicCountMap = new HashMap<String, Integer>();
        topicCountMap.put(topic, new Integer(a_numThreads));
        Map<String, List<KafkaStream<byte[], byte[]>>> consumerMap = consumer.createMessageStreams(topicCountMap);
        List<KafkaStream<byte[], byte[]>> streams = consumerMap.get(topic);
 
        // now launch all the threads
        //
        executor = Executors.newFixedThreadPool(a_numThreads);
 
        // now create an object to consume the messages
        //
        int threadNumber = 0;
        for (final KafkaStream stream : streams) {
            executor.submit(new ConsumerTest(stream, threadNumber));
            threadNumber++;
        }
    }
    private static ConsumerConfig createConsumerConfig(String a_zookeeper, String a_groupId) {
        Properties props = new Properties();
        props.put("zookeeper.connect", a_zookeeper);
        props.put("group.id", a_groupId);
        props.put("zookeeper.session.timeout.ms", "40000");
        props.put("zookeeper.sync.time.ms", "2000");
        props.put("auto.commit.interval.ms", "1000");
 
        return new ConsumerConfig(props);
    }
    
    public static void main(String[] args) {
        if(args.length < 1){
            System.out.println("Please assign partition number.");
        }
        
        String zooKeeper = "10.206.216.13:12181,10.206.212.14:12181,10.206.209.25:12181";
        String groupId = "jikegrouptest";
        String topic = "jiketest";
        int threads = Integer.parseInt(args[0]);
 
        GroupConsumerTest example = new GroupConsumerTest(zooKeeper, groupId, topic);
        example.run(threads);
 
        try {
            Thread.sleep(Long.MAX_VALUE);
        } catch (InterruptedException ie) {
 
        }
        example.shutdown();
    }
}

分区消费模式的java实现

package kafka.consumer.partition;

import kafka.api.FetchRequest;
import kafka.api.FetchRequestBuilder;
import kafka.api.PartitionOffsetRequestInfo;
import kafka.common.ErrorMapping;
import kafka.common.TopicAndPartition;
import kafka.javaapi.*;
import kafka.javaapi.consumer.SimpleConsumer;
import kafka.message.MessageAndOffset;
 
import java.nio.ByteBuffer;
import java.util.ArrayList;
import java.util.Collections;
import java.util.HashMap;
import java.util.List;
import java.util.Map;

public class PartitionConsumerTest {
    public static void main(String args[]) {
        PartitionConsumerTest example = new PartitionConsumerTest();
        long maxReads = Long.MAX_VALUE;
        String topic = "jiketest";
        if(args.length < 1){
            System.out.println("Please assign partition number.");
        }
        
        List<String> seeds = new ArrayList<String>();
        String hosts="10.206.216.13,10.206.212.14,10.206.209.25";
        String[] hostArr = hosts.split(",");
        for(int index = 0;index < hostArr.length;index++){
            seeds.add(hostArr[index].trim());
        }
        
        int port = 19092;
         
        int partLen = Integer.parseInt(args[0]);
        for(int index=0;index < partLen;index++){
            try {
                example.run(maxReads, topic, index/*partition*/, seeds, port);
            } catch (Exception e) {
                System.out.println("Oops:" + e);
                 e.printStackTrace();
            }
        }
    }
    
    private List<String> m_replicaBrokers = new ArrayList<String>();
     
        public PartitionConsumerTest() {
            m_replicaBrokers = new ArrayList<String>();
        }
     
        public void run(long a_maxReads, String a_topic, int a_partition, List<String> a_seedBrokers, int a_port) throws Exception {
            // find the meta data about the topic and partition we are interested in
            //
            PartitionMetadata metadata = findLeader(a_seedBrokers, a_port, a_topic, a_partition);
            if (metadata == null) {
                System.out.println("Can't find metadata for Topic and Partition. Exiting");
                return;
            }
            if (metadata.leader() == null) {
                System.out.println("Can't find Leader for Topic and Partition. Exiting");
                return;
            }
            String leadBroker = metadata.leader().host();
            String clientName = "Client_" + a_topic + "_" + a_partition;
     
            SimpleConsumer consumer = new SimpleConsumer(leadBroker, a_port, 100000, 64 * 1024, clientName);
            long readOffset = getLastOffset(consumer,a_topic, a_partition, kafka.api.OffsetRequest.EarliestTime(), clientName);
     
            int numErrors = 0;
            while (a_maxReads > 0) {
                if (consumer == null) {
                    consumer = new SimpleConsumer(leadBroker, a_port, 100000, 64 * 1024, clientName);
                }
                FetchRequest req = new FetchRequestBuilder()
                        .clientId(clientName)
                        .addFetch(a_topic, a_partition, readOffset, 100000) // Note: this fetchSize of 100000 might need to be increased if large batches are written to Kafka
                        .build();
                FetchResponse fetchResponse = consumer.fetch(req);
     
                if (fetchResponse.hasError()) {
                    numErrors++;
                    // Something went wrong!
                    short code = fetchResponse.errorCode(a_topic, a_partition);
                    System.out.println("Error fetching data from the Broker:" + leadBroker + " Reason: " + code);
                    if (numErrors > 5) break;
                    if (code == ErrorMapping.OffsetOutOfRangeCode())  {
                        // We asked for an invalid offset. For simple case ask for the last element to reset
                        readOffset = getLastOffset(consumer,a_topic, a_partition, kafka.api.OffsetRequest.LatestTime(), clientName);
                        continue;
                    }
                    consumer.close();
                    consumer = null;
                    leadBroker = findNewLeader(leadBroker, a_topic, a_partition, a_port);
                    continue;
                }
                numErrors = 0;
     
                long numRead = 0;
                for (MessageAndOffset messageAndOffset : fetchResponse.messageSet(a_topic, a_partition)) {
                    long currentOffset = messageAndOffset.offset();
                    if (currentOffset < readOffset) {
                        System.out.println("Found an old offset: " + currentOffset + " Expecting: " + readOffset);
                        continue;
                    }
                    readOffset = messageAndOffset.nextOffset();
                    ByteBuffer payload = messageAndOffset.message().payload();
     
                    byte[] bytes = new byte[payload.limit()];
                    payload.get(bytes);
                    System.out.println(String.valueOf(messageAndOffset.offset()) + ": " + new String(bytes, "UTF-8"));
                    numRead++;
                    a_maxReads--;
                }
     
                if (numRead == 0) {
                    try {
                        Thread.sleep(1000);
                    } catch (InterruptedException ie) {
                    }
                }
            }
            if (consumer != null) consumer.close();
        }
     
        public static long getLastOffset(SimpleConsumer consumer, String topic, int partition,
                                         long whichTime, String clientName) {
            TopicAndPartition topicAndPartition = new TopicAndPartition(topic, partition);
            Map<TopicAndPartition, PartitionOffsetRequestInfo> requestInfo = new HashMap<TopicAndPartition, PartitionOffsetRequestInfo>();
            requestInfo.put(topicAndPartition, new PartitionOffsetRequestInfo(whichTime, 1));
            kafka.javaapi.OffsetRequest request = new kafka.javaapi.OffsetRequest(
                    requestInfo, kafka.api.OffsetRequest.CurrentVersion(), clientName);
            OffsetResponse response = consumer.getOffsetsBefore(request);
     
            if (response.hasError()) {
                System.out.println("Error fetching data Offset Data the Broker. Reason: " + response.errorCode(topic, partition) );
                return 0;
            }
            long[] offsets = response.offsets(topic, partition);
            return offsets[0];
        }
     
        private String findNewLeader(String a_oldLeader, String a_topic, int a_partition, int a_port) throws Exception {
            for (int i = 0; i < 3; i++) {
                boolean goToSleep = false;
                PartitionMetadata metadata = findLeader(m_replicaBrokers, a_port, a_topic, a_partition);
                if (metadata == null) {
                    goToSleep = true;
                } else if (metadata.leader() == null) {
                    goToSleep = true;
                } else if (a_oldLeader.equalsIgnoreCase(metadata.leader().host()) && i == 0) {
                    // first time through if the leader hasn't changed give ZooKeeper a second to recover
                    // second time, assume the broker did recover before failover, or it was a non-Broker issue
                    //
                    goToSleep = true;
                } else {
                    return metadata.leader().host();
                }
                if (goToSleep) {
                    try {
                        Thread.sleep(1000);
                    } catch (InterruptedException ie) {
                    }
                }
            }
            System.out.println("Unable to find new leader after Broker failure. Exiting");
            throw new Exception("Unable to find new leader after Broker failure. Exiting");
        }
     
        private PartitionMetadata findLeader(List<String> a_seedBrokers, int a_port, String a_topic, int a_partition) {
            PartitionMetadata returnMetaData = null;
            loop:
            for (String seed : a_seedBrokers) {
                SimpleConsumer consumer = null;
                try {
                    consumer = new SimpleConsumer(seed, a_port, 100000, 64 * 1024, "leaderLookup");
                    List<String> topics = Collections.singletonList(a_topic);
                    TopicMetadataRequest req = new TopicMetadataRequest(topics);
                    kafka.javaapi.TopicMetadataResponse resp = consumer.send(req);
     
                    List<TopicMetadata> metaData = resp.topicsMetadata();
                    for (TopicMetadata item : metaData) {
                        for (PartitionMetadata part : item.partitionsMetadata()) {
                            if (part.partitionId() == a_partition) {
                                returnMetaData = part;
                                break loop;
                            }
                        }
                    }
                } catch (Exception e) {
                    System.out.println("Error communicating with Broker [" + seed + "] to find Leader for [" + a_topic
                            + ", " + a_partition + "] Reason: " + e);
                } finally {
                    if (consumer != null) consumer.close();
                }
            }
            if (returnMetaData != null) {
                m_replicaBrokers.clear();
                for (kafka.cluster.Broker replica : returnMetaData.replicas()) {
                    m_replicaBrokers.add(replica.host());
                }
            }
            return returnMetaData;
        }
}

java客户端的参数调优

• fetchSize:从服务器获取的单包大小
• bufferSize:kafka客户端缓冲区大小
• group.id:分组消费分组名（用于实现复制消费，每个分组都能取得全量的数）