kafka API

Producer API:

消息发送流程:

KafkaProducer发送消息采用的是异步发送的方式。在消息发送的过程中,涉及到了两个线程——main线程和Sender线程

以及一个线程共享变量——RecordAccumulatormain线程将消息发送给RecordAccumulatorSender线程不断从RecordAccumulator中拉取消息发送到Kafka broker

 

 

 

相关参数:

batch.size只有数据积累到batch.size之后,sender才会发送数据。

linger.ms如果数据迟迟未达到batch.sizesender等待linger.time之后就会发送数据。

 

异步发送API:

1)导入依赖

<dependency>

<groupId>org.apache.kafka</groupId>

<artifactId>kafka-clients</artifactId>

<version>0.11.0.0</version>

</dependency>

2)编写代码

需要用到的类:

KafkaProducer:需要创建一个生产者对象,用来发送数据

ProducerConfig:获取所需的一系列配置参数

ProducerRecord:每条数据都要封装成一个ProducerRecord对象

1.不带回调函数的API

package com.bigdata.kafka;

 

import org.apache.kafka.clients.producer.*;

 

import java.util.Properties;

import java.util.concurrent.ExecutionException;

 

public class CustomProducer {

 

    public static void main(String[] args) throws ExecutionException, InterruptedException {

        Properties props = new Properties();

 

        //kafka集群,broker-list

        props.put("bootstrap.servers", "hadoop102:9092");

 

        props.put("acks", "all");

 

        //重试次数

        props.put("retries", 1); 

 

        //批次大小

        props.put("batch.size", 16384); 

 

        //等待时间

        props.put("linger.ms", 1); 

 

        //RecordAccumulator缓冲区大小

        props.put("buffer.memory", 33554432);

 

        props.put("key.serializer", "org.apache.kafka.common.serialization.StringSerializer");

        props.put("value.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

        Producer<String, String> producer = new KafkaProducer<>(props);

 

        for (int i = 0; i < 100; i++) {

            producer.send(new ProducerRecord<String, String>("first", Integer.toString(i), Integer.toString(i)));

        }

 

        producer.close();

    }

}

2.带回调函数的API

回调函数会在producer收到ack时调用,为异步调用,该方法有两个参数,分别是RecordMetadataException,如果Exceptionnull,说明消息发送成功,如果Exception不为null,说明消息发送失败。

注意:消息发送失败会自动重试,不需要我们在回调函数中手动重试。

package com.bigdata.kafka;

 

 

 

import org.apache.kafka.clients.producer.*;

 

 

 

import java.util.Properties;

 

import java.util.concurrent.ExecutionException;

 

 

 

public class CustomProducer {

 

 

 

public static void main(String[] args) throws ExecutionException, InterruptedException {

 

 

 

        Properties props = new Properties();

 

 

 

        props.put("bootstrap.servers", "hadoop102:9092");//kafka集群,broker-list

 

 

 

        props.put("acks", "all");

 

 

 

        props.put("retries", 1);//重试次数

 

 

 

        props.put("batch.size", 16384);//批次大小

 

 

 

        props.put("linger.ms", 1);//等待时间

 

 

 

        props.put("buffer.memory", 33554432);//RecordAccumulator缓冲区大小

 

 

 

        props.put("key.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

        props.put("value.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

 

 

        Producer<String, String> producer = new KafkaProducer<>(props);

 

 

 

        for (int i = 0; i < 100; i++) {

 

            producer.send(new ProducerRecord<String, String>("first", Integer.toString(i), Integer.toString(i)), new Callback() {

 

 

 

                //回调函数,该方法会在Producer收到ack时调用,为异步调用

 

                @Override

 

                public void onCompletion(RecordMetadata metadata, Exception exception) {

 

                    if (exception == null) {

 

                        System.out.println("success->" + metadata.offset());

 

                    } else {

 

                        exception.printStackTrace();

 

                    }

 

                }

 

            });

 

        }

 

        producer.close();

 

    }

 

}

 

 

同步发送API:

同步发送的意思就是,一条消息发送之后,会阻塞当前线程,直至返回ack

由于send方法返回的是一个Future对象,根据Futrue对象的特点,我们也可以实现同步发送的效果,只需在调用Future对象的get方发即可。

 

package com.bigdata.kafka;

 

 

 

import org.apache.kafka.clients.producer.KafkaProducer;

 

import org.apache.kafka.clients.producer.Producer;

 

import org.apache.kafka.clients.producer.ProducerRecord;

 

 

 

import java.util.Properties;

 

import java.util.concurrent.ExecutionException;

 

 

 

public class CustomProducer {

 

 

 

public static void main(String[] args) throws ExecutionException, InterruptedException {

 

 

 

        Properties props = new Properties();

 

 

 

        props.put("bootstrap.servers", "hadoop102:9092");//kafka集群,broker-list

 

 

 

        props.put("acks", "all");

 

 

 

        props.put("retries", 1);//重试次数

 

 

 

        props.put("batch.size", 16384);//批次大小

 

 

 

        props.put("linger.ms", 1);//等待时间

 

 

 

        props.put("buffer.memory", 33554432);//RecordAccumulator缓冲区大小

 

 

 

        props.put("key.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

        props.put("value.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

 

 

        Producer<String, String> producer = new KafkaProducer<>(props);

 

        for (int i = 0; i < 100; i++) {

 

            producer.send(new ProducerRecord<String, String>("first", Integer.toString(i), Integer.toString(i))).get();

 

        }

 

        producer.close();

 

    }

 

}

 

Consumer API:

 

Consumer消费数据时的可靠性是很容易保证的,因为数据在Kafka中是持久化的,故不用担心数据丢失问题。

 

由于consumer在消费过程中可能会出现断电宕机等故障,consumer恢复后,需要从故障前的位置的继续消费,所以consumer需要实时记录自己消费到了哪个offset,以便故障恢复后继续消费。

 

所以offset的维护是Consumer消费数据是必须考虑的问题。

自动提交offset

1)导入依赖

<dependency>

<groupId>org.apache.kafka</groupId>

<artifactId>kafka-clients</artifactId>

<version>0.11.0.0</version>

</dependency>

2)编写代码

需要用到的类:

KafkaConsumer:需要创建一个消费者对象,用来消费数据

ConsumerConfig:获取所需的一系列配置参数

ConsuemrRecord:每条数据都要封装成一个ConsumerRecord对象

为了使我们能够专注于自己的业务逻辑,Kafka提供了自动提交offset的功能。 

自动提交offset的相关参数:

enable.auto.commit是否开启自动提交offset功能

auto.commit.interval.ms自动提交offset的时间间隔

 

package com.bigdata.kafka;

 

 

 

import org.apache.kafka.clients.consumer.ConsumerRecord;

 

import org.apache.kafka.clients.consumer.ConsumerRecords;

 

import org.apache.kafka.clients.consumer.KafkaConsumer;

 

 

 

import java.util.Arrays;

 

import java.util.Properties;

 

 

 

public class CustomConsumer {

 

 

 

public static void main(String[] args) {

 

 

 

        Properties props = new Properties();

 

 

 

        props.put("bootstrap.servers", "hadoop102:9092");

 

 

 

        props.put("group.id", "test");

 

 

 

        props.put("enable.auto.commit", "true");

 

 

 

        props.put("auto.commit.interval.ms", "1000");

 

 

 

        props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

        props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

 

 

        KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);

 

 

 

        consumer.subscribe(Arrays.asList("first"));

 

 

 

        while (true) {

 

 

 

            ConsumerRecords<String, String> records = consumer.poll(100);

 

 

 

            for (ConsumerRecord<String, String> record : records)

 

 

 

                System.out.printf("offset = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());

 

        }

 

    }

 

}

 

手动提交offset

 

虽然自动提交offset十分简介便利,但由于其是基于时间提交的,开发人员难以把握offset提交的时机。因此Kafka还提供了手动提交offsetAPI

 

手动提交offset的方法有两种:分别是commitSync(同步提交)commitAsync(异步提交)。两者的相同点是,都会将本次poll的一批数据最高的偏移量提交;不同点是,commitSync阻塞当前线程,一直到提交成功,并且会自动失败重试(由不可控因素导致,也会出现提交失败);而commitAsync则没有失败重试机制,故有可能提交失败。

 

同步提交offset

由于同步提交offset有失败重试机制,故更加可靠,以下为同步提交offset的示例。

package com.bigdata.kafka.consumer;

 

import org.apache.kafka.clients.consumer.ConsumerRecord;

import org.apache.kafka.clients.consumer.ConsumerRecords;

import org.apache.kafka.clients.consumer.KafkaConsumer;

 

import java.util.Arrays;

import java.util.Properties;

 

public class CustomComsumer {

 

    public static void main(String[] args) {

 

        Properties props = new Properties();

 

//Kafka集群

        props.put("bootstrap.servers", "hadoop102:9092"); 

 

//消费者组,只要group.id相同,就属于同一个消费者组

        props.put("group.id", "test"); 

 

        props.put("enable.auto.commit", "false");//关闭自动提交offset

 

        props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

        props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

        KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);

 

        consumer.subscribe(Arrays.asList("first"));//消费者订阅主题

 

        while (true) {

 

//消费者拉取数据

            ConsumerRecords<String, String> records = consumer.poll(100); 

 

            for (ConsumerRecord<String, String> record : records) {

 

                System.out.printf("offset = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());

 

            }

 

//同步提交,当前线程会阻塞直到offset提交成功

            consumer.commitSync();

        }

    }

}

异步提交offset

虽然同步提交offset更可靠一些,但是由于其会阻塞当前线程,直到提交成功。因此吞吐量会收到很大的影响。因此更多的情况下,会选用异步提交offset的方式。

以下为异步提交offset的示例:

 

package com.bigdata.kafka.consumer;

 

 

 

import org.apache.kafka.clients.consumer.*;

 

import org.apache.kafka.common.TopicPartition;

 

 

 

import java.util.Arrays;

 

import java.util.Map;

 

import java.util.Properties;

 

 

 

 

 

public class CustomConsumer {

 

 

 

    public static void main(String[] args) {

 

 

 

        Properties props = new Properties();

 

 

 

        //Kafka集群

 

        props.put("bootstrap.servers", "hadoop102:9092"); 

 

 

 

        //消费者组,只要group.id相同,就属于同一个消费者组

 

        props.put("group.id", "test"); 

 

 

 

        //关闭自动提交offset

 

        props.put("enable.auto.commit", "false");

 

 

 

        props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

        props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

 

 

        KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);

 

        consumer.subscribe(Arrays.asList("first"));//消费者订阅主题

 

 

 

        while (true) {

 

            ConsumerRecords<String, String> records = consumer.poll(100);//消费者拉取数据

 

            for (ConsumerRecord<String, String> record : records) {

 

                System.out.printf("offset = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());

 

            }

 

 

 

//异步提交

 

            consumer.commitAsync(new OffsetCommitCallback() {

 

                @Override

 

                public void onComplete(Map<TopicPartition, OffsetAndMetadata> offsets, Exception exception) {

 

                    if (exception != null) {

 

                        System.err.println("Commit failed for" + offsets);

 

                    }

 

                }

 

            }); 

 

        }

 

    }

 

}

数据漏消费和重复消费分析

无论是同步提交还是异步提交offset,都有可能会造成数据的漏消费或者重复消费。先提交offset后消费,有可能造成数据的漏消费;而先消费后提交offset,有可能会造成数据的重复消费。

 

 

自定义存储offset

Kafka 0.9版本之前,offset存储在zookeeper0.9版本之后,默认将offset存储在Kafka的一个内置的topic中。除此之外,Kafka还可以选择自定义存储offset

offset的维护是相当繁琐的,因为需要考虑到消费者的Rebalace

当有新的消费者加入消费者组、已有的消费者推出消费者组或者所订阅的主题的分区发生变化,就会触发到分区的重新分配,重新分配的过程叫做Rebalance

消费者发生Rebalance之后,每个消费者消费的分区就会发生变化。因此消费者要首先获取到自己被重新分配到的分区,并且定位到每个分区最近提交的offset位置继续消费。

要实现自定义存储offset,需要借助ConsumerRebalanceListener,以下为示例代码,其中提交和获取offset的方法,需要根据所选的offset存储系统自行实现。

 

package com.bigdata.kafka.consumer;

 

 

 

import org.apache.kafka.clients.consumer.*;

 

import org.apache.kafka.common.TopicPartition;

 

 

 

import java.util.*;

 

 

 

public class CustomConsumer {

 

 

 

    private static Map<TopicPartition, Long> currentOffset = new HashMap<>();

 

 

 

public static void main(String[] args) {

 

 

 

//创建配置信息

 

        Properties props = new Properties();

 

 

 

//Kafka集群

 

        props.put("bootstrap.servers", "hadoop102:9092"); 

 

 

 

//消费者组,只要group.id相同,就属于同一个消费者组

 

        props.put("group.id", "test"); 

 

 

 

//关闭自动提交offset

 

        props.put("enable.auto.commit", "false");

 

 

 

        //KeyValue的反序列化类

 

        props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

        props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");

 

 

 

        //创建一个消费者

 

        KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);

 

 

 

        //消费者订阅主题

 

        consumer.subscribe(Arrays.asList("first"), new ConsumerRebalanceListener() {

 

            

 

            //该方法会在Rebalance之前调用

 

            @Override

 

            public void onPartitionsRevoked(Collection<TopicPartition> partitions) {

 

                commitOffset(currentOffset);

 

            }

 

 

 

            //该方法会在Rebalance之后调用

 

            @Override

 

            public void onPartitionsAssigned(Collection<TopicPartition> partitions) {

 

                currentOffset.clear();

 

                for (TopicPartition partition : partitions) {

 

                    consumer.seek(partition, getOffset(partition));//定位到最近提交的offset位置继续消费

 

                }

 

            }

 

        });

 

 

 

        while (true) {

 

            ConsumerRecords<String, String> records = consumer.poll(100);//消费者拉取数据

 

            for (ConsumerRecord<String, String> record : records) {

 

                System.out.printf("offset = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());

 

                currentOffset.put(new TopicPartition(record.topic(), record.partition()), record.offset());

 

            }

 

            commitOffset(currentOffset);//异步提交

 

        }

 

    }

 

 

 

    //获取某分区的最新offset

 

    private static long getOffset(TopicPartition partition) {

 

        return 0;

 

    }

 

 

 

    //提交该消费者所有分区的offset

 

    private static void commitOffset(Map<TopicPartition, Long> currentOffset) {

 

 

 

    }

 

}

 

 

 

 

自定义Interceptor:

拦截器原理

Producer拦截器(interceptor)是在Kafka 0.10版本被引入的,主要用于实现clients端的定制化控制逻辑。

对于producer而言,interceptor使得用户在消息发送前以及producer回调逻辑前有机会对消息做一些定制化需求,比如修改消息等。同时,producer允许用户指定多个interceptor按序作用于同一条消息从而形成一个拦截链(interceptor chain)Intercetpor的实现接口是org.apache.kafka.clients.producer.ProducerInterceptor,其定义的方法包括:

1configure(configs)

获取配置信息初始化数据时调用

2onSend(ProducerRecord)

该方法封装进KafkaProducer.send方法中,即它运行在用户主线程中。Producer确保在消息被序列化以及计算分区前调用该方法。用户可以在该方法中对消息做任何操作,但最好保证不要修改消息所属的topic和分区,否则会影响目标分区的计算。

3onAcknowledgement(RecordMetadata, Exception)

该方法会在消息从RecordAccumulator成功发送到Kafka Broker之后,或者在发送过程中失败时调用。并且通常都是在producer回调逻辑触发之前。onAcknowledgement运行在producerIO线程中,因此不要在该方法中放入很重的逻辑,否则会拖慢producer的消息发送效率。

4close

关闭interceptor,主要用于执行一些资源清理工作

如前所述,interceptor可能被运行在多个线程中,因此在具体实现时用户需要自行确保线程安全。另外倘若指定了多个interceptor,则producer将按照指定顺序调用它们,并仅仅是捕获每个interceptor可能抛出的异常记录到错误日志中而非在向上传递。这在使用过程中要特别留意。

拦截器案例

1需求:

 

实现一个简单的双interceptor组成的拦截链。第一个interceptor会在消息发送前将时间戳信息加到消息value的最前部;第二个interceptor会在消息发送后更新成功发送消息数或失败发送消息数。

 

 

案例实操

 

1)增加时间戳拦截器

 

package com.bigdata.kafka.interceptor;

 

import java.util.Map;

 

import org.apache.kafka.clients.producer.ProducerInterceptor;

 

import org.apache.kafka.clients.producer.ProducerRecord;

 

import org.apache.kafka.clients.producer.RecordMetadata;

 

 

 

public class TimeInterceptor implements ProducerInterceptor<String, String> {

 

 

 

@Override

 

public void configure(Map<String, ?> configs) {

 

 

 

}

 

 

 

@Override

 

public ProducerRecord<String, String> onSend(ProducerRecord<String, String> record) {

 

 

 

// 创建一个新的record,把时间戳写入消息体的最前部

 

return new ProducerRecord(record.topic(), record.partition(), record.timestamp(), record.key(),

 

System.currentTimeMillis() + "," + record.value().toString());

 

}

 

 

 

@Override

 

public void onAcknowledgement(RecordMetadata metadata, Exception exception) {

 

 

 

}

 

 

 

@Override

 

public void close() {

 

 

 

}

 

}

2)统计发送消息成功和发送失败消息数,并在producer关闭时打印这两个计数器

 

package com.bigdata.kafka.interceptor;

 

import java.util.Map;

 

import org.apache.kafka.clients.producer.ProducerInterceptor;

 

import org.apache.kafka.clients.producer.ProducerRecord;

 

import org.apache.kafka.clients.producer.RecordMetadata;

 

 

 

public class CounterInterceptor implements ProducerInterceptor<String, String>{

 

    private int errorCounter = 0;

 

    private int successCounter = 0;

 

 

 

@Override

 

public void configure(Map<String, ?> configs) {

 

 

 

}

 

 

 

@Override

 

public ProducerRecord<String, String> onSend(ProducerRecord<String, String> record) {

 

 return record;

 

}

 

 

 

@Override

 

public void onAcknowledgement(RecordMetadata metadata, Exception exception) {

 

// 统计成功和失败的次数

 

        if (exception == null) {

 

            successCounter++;

 

        } else {

 

            errorCounter++;

 

        }

 

}

 

 

 

@Override

 

public void close() {

 

        // 保存结果

 

        System.out.println("Successful sent: " + successCounter);

 

        System.out.println("Failed sent: " + errorCounter);

 

}

 

}

3producer主程序

 

package com.bigdata.kafka.interceptor;

 

import java.util.ArrayList;

 

import java.util.List;

 

import java.util.Properties;

 

import org.apache.kafka.clients.producer.KafkaProducer;

 

import org.apache.kafka.clients.producer.Producer;

 

import org.apache.kafka.clients.producer.ProducerConfig;

 

import org.apache.kafka.clients.producer.ProducerRecord;

 

 

 

public class InterceptorProducer {

 

 

 

public static void main(String[] args) throws Exception {

 

// 1 设置配置信息

 

Properties props = new Properties();

 

props.put("bootstrap.servers", "hadoop102:9092");

 

props.put("acks", "all");

 

props.put("retries", 0);

 

props.put("batch.size", 16384);

 

props.put("linger.ms", 1);

 

props.put("buffer.memory", 33554432);

 

props.put("key.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

props.put("value.serializer", "org.apache.kafka.common.serialization.StringSerializer");

 

 

 

// 2 构建拦截链

 

List<String> interceptors = new ArrayList<>();

 

interceptors.add("com.bigdata.kafka.interceptor.TimeInterceptor"); interceptors.add("com.bigdata.kafka.interceptor.CounterInterceptor");

 

props.put(ProducerConfig.INTERCEPTOR_CLASSES_CONFIG, interceptors);

 

 

 

String topic = "first";

 

Producer<String, String> producer = new KafkaProducer<>(props);

 

 

 

// 3 发送消息

 

for (int i = 0; i < 10; i++) {

 

 

 

    ProducerRecord<String, String> record = new ProducerRecord<>(topic, "message" + i);

 

    producer.send(record);

 

}

 

 

 

// 4 一定要关闭producer,这样才会调用interceptorclose方法

 

producer.close();

 

}

 

}

3)测试

1)在kafka上启动消费者然后运行客户端java程序。

[hadoop@hadoop102 kafka]$ bin/kafka-console-consumer.sh \

--bootstrap-server hadoop102:9092 --from-beginning --topic first

 

1501904047034,message0

1501904047225,message1

1501904047230,message2

1501904047234,message3

1501904047236,message4

1501904047240,message5

1501904047243,message6

1501904047246,message7

1501904047249,message8

1501904047252,message9

 

posted @ 2020-12-29 10:32  芒果不氓  阅读(98)  评论(0编辑  收藏  举报