系统学习三步骤走:理解原理、搭建系统、Api练习。
从哪里找到Api?Document和git。
例如,Kafka在github上的地址github.com/apache/kafka,找到example目录。
这也算是一个小技巧/apache/xxx,就是XXX的git目录。
Kafka文档路径更好找,就在kafka.apache.org。
别用百度搜索,再跳转一次,记住xxx.apache.org就是apache项目的主目录。
如图,Kafka系统中包含三种角色,(1)producer生产者(2)Kafka Cluster消息队列(3)consumer消费者。
在上篇文章中,介绍了Kafka安装,通过启动Kafka server,实现了Kafka Cluster。而生产者消费者,可以通过Api实现写入和读取消息队列。
一、 pom.xml文件,引入依赖
Kafka Api 被包含在Kafka-clients包中,修改pom.xml文件。
<dependency>
<groupId>org.apache.kafka</groupId>
<artifactId>kafka-clients</artifactId>
<version>0.10.0.1</version>
</dependency>
二、编写Producer
1.Producer 配置
Properties props = new Properties();
props.put("bootstrap.servers", "hbase:9092,datanode2:9092,datanode3:9092");
props.put("acks", "all");
props.put("retries", 3);
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");
- bootstrap.servers:kafka server的地址
- acks:写入kafka时,leader负责一个该partion读写,当写入partition时,需要将记录同步到repli节点,all是全部同步节点都返回成功,leader才返回ack。
- retris:写入失败时,重试次数。当leader节点失效,一个repli节点会替代成为leader节点,此时可能出现写入失败,当retris为0时,produce不会重复。retirs重发,此时repli节点完全成为leader节点,不会产生消息丢失。
- batch.size:produce积累到一定数据,一次发送。
- buffer.memory: produce积累数据一次发送,缓存大小达到buffer.memory就发送数据。
- linger.ms :当设置了缓冲区,消息就不会即时发送,如果消息总不够条数、或者消息不够buffer大小就不发送了吗?当消息超过linger时间,也会发送。
- key/value serializer:序列化类。
2.KafkaProducer
- KafkaProducer
import org.apache.kafka.clients.producer.KafkaProducer;
Properties props = getConfig();
Producer<String, String> producer =
new KafkaProducer<String, String>(props);
- Producer是一个接口,声明了同步send和异步send两个重要方法。
public Future<RecordMetadata> send(ProducerRecord<K, V> record);
public Future<RecordMetadata> send(ProducerRecord<K, V> record, Callback callback);
- ProducerRecord 消息实体类,每条消息由(topic,key,value,timestamp)四元组封装。一条消息key可以为空和timestamp可以设置当前时间为默认值。
ProducerRecord record = new ProducerRecord<String, String>
("exam2", Integer.toString(i), Integer.toString(i));//exam2为topic
producer.send(record);
异步发送
long startTime = System.currentTimeMillis();
producer.send(new ProducerRecord<>(topic,messagekey,messageValue),
new DemoCallBack(startTime, messagekey, messageValue));
DemoCallBack异步回调接口,包含2个函数,构造函数和onCompletion函数。
返回的对象RecordMetadata包含partition和offset两个信息。
class DemoCallBack implements Callback {
private final long startTime;
private final String key;
private final String message;
public DemoCallBack(long startTime, String key, String message) {
this.startTime = startTime;
this.key = key;
this.message = message;
}
/**
* @param metadata The metadata for the record that was sent (i.e. the partition and offset). Null if an error
* occurred.
* @param exception The exception thrown during processing of this record. Null if no error occurred.
*/
public void onCompletion(RecordMetadata metadata, Exception exception) {
long elapsedTime = System.currentTimeMillis() - startTime;
if (metadata != null) {
System.out.println(
"message(" + key + ", " + message + ") sent to partition(" + metadata.partition() +
"), " +
"offset(" + metadata.offset() + ") in " + elapsedTime + " ms");
} else {
exception.printStackTrace();
}
}
}
控制台输出结果,能够看出回调函数不是异步执行的。
i:0
i:1
message(0, 0) sent to partition(6), offset(303) in 680 ms
i:2
message(1, 1) sent to partition(9), offset(295) in 126 ms
message(2, 2) sent to partition(8), offset(343) in 53 ms
i:3
message(3, 3) sent to partition(3), offset(331) in 18 ms
i:4
message(4, 4) sent to partition(3), offset(332) in 8 ms
i:5
message(5, 5) sent to partition(0), offset(310) in 22 ms
i:6
message(6, 6) sent to partition(8), offset(344) in 8 ms
i:7
message(7, 7) sent to partition(9), offset(296) in 19 ms
i:8
i:9
message(9, 9) sent to partition(3), offset(333) in 23 ms
message(8, 8) sent to partition(7), offset(287) in 136 ms
i:10
message(10, 10) sent to partition(6), offset(304) in 21 ms
三、编写Consumer
1.Consumer 配置
Properties props = new Properties();
props.put("bootstrap.servers", "hbase:9092,datanode2:9092,datanode3:9092");
props.put("group.id", "testGroup");
props.put("enable.auto.commit", "true");
props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
- group.id:testGroup。由于在kafka中,同一组中的consumer不会读取到同一个消息,依靠groud.id设置组名。
- enable.auto.commit:true。设置自动提交offset。
2.KafkaConsumer
KafkaConsumer
import org.apache.kafka.clients.consumer.KafkaConsumer;
Properties props = getConfig();
consumer = new KafkaConsumer<String, String>(props);
Consumer接口,声明了subscribe和poll两个重要方法。KafkaConsumer实现了Consumer接口。
public void subscribe(Collection<String> topics);
public ConsumerRecords<K, V> poll(long timeout);
可以创建多个consumer线程,并发拉取消息。由于consumer是线程不安全的,合适的做法是每个线程创建并维护一个consumer对象。
自定义KafkaConsumerRunner是一个多线程类,维护一个KafkaConsumer对象。
// Thread to consume kafka data
public static class KafkaConsumerRunner implements Runnable
{
private final AtomicBoolean closed = new AtomicBoolean(false);
private final KafkaConsumer<String, String> consumer;
private final String topic;
public KafkaConsumerRunner(String topic)
{
Properties props = getConfig();
consumer = new KafkaConsumer<String, String>(props);
this.topic = topic;
}
public void handleRecord(ConsumerRecord record)
{
System.out.println("name: " + Thread.currentThread().getName()
+ " ; topic: " + record.topic() + "; partition:"+record.partition()+
" ; offset" + record.offset() + " ; key: " + record.key() + " ; value: " + record.value());
}
public void run()
{
try {
// subscribe
consumer.subscribe(Arrays.asList(topic));
while (!closed.get()) {
//read data
ConsumerRecords<String, String> records = consumer.poll(1000);
// Handle new records
for (ConsumerRecord<String, String> record : records) {
handleRecord(record);
}
}
}
catch (WakeupException e) {
// Ignore exception if closing
if (!closed.get()) {
throw e;
}
}
finally {
consumer.close();
}
}
// Shutdown hook which can be called from a separate thread
public void shutdown()
{
closed.set(true);
consumer.wakeup();
}
}
线程池启动多个consumer线程,
int numConsumers = 3;
final String topic = "exam2";
final ExecutorService executor = Executors.newFixedThreadPool(numConsumers);
final List<KafkaConsumerRunner> consumers = new ArrayList<KafkaConsumerRunner>();
for (int i = 0; i < numConsumers; i++) {
KafkaConsumerRunner consumer = new KafkaConsumerRunner(topic);
consumers.add(consumer);
executor.submit(consumer);
}
执行结果:
name: pool-1-thread-3 ; topic: exam2; partition:9 ; offset445 ; key: 1 ; value: 1
name: pool-1-thread-2 ; topic: exam2; partition:6 ; offset448 ; key: 0 ; value: 0
name: pool-1-thread-3 ; topic: exam2; partition:8 ; offset508 ; key: 2 ; value: 2
name: pool-1-thread-1 ; topic: exam2; partition:3 ; offset495 ; key: 3 ; value: 3
name: pool-1-thread-1 ; topic: exam2; partition:3 ; offset496 ; key: 4 ; value: 4
name: pool-1-thread-1 ; topic: exam2; partition:0 ; offset461 ; key: 5 ; value: 5
name: pool-1-thread-3 ; topic: exam2; partition:8 ; offset509 ; key: 6 ; value: 6
name: pool-1-thread-3 ; topic: exam2; partition:9 ; offset446 ; key: 7 ; value: 7
name: pool-1-thread-3 ; topic: exam2; partition:7 ; offset428 ; key: 8 ; value: 8
name: pool-1-thread-1 ; topic: exam2; partition:3 ; offset497 ; key: 9 ; value: 9
name: pool-1-thread-2 ; topic: exam2; partition:6 ; offset449 ; key: 10 ; value: 10
name: pool-1-thread-3 ; topic: exam2; partition:8 ; offset510 ; key: 11 ; value: 11
观察结果
- 保证每个consumer线程消费不同的partition。
- partition之间不能保证顺序进行,里如key:1和key:0
- 同一个partition内保证顺序性,即offset保证在同一partition内顺序进行。
优雅的关闭子线程
在main函数中,添加hook进程关闭的函数。new Thread 在进程关闭时触发,调用Consumer的shutdown函数,设置while循环的退出条件while (!closed.get())
。
Runtime.getRuntime().addShutdownHook(new Thread()
{
@Override
public void run()
{
for (KafkaConsumerRunner consumer : consumers) {
consumer.shutdown();
}
executor.shutdown();
try {
executor.awaitTermination(5000, TimeUnit.MILLISECONDS);
}
catch (InterruptedException e) {
e.printStackTrace();
}
System.out.println("process quit");
}
});
手动控制offset
//设置由用户触发提交offset
props.put("enable.auto.commit", "false");
for (ConsumerRecord<String, String> record : records) {
handleRecord(record);
}
consumer.commitAsync();
运行结果:
- poll拉取的数据还是顺序返回,不会反复拉取offset的数据。
- 重启进程,由于offset没有提交,会重头处理offset。
四、总结
本文测试了kafka提供的Api。
在实际应用中kafka会和spark stream结合,采用流式计算的方式处理kafka中数据。