在上一章《Docker下的Kafka学习之二:搭建集群环境》中我们学会了搭建kafka集群环境,今天我们来实战集群环境下的用java发送和消费kafka的消息;
本次实战我们要搭建的环境略有一些复杂,整体环境如下图:
如上图所示,从浏览器发起一个请求会经历以下历程: 1. 请求到nginx后,由nginx转发到tomcat,nginx后面接了两个tomcat,容器名分别是producer1和producer2,部署的都是kafkaclusterproducerdemo这个应用的war包; 2. producer1和producer2收到消息后,向broker投递消息; 3. consumer1、consumer2、consumer3是三个tomcat,上面都部署了kafkaclusterconsumerdemo这个应用,连接了kafka的partition,收到消息后消费这些消息,这三个consumer属于同一个group,共同消息主题”test002”的消息;
整个环境的ip和功能说明列表如下:
容器name | 镜像 | IP | 功能 | 端口映射 | 端口功能 |
---|---|---|---|---|---|
broker1 | bolingcavalry/ssh-kafka292081-zk346:0.0.1 | 172.18.0.2 | 一号broker | 19011:22 | ssh |
broker2 | bolingcavalry/ssh-kafka292081-zk346:0.0.1 | 172.18.0.3 | 二号broker | 19012:22 | ssh |
broker3 | bolingcavalry/ssh-kafka292081-zk346:0.0.1 | 172.18.0.4 | 三号broker | 19013:22 | ssh |
producer1 | bolingcavalry/online_deploy_tomcat:0.0.1 | 172.18.0.5 | 一号消息制造者 | 19014:8080 | 页面访问和在线web部署 |
producer2 | bolingcavalry/online_deploy_tomcat:0.0.1 | 172.18.0.6 | 二号消息制造者 | 19015:8080 | 页面访问和在线web部署 |
nginx | daocloud.io/library/nginx:latest | 172.18.0.7 | 入口 | 19016:80 | 页面访问 |
consumer1 | bolingcavalry/online_deploy_tomcat:0.0.1 | 172.18.0.8 | 一号消息消费者 | 19017:8080 | 页面访问和在线web部署 |
consumer2 | bolingcavalry/online_deploy_tomcat:0.0.1 | 172.18.0.9 | 二号消息消费者 | 19018:8080 | 页面访问和在线web部署 |
consumer3 | bolingcavalry/online_deploy_tomcat:0.0.1 | 172.18.0.10 | 三号消息消费者 | 19019:8080 | 页面访问和在线web部署 |
这么多容器,还是用docker-compose来配置和管理比较方便,配置好的docker-compose.yml文件如下所示:
version: '2'
services:
broker1:
image: bolingcavalry/ssh-kafka292081-zk346:0.0.1
container_name: broker1
ports:
- "19011:22"
restart: always
broker2:
image: bolingcavalry/ssh-kafka292081-zk346:0.0.1
container_name: broker2
depends_on:
- broker1
ports:
- "19012:22"
restart: always
broker3:
image: bolingcavalry/ssh-kafka292081-zk346:0.0.1
container_name: broker3
depends_on:
- broker2
ports:
- "19013:22"
restart: always
producer1:
image: bolingcavalry/online_deploy_tomcat:0.0.1
container_name: producer1
depends_on:
- broker3
links:
- broker1:hostb1
- broker2:hostb2
- broker3:hostb3
ports:
- "19014:8080"
environment:
TOMCAT_SERVER_ID: producer1
restart: always
producer2:
image: bolingcavalry/online_deploy_tomcat:0.0.1
container_name: producer2
depends_on:
- producer1
links:
- broker1:hostb1
- broker2:hostb2
- broker3:hostb3
ports:
- "19015:8080"
environment:
TOMCAT_SERVER_ID: producer2
restart: always
nginx:
image: daocloud.io/library/nginx:latest
container_name: nginx
depends_on:
- producer2
links:
- producer1:t01
- producer2:t02
ports:
- "19016:80"
restart: always
consumer1:
image: bolingcavalry/online_deploy_tomcat:0.0.1
container_name: consumer1
depends_on:
- nginx
links:
- broker1:hostb1
- broker2:hostb2
- broker3:hostb3
ports:
- "19017:8080"
environment:
TOMCAT_SERVER_ID: consumer1
restart: always
consumer2:
image: bolingcavalry/online_deploy_tomcat:0.0.1
container_name: consumer2
depends_on:
- consumer1
ports:
- "19018:8080"
links:
- broker1:hostb1
- broker2:hostb2
- broker3:hostb3
environment:
TOMCAT_SERVER_ID: consumer2
restart: always
consumer3:
image: bolingcavalry/online_deploy_tomcat:0.0.1
container_name: consumer3
depends_on:
- consumer2
ports:
- "19019:8080"
links:
- broker1:hostb1
- broker2:hostb2
- broker3:hostb3
environment:
TOMCAT_SERVER_ID: consumer3
restart: always
如上所示,broker1、broker2、broker3是用来搭建kafka集群环境的,作为生产消息的tomcat有两个,消费消息的tomcat有三个,再加上一个nginx,就是全部的容器了;
在docker-compose.yml所在目录下执行以下命令即可启动所有容器:
docker-compose up -d
启动的过程下图所示:
详细的搭建步骤,请看《Docker下的Kafka学习之二:搭建集群环境》,所有操作都在broker1、broker2、broker3这三个容器上进行;
在容器broker1上执行以下命令,就会在kafka上创建一个主题,有6个patition:
/usr/local/work/kafka_2.9.2-0.8.1/bin/kafka-topics.sh --create --zookeeper broker1:2181,broker2:2181,broker3:2181 --replication-factor 1 --partitions 6 --topic test002
查看broker1的/tmp/kafka-logs/目录,就能看到partition目录,如下图:
如上图,broker1负责的是partition2和partition5; 去broker2、borker3检查后,得到每个broker和partition的关系如下:
容器 | partition |
---|---|
broker1 | partition2、partition5 |
broker2 | partition0、partition3 |
broker3 | partition1、partition4 |
遗憾的是nginx容器里面没有安装vim工具,需要先执行apt-get update,再执行apt-get install -y vim,将vim装好,再去修改/etc/nginx/nginx.conf文件,修改后的内容如下:
user nginx;
worker_processes 1;
error_log /var/log/nginx/error.log warn;
pid /var/run/nginx.pid;
events {
worker_connections 1024;
}
http {
include /etc/nginx/mime.types;
default_type application/octet-stream;
log_format main '$remote_addr - $remote_user [$time_local] "$request" '
'$status $body_bytes_sent "$http_referer" '
'"$http_user_agent" "$http_x_forwarded_for"';
access_log /var/log/nginx/access.log main;
sendfile on;
#tcp_nopush on;
keepalive_timeout 65;
#gzip on;
#include /etc/nginx/conf.d/*.conf;
upstream tomcat_client {
server t01:8080 weight=1;
server t02:8080 weight=1;
}
server {
server_name "";
listen 80 default_server;
listen [::]:80 default_server ipv6only=on;
location / {
proxy_pass http://tomcat_client;
proxy_redirect default;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
}
}
}
和原来的内容相比,主要的修改点就是注释掉“include /etc/nginx/conf.d/*.conf”,再增加upstream和server的配置; 修改完成后,执行/usr/sbin/nginx -s reload命令让nginx加载修改后的配置; 在当前电脑的浏览器输入“http://localhost:19016/examples/servlets/servlet/HelloWorldExample”试试,可以看到请求已经被nginx转发到tomcat上去了,如下图:
本次实战要开发两个java的web应用,然后将构建的war在线部署到tomcat上去,关于在线部署的详情请参照《实战docker,编写Dockerfile定制tomcat镜像,实现web应用在线部署》,本次开发的两个java应用的pom.xml中已经配置好了在线部署的插件和参数,读者们只需要将本地maven配置好部署所需的用户名和密码即可;
本次开发的两个java工程的源码都可以在github下载,地址是:git@github.com:zq2599/blog_demos.git,这里面有多个工程,本次用到的两个工程如下图红框所示:
接下来开始编码了;
<!--fastjson-->
<dependency>
<groupId>com.alibaba</groupId>
<artifactId>fastjson</artifactId>
<version>1.2.39</version>
</dependency>
<dependency>
<groupId>org.apache.commons</groupId>
<artifactId>commons-lang3</artifactId>
<version>3.5</version>
</dependency>
<!--kafka-->
<dependency>
<groupId>org.apache.kafka</groupId>
<artifactId>kafka_2.9.2</artifactId>
<version>0.8.1</version>
</dependency>
如上所示,除了kafka的依赖,fastjson和commons-lang3也会用到; 3. 自定义partition规则,创建一个实现了Partitioner接口的类BusinessPartition,可以通过key的值来决定将消息投递到那个partition,这里的做法是直接用key的值来代表partition,源码如下:
public class BusinessPartition implements Partitioner {
/**
* 构造函数的函数体没有东西,但是不能没有构造函数
* @param props
*/
public BusinessPartition(VerifiableProperties props) {
super();
}
public int partition(Object o, int i) {
int partitionValue = 0;
if(o instanceof String && StringUtils.isNoneBlank((String)o)){
partitionValue = Integer.valueOf((String)o);
}
return partitionValue;
}
}
注意:带VerifiableProperties的构造方法一定要写! 4. 消息服务初始化: 初始化工作是放在一个Bean的init方法中进行的,如下:
@PostConstruct
public void init(){
try {
Properties props = new Properties();
props.put("serializer.class", "kafka.serializer.StringEncoder");
props.put("zk.connect", "hostb1:2181,hostb1:2181,hostb1:2181");
props.put("metadata.broker.list", "hostb1:9092,hostb1:9092,hostb1:9092");
props.put("partitioner.class","com.bolingcavalry.service.BusinessPartition");
producer = new kafka.javaapi.producer.Producer<String, String>(new ProducerConfig(props));
} catch (Exception e) {
e.printStackTrace();
}
}
hostb1、hostb2、hostb3都是docker-compose.yml配置的link参数; 值得注意的是”partitioner.class”这个参数的值,是我们刚刚创建的BusinessPartition这个类,这样kafka就知道用哪个自定义类来处理partition的逻辑了; 5. 发送消息: 发送消息的方法有两个,第二个接受外部传来的key,用来确定当前消息发往哪个partition:
public void sendSimpleMsg(String topic, String message) {
//producer的内部实现中,已经考虑了线程安全,所以此处不用加锁了
producer.send(new KeyedMessage<String, String>(topic, message));
}
public void sendKeyMsg(String topic, String key, String message) {
//producer的内部实现中,已经考虑了线程安全,所以此处不用加锁了
producer.send(new KeyedMessage<String, String>(topic, key, message));
}
@RequestMapping("/keymessage")
@ResponseBody
public String keymessage(HttpServletRequest request, Model model) {
String topic = get(request, "topic");
String content = get(request, "content");
String keyStr = get(request, "key");
SimpleMessage simpleMessage = new SimpleMessage();
simpleMessage.setContent(content);
simpleMessage.setFrom(TOMCAT_ID);
String message = JSON.toJSONString(simpleMessage);
logger.info("start simple, topic [{}], key [{}], message [{}]", topic, keyStr, message);
messageService.sendKeyMsg(topic, keyStr, message);
logger.info("end simple, topic [{}], key [{}], message [{}]", topic, keyStr, message);
return String.format("success [%s], topic [%s], key [%s], content [%s]", tag(), topic, keyStr, content);
}
如上所示,收到web请求时,会构造一个SimpleMessage对象,里面有消息内容和当前tomcat的标识(因为有两个tomcat,通过此属性可以区分是哪个发的消息),再将这个对象转成字符串,然后请求kafka发送该消息,key也是从请求参数中取得的; 7. 部署到tomcat上去: 按照我们之前的规划,kafkaclusterproducerdemo要部署到producer1和producer2这两个容器上,producer1和producer2的8080端口分别映射到了当前电脑的19014和19015端口,所以我们在pom.xml中,tomcat7-maven-plugin插件的url参数中端口改为19014和19015分别部署一次,就能将war在线部署到两个tomcat上去了,如下图:
<dependency>
<groupId>org.apache.zookeeper</groupId>
<artifactId>zookeeper</artifactId>
<version>3.4.6</version>
</dependency>
<!--kafka-->
<dependency>
<groupId>org.apache.kafka</groupId>
<artifactId>kafka_2.9.2</artifactId>
<version>0.8.1</version>
<exclusions>
<exclusion>
<artifactId>org.apache.zookeeper</artifactId>
<groupId>zookeeper</groupId>
</exclusion>
</exclusions>
</dependency>
如上所示,zookeeper的版本是3.4.6,而kafka中对zookeeper的间接依赖已经被排除了; 3. 对kafka消息的消费,已经封装在一个Bean中,初始化逻辑如下:
private static final String GROUP_ID = "testgroup001";
private static final String ZK = "hostb1:2181,hostb2:2181,hostb3:2181";
private static final String TOPIC = "test002";
private static final int THREAD_NUM = 2;
@PostConstruct
public void init(){
logger.info("start init kafka consumer service");
// 1. 创建Kafka连接器
consumer = Consumer.createJavaConsumerConnector(createConsumerConfig(ZK, GROUP_ID));
Map<String, Integer> topicCountMap = new HashMap<String, Integer>();
topicCountMap.put(TOPIC, THREAD_NUM);
// 2. 指定数据的解码器
StringDecoder keyDecoder = new StringDecoder(new VerifiableProperties());
StringDecoder valueDecoder = new StringDecoder(new VerifiableProperties());
// 3. 获取连接数据的迭代器对象集合
/**
* Key: Topic主题
* Value: 对应Topic的数据流读取器,大小是topicCountMap中指定的topic大小
*/
Map<String, List<KafkaStream<String, String>>> consumerMap = this.consumer.createMessageStreams(topicCountMap, keyDecoder, valueDecoder);
// 4. 从返回结果中获取对应topic的数据流处理器
List<KafkaStream<String, String>> streams = consumerMap.get(TOPIC);
logger.info("streams size {}", streams.size());
// 5. 创建线程池
this.executorPool = new ThreadPoolExecutor(THREAD_NUM, THREAD_NUM,
0,
TimeUnit.MILLISECONDS,
new LinkedBlockingQueue<Runnable>(),
new CustomThreadFactory(),
new ThreadPoolExecutor.AbortPolicy());
// 6. 构建数据输出对象
int threadNumber = 0;
for (final KafkaStream<String, String> stream : streams) {
this.executorPool.submit(new Processer(stream, threadNumber));
threadNumber++;
}
logger.info("end init kafka consumer service");
}
如上所示,我们用的是High level的方式,很多细节的东西就不用关心了,例如commit,offset等; 由于本次实战的消息打算分成6个partition,一共有三个消费tomcat,所以每个tomcat上启动两个消费者线程来处理消息,这样就保证了每个partition都有一个单独的线程来处理; 4. 消息处理的业务类Processer实现了Runnable,关键代码如下:
public void run() {
// 1. 获取数据迭代器
ConsumerIterator<String, String> iter = this.stream.iterator();
logger.info("server [{}] start run", TOMCAT_ID);
// 2. 迭代输出数据
while (iter.hasNext()) {
// 2.1 获取数据值
MessageAndMetadata value = iter.next();
// 2.2 输出
logger.info("server [{}], threadNumber [{}], offset [{}], key [{}], message[{}]",
TOMCAT_ID,
threadNumber,
value.offset(),
value.key(),
value.message());
}
// 3. 表示当前线程执行完成
logger.info("Shutdown Thread:" + this.threadNumber);
}
如上,每当从kafka中取得了消息,就通过日志打印出来; 7. 部署到tomcat上去: 按照我们之前的规划,kafkaclusterconsumerdemo要部署到consumer1、consumer2、consumer3这三个容器上,它们的8080端口分别映射到了当前电脑的19017、19018、19019端口上,所以我们在pom.xml中,tomcat7-maven-plugin插件的url参数中端口改为这三个端口分别部署一次,就能将war在线部署到三个tomcat上去了;
推荐同时打开五个控制台,分别登上producer1、producer2、consumer1、consumer2、consumer3这五个容器,查看日志时分别做如下操作: 1. 对producer1和producer2,执行:
tail -f /usr/local/tomcat/logs/kafkaclusterproducerdemo/output.2017-10-29.log
output.xxxxxx.log要用当天的日期; 2. 对consumer1、consumer2、consumer3,执行:
tail -f /usr/local/tomcat/logs/kafkaclusterconsumerdemo/output.2017-10-29.log
这样就能将日志实时打印出来了;
实战中经常出现consumer1、consumer2、consumer3等容器在部署了war包后启动失败的情况,请浏览器输入http://localhost:19017/kafkaclusterconsumerdemo来检查consumer1是否启动成功,启动成功后的效果如下:
如果启动失败页面就404错误了,检查日志发现是连接zookeeper失败,如下图:
这时候建议多部署几次,就能连接成功了,启动成功的日志如下图所示:
用浏览器分别访问以下六个地址: 1. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message001&key=1 2. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message002&key=2 3. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message003&key=3 4. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message004&key=4 5. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message005&key=5 6. http://localhost:19016/kafkaclusterproducerdemo/keymessage?topic=test002&content=message006&key=6
上面六个地址代表发送了六个消息,例如第一个消息的key是1,消息内容是message001,在consumer3的日志中我们看到了这个消息,如下图所示:
从from字段我们还能发现这个消息是从producer1发出的;
从每个consumer日志中的key可以将consumer1、consumer2、consumer3和partition的关系梳理如下表:
容器 | partition |
---|---|
consumer1 | partition4、partition5 |
consumer2 | partition0、partition1 |
consumer3 | partition2、partition3 |
至此,kafka集群环境下的java开发实战就全部结束了,和之前的入门实战相比稍微复杂了一些,但也更接近实际生产环境的操作了,希望能对读者您的学习和开发有所帮助;