在基于Hadoop平台的很多应用场景中,我们需要对数据进行离线和实时分析,离线分析可以很容易地借助于Hive来实现统计分析,但是对于实时的需求Hive就不合适了。实时应用场景可以使用Storm,它是一个实时处理系统,它为实时处理类应用提供了一个计算模型,可以很容易地进行编程处理。为了统一离线和实时计算,一般情况下,我们都希望将离线和实时计算的数据源的集合统一起来作为输入,然后将数据的流向分别经由实时系统和离线分析系统,分别进行分析处理,这时我们可以考虑将数据源(如使用Flume收集日志)直接连接一个消息中间件,如Kafka,可以整合Flume+Kafka,Flume作为消息的Producer,生产的消息数据(日志数据、业务请求数据等等)发布到Kafka中,然后通过订阅的方式,使用Storm的Topology作为消息的Consumer,在Storm集群中分别进行如下两个需求场景的处理:

  • 直接使用Storm的Topology对数据进行实时分析处理
  • 整合Storm+HDFS,将消息处理后写入HDFS进行离线分析处理

实时处理,只要开发满足业务需要的Topology即可,不做过多说明。这里,我们主要从安装配置Kafka、Storm,以及整合Kafka+Storm、整合Storm+HDFS、整合Kafka+Storm+HDFS这几点来配置实践,满足上面提出的一些需求。配置实践使用的软件包如下所示:

  • zookeeper-3.4.5.tar.gz
  • kafka_2.9.2-0.8.1.1.tgz
  • apache-storm-0.9.2-incubating.tar.gz
  • hadoop-2.2.0.tar.gz

程序配置运行所基于的操作系统为CentOS 5.11。

Kafka安装配置

我们使用3台机器搭建Kafka集群:

192.168.4.142   h1
192.168.4.143   h2
192.168.4.144   h3

在安装Kafka集群之前,这里没有使用Kafka自带的Zookeeper,而是独立安装了一个Zookeeper集群,也是使用这3台机器,保证Zookeeper集群正常运行。 首先,在h1上准备Kafka安装文件,执行如下命令:

cd /usr/local/
wget http://mirror.bit.edu.cn/apache/kafka/0.8.1.1/kafka_2.9.2-0.8.1.1.tgz
tar xvzf kafka_2.9.2-0.8.1.1.tgz
ln -s /usr/local/kafka_2.9.2-0.8.1.1 /usr/local/kafka
chown -R kafka:kafka /usr/local/kafka_2.9.2-0.8.1.1 /usr/local/kafka

修改配置文件/usr/local/kafka/config/server.properties,修改如下内容:

broker.id=0
zookeeper.connect=h1:2181,h2:2181,h3:2181

然后,将配置好的安装文件同步到其他的h2、h3节点上:

scp -r /usr/local/kafka_2.9.2-0.8.1.1/ h2:/usr/local/
scp -r /usr/local/kafka_2.9.2-0.8.1.1/ h3:/usr/local/

最后,在h2、h3节点上配置,执行如下命令:

cd /usr/local/
ln -s /usr/local/kafka_2.9.2-0.8.1.1 /usr/local/kafka
chown -R kafka:kafka /usr/local/kafka_2.9.2-0.8.1.1 /usr/local/kafka

并修改配置文件/usr/local/kafka/config/server.properties内容如下所示:

broker.id=1  # 在h1修改

broker.id=2  # 在h2修改

因为Kafka集群需要保证各个Broker的id在整个集群中必须唯一,需要调整这个配置项的值(如果在单机上,可以通过建立多个Broker进程来模拟分布式的Kafka集群,也需要Broker的id唯一,还需要修改一些配置目录的信息)。 在集群中的h1、h2、h3这三个节点上分别启动Kafka,分别执行如下命令:

bin/kafka-server-start.sh /usr/local/kafka/config/server.properties &

可以通过查看日志,或者检查进程状态,保证Kafka集群启动成功。 我们创建一个名称为my-replicated-topic5的Topic,5个分区,并且复制因子为3,执行如下命令:

bin/kafka-topics.sh --create --zookeeper h1:2181,h2:2181,h3:2181 --replication-factor 3 --partitions 5 --topic my-replicated-topic5

查看创建的Topic,执行如下命令:

bin/kafka-topics.sh --describe --zookeeper h1:2181,h2:2181,h3:2181 --topic my-replicated-topic5

结果信息如下所示:

Topic:my-replicated-topic5	PartitionCount:5	ReplicationFactor:3	Configs:
	Topic: my-replicated-topic5	Partition: 0	Leader: 0	Replicas: 0,2,1	Isr: 0,2,1
	Topic: my-replicated-topic5	Partition: 1	Leader: 0	Replicas: 1,0,2	Isr: 0,2,1
	Topic: my-replicated-topic5	Partition: 2	Leader: 2	Replicas: 2,1,0	Isr: 2,0,1
	Topic: my-replicated-topic5	Partition: 3	Leader: 0	Replicas: 0,1,2	Isr: 0,2,1
	Topic: my-replicated-topic5	Partition: 4	Leader: 2	Replicas: 1,2,0	Isr: 2,0,1

上面Leader、Replicas、Isr的含义如下:

Partition: 分区
Leader   : 负责读写指定分区的节点
Replicas : 复制该分区log的节点列表
Isr      : "in-sync" replicas,当前活跃的副本列表(是一个子集),并且可能成为Leader

我们可以通过Kafka自带的bin/kafka-console-producer.sh和bin/kafka-console-consumer.sh脚本,来验证演示如果发布消息、消费消息。 在一个终端,启动Producer,并向我们上面创建的名称为my-replicated-topic5的Topic中生产消息,执行如下脚本:

bin/kafka-console-producer.sh --broker-list h1:9092,h2:9092,h3:9092 --topic my-replicated-topic5

在另一个终端,启动Consumer,并订阅我们上面创建的名称为my-replicated-topic5的Topic中生产的消息,执行如下脚本:

bin/kafka-console-consumer.sh --zookeeper h1:2181,h2:2181,h3:2181 --from-beginning --topic my-replicated-topic5

可以在Producer终端上输入字符串消息行,然后回车,就可以在Consumer终端上看到消费者消费的消息内容。 也可以参考Kafka的Producer和Consumer的Java API,通过API编码的方式来实现消息生产和消费的处理逻辑。

Storm安装配置

Storm集群也依赖Zookeeper集群,要保证Zookeeper集群正常运行。Storm的安装配置比较简单,我们仍然使用下面3台机器搭建:

192.168.4.142   h1
192.168.4.143   h2
192.168.4.144   h3

首先,在h1节点上,执行如下命令安装:

cd /usr/local/
wget http://mirror.bit.edu.cn/apache/incubator/storm/apache-storm-0.9.2-incubating/apache-storm-0.9.2-incubating.tar.gz
tar xvzf apache-storm-0.9.2-incubating.tar.gz
ln -s /usr/local/apache-storm-0.9.2-incubating /usr/local/storm
chown -R storm:storm /usr/local/apache-storm-0.9.2-incubating /usr/local/storm

然后,修改配置文件conf/storm.yaml,内容如下所示:

storm.zookeeper.servers:
	- "h1"
	- "h2"
	- "h3"
storm.zookeeper.port: 2181
#
nimbus.host: "h1"

supervisor.slots.ports:
    - 6700
    - 6701
    - 6702
    - 6703

storm.local.dir: "/tmp/storm"

将配置好的安装文件,分发到其他节点上:

scp -r /usr/local/apache-storm-0.9.2-incubating/ h2:/usr/local/
scp -r /usr/local/apache-storm-0.9.2-incubating/ h3:/usr/local/

最后,在h2、h3节点上配置,执行如下命令:

cd /usr/local/
ln -s /usr/local/apache-storm-0.9.2-incubating /usr/local/storm
chown -R storm:storm /usr/local/apache-storm-0.9.2-incubating /usr/local/storm

Storm集群的主节点为Nimbus,从节点为Supervisor,我们需要在h1上启动Nimbus服务,在从节点h2、h3上启动Supervisor服务:

bin/storm nimbus &
bin/storm supervisor &

为了方便监控,可以启动Storm UI,可以从Web页面上监控Storm Topology的运行状态,例如在h2上启动:

bin/storm ui &

这样可以通过访问 http://h2:8080/ 来查看Topology的运行状况。

整合Kafka+Storm

消息通过各种方式进入到Kafka消息中间件,比如可以通过使用Flume来收集日志数据,然后在Kafka中路由暂存,然后再由实时计算程序Storm做实时分析,这时我们就需要将在Storm的Spout中读取Kafka中的消息,然后交由具体的Spot组件去分析处理。实际上,apache-storm-0.9.2-incubating这个版本的Storm已经自带了一个集成Kafka的外部插件程序storm-kafka,可以直接使用,例如我使用的Maven依赖配置,如下所示:

<dependency>
			<groupId>org.apache.storm</groupId>
			<artifactId>storm-core</artifactId>
			<version>0.9.2-incubating</version>
			<scope>provided</scope>
		</dependency>
		<dependency>
			<groupId>org.apache.storm</groupId>
			<artifactId>storm-kafka</artifactId>
			<version>0.9.2-incubating</version>
		</dependency>
		<dependency>
			<groupId>org.apache.kafka</groupId>
			<artifactId>kafka_2.9.2</artifactId>
			<version>0.8.1.1</version>
			<exclusions>
				<exclusion>
					<groupId>org.apache.zookeeper</groupId>
					<artifactId>zookeeper</artifactId>
				</exclusion>
				<exclusion>
					<groupId>log4j</groupId>
					<artifactId>log4j</artifactId>
				</exclusion>
			</exclusions>
		</dependency>

下面,我们开发了一个简单WordCount示例程序,从Kafka读取订阅的消息行,通过空格拆分出单个单词,然后再做词频统计计算,实现的Topology的代码,如下所示:

package org.shirdrn.storm.examples;

import java.util.Arrays;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.Map.Entry;
import java.util.concurrent.atomic.AtomicInteger;

import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;

import storm.kafka.BrokerHosts;
import storm.kafka.KafkaSpout;
import storm.kafka.SpoutConfig;
import storm.kafka.StringScheme;
import storm.kafka.ZkHosts;
import backtype.storm.Config;
import backtype.storm.LocalCluster;
import backtype.storm.StormSubmitter;
import backtype.storm.generated.AlreadyAliveException;
import backtype.storm.generated.InvalidTopologyException;
import backtype.storm.spout.SchemeAsMultiScheme;
import backtype.storm.task.OutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.topology.TopologyBuilder;
import backtype.storm.topology.base.BaseRichBolt;
import backtype.storm.tuple.Fields;
import backtype.storm.tuple.Tuple;
import backtype.storm.tuple.Values;

public class MyKafkaTopology {

	public static class KafkaWordSplitter extends BaseRichBolt {

		private static final Log LOG = LogFactory.getLog(KafkaWordSplitter.class);
		private static final long serialVersionUID = 886149197481637894L;
		private OutputCollector collector;
	    
		@Override
		public void prepare(Map stormConf, TopologyContext context,
				OutputCollector collector) {
			this.collector = collector;		    
		}

		@Override
		public void execute(Tuple input) {
			String line = input.getString(0);
			LOG.info("RECV[kafka -> splitter] " + line);
			String[] words = line.split("\\s+");
			for(String word : words) {
				LOG.info("EMIT[splitter -> counter] " + word);
				collector.emit(input, new Values(word, 1));
			}
			collector.ack(input);
		}

		@Override
		public void declareOutputFields(OutputFieldsDeclarer declarer) {
			declarer.declare(new Fields("word", "count"));	    
		}
	    
	}
    
	public static class WordCounter extends BaseRichBolt {

		private static final Log LOG = LogFactory.getLog(WordCounter.class);
		private static final long serialVersionUID = 886149197481637894L;
		private OutputCollector collector;
		private Map<String, AtomicInteger> counterMap;
	    
		@Override
		public void prepare(Map stormConf, TopologyContext context,
				OutputCollector collector) {
			this.collector = collector;    
			this.counterMap = new HashMap<String, AtomicInteger>();
		}

		@Override
		public void execute(Tuple input) {
			String word = input.getString(0);
			int count = input.getInteger(1);
			LOG.info("RECV[splitter -> counter] " + word + " : " + count);
			AtomicInteger ai = this.counterMap.get(word);
			if(ai == null) {
				ai = new AtomicInteger();
				this.counterMap.put(word, ai);
			}
			ai.addAndGet(count);
			collector.ack(input);
			LOG.info("CHECK statistics map: " + this.counterMap);
		}

		@Override
		public void cleanup() {
			LOG.info("The final result:");
			Iterator<Entry<String, AtomicInteger>> iter = this.counterMap.entrySet().iterator();
			while(iter.hasNext()) {
				Entry<String, AtomicInteger> entry = iter.next();
				LOG.info(entry.getKey() + "\t:\t" + entry.getValue().get());
			}
		    
		}

		@Override
		public void declareOutputFields(OutputFieldsDeclarer declarer) {
			declarer.declare(new Fields("word", "count"));	    
		}
	}
    
	public static void main(String[] args) throws AlreadyAliveException, InvalidTopologyException, InterruptedException {
		String zks = "h1:2181,h2:2181,h3:2181";
		String topic = "my-replicated-topic5";
		String zkRoot = "/storm"; // default zookeeper root configuration for storm
		String id = "word";
	    
		BrokerHosts brokerHosts = new ZkHosts(zks);
		SpoutConfig spoutConf = new SpoutConfig(brokerHosts, topic, zkRoot, id);
		spoutConf.scheme = new SchemeAsMultiScheme(new StringScheme());
		spoutConf.forceFromStart = true;
		spoutConf.zkServers = Arrays.asList(new String[] {"h1", "h2", "h3"});
		spoutConf.zkPort = 2181;
	    
		TopologyBuilder builder = new TopologyBuilder();
		builder.setSpout("kafka-reader", new KafkaSpout(spoutConf), 5); // Kafka我们创建了一个5分区的Topic,这里并行度设置为5
		builder.setBolt("word-splitter", new KafkaWordSplitter(), 2).shuffleGrouping("kafka-reader");
		builder.setBolt("word-counter", new WordCounter()).fieldsGrouping("word-splitter", new Fields("word"));
	    
		Config conf = new Config();
	    
		String name = MyKafkaTopology.class.getSimpleName();
		if (args != null && args.length > 0) {
			// Nimbus host name passed from command line
			conf.put(Config.NIMBUS_HOST, args[0]);
			conf.setNumWorkers(3);
			StormSubmitter.submitTopologyWithProgressBar(name, conf, builder.createTopology());
		} else {
			conf.setMaxTaskParallelism(3);
			LocalCluster cluster = new LocalCluster();
			cluster.submitTopology(name, conf, builder.createTopology());
			Thread.sleep(60000);
			cluster.shutdown();
		}
	}
}

上面程序,在本地调试(使用LocalCluster)不需要输入任何参数,提交到实际集群中运行时,需要传递一个参数,该参数为Nimbus的主机名称。 通过Maven构建,生成一个包含依赖的single jar文件(不要把Storm的依赖包添加进去),例如storm-examples-0.0.1-SNAPSHOT.jar,在提交Topology程序到Storm集群之前,因为用到了Kafka,需要拷贝一下依赖jar文件到Storm集群中的lib目录下面:

cp /usr/local/kafka/libs/kafka_2.9.2-0.8.1.1.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/scala-library-2.9.2.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/metrics-core-2.2.0.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/snappy-java-1.0.5.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/zkclient-0.3.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/log4j-1.2.15.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/slf4j-api-1.7.2.jar /usr/local/storm/lib/
cp /usr/local/kafka/libs/jopt-simple-3.2.jar /usr/local/storm/lib/

然后,就可以提交我们开发的Topology程序了:

bin/storm jar /home/storm/storm-examples-0.0.1-SNAPSHOT.jar org.shirdrn.storm.examples.MyKafkaTopology h1

可以通过查看日志文件(logs/目录下)或者Storm UI来监控Topology的运行状况。如果程序没有错误,可以使用前面我们使用的Kafka Producer来生成消息,就能看到我们开发的Storm Topology能够实时接收到并进行处理。

整合Storm+HDFS

Storm实时计算集群从Kafka消息中间件中消费消息,有实时处理需求的可以走实时处理程序,还有需要进行离线分析的需求,如写入到HDFS进行分析。下面实现了一个Topology,代码如下所示:

package org.shirdrn.storm.examples;

import java.text.DateFormat;
import java.text.SimpleDateFormat;
import java.util.Date;
import java.util.Map;
import java.util.Random;

import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.storm.hdfs.bolt.HdfsBolt;
import org.apache.storm.hdfs.bolt.format.DefaultFileNameFormat;
import org.apache.storm.hdfs.bolt.format.DelimitedRecordFormat;
import org.apache.storm.hdfs.bolt.format.FileNameFormat;
import org.apache.storm.hdfs.bolt.format.RecordFormat;
import org.apache.storm.hdfs.bolt.rotation.FileRotationPolicy;
import org.apache.storm.hdfs.bolt.rotation.TimedRotationPolicy;
import org.apache.storm.hdfs.bolt.rotation.TimedRotationPolicy.TimeUnit;
import org.apache.storm.hdfs.bolt.sync.CountSyncPolicy;
import org.apache.storm.hdfs.bolt.sync.SyncPolicy;

import backtype.storm.Config;
import backtype.storm.LocalCluster;
import backtype.storm.StormSubmitter;
import backtype.storm.generated.AlreadyAliveException;
import backtype.storm.generated.InvalidTopologyException;
import backtype.storm.spout.SpoutOutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.topology.TopologyBuilder;
import backtype.storm.topology.base.BaseRichSpout;
import backtype.storm.tuple.Fields;
import backtype.storm.tuple.Values;
import backtype.storm.utils.Utils;

public class StormToHDFSTopology {

     public static class EventSpout extends BaseRichSpout {

	private static final Log LOG = LogFactory.getLog(EventSpout.class);
	private static final long serialVersionUID = 886149197481637894L;
	private SpoutOutputCollector collector;
	private Random rand;
	private String[] records;
         
	@Override
	public void open(Map conf, TopologyContext context,
		SpoutOutputCollector collector) {
	     this.collector = collector;    
	     rand = new Random();
	     records = new String[] {
		     "10001     ef2da82d4c8b49c44199655dc14f39f6     4.2.1     HUAWEI G610-U00     HUAWEI     2     70:72:3c:73:8b:22     2014-10-13 12:36:35",
		     "10001     ffb52739a29348a67952e47c12da54ef     4.3     GT-I9300     samsung     2     50:CC:F8:E4:22:E2     2014-10-13 12:36:02",
		     "10001     ef2da82d4c8b49c44199655dc14f39f6     4.2.1     HUAWEI G610-U00     HUAWEI     2     70:72:3c:73:8b:22     2014-10-13 12:36:35"
	     };
	}


	@Override
	public void nextTuple() {
	     Utils.sleep(1000);
	     DateFormat df = new SimpleDateFormat("yyyy-MM-dd_HH-mm-ss");
	     Date d = new Date(System.currentTimeMillis());
	     String minute = df.format(d);
	     String record = records[rand.nextInt(records.length)];
	     LOG.info("EMIT[spout -> hdfs] " + minute + " : " + record);
	     collector.emit(new Values(minute, record));
	}

	@Override
	public void declareOutputFields(OutputFieldsDeclarer declarer) {
	     declarer.declare(new Fields("minute", "record"));         
	}


     }
    
     public static void main(String[] args) throws AlreadyAliveException, InvalidTopologyException, InterruptedException {
	// use "|" instead of "," for field delimiter
	RecordFormat format = new DelimitedRecordFormat()
	        .withFieldDelimiter(" : ");

	// sync the filesystem after every 1k tuples
	SyncPolicy syncPolicy = new CountSyncPolicy(1000);

	// rotate files 
	FileRotationPolicy rotationPolicy = new TimedRotationPolicy(1.0f, TimeUnit.MINUTES);

	FileNameFormat fileNameFormat = new DefaultFileNameFormat()
	        .withPath("/storm/").withPrefix("app_").withExtension(".log");

	HdfsBolt hdfsBolt = new HdfsBolt()
	        .withFsUrl("hdfs://h1:8020")
	        .withFileNameFormat(fileNameFormat)
	        .withRecordFormat(format)
	        .withRotationPolicy(rotationPolicy)
	        .withSyncPolicy(syncPolicy);
         
	TopologyBuilder builder = new TopologyBuilder();
	builder.setSpout("event-spout", new EventSpout(), 3);
	builder.setBolt("hdfs-bolt", hdfsBolt, 2).fieldsGrouping("event-spout", new Fields("minute"));
         
	Config conf = new Config();
         
	String name = StormToHDFSTopology.class.getSimpleName();
	if (args != null && args.length > 0) {
	     conf.put(Config.NIMBUS_HOST, args[0]);
	     conf.setNumWorkers(3);
	     StormSubmitter.submitTopologyWithProgressBar(name, conf, builder.createTopology());
	} else {
	     conf.setMaxTaskParallelism(3);
	     LocalCluster cluster = new LocalCluster();
	     cluster.submitTopology(name, conf, builder.createTopology());
	     Thread.sleep(60000);
	     cluster.shutdown();
	}
     }

}

上面的处理逻辑,可以对HdfsBolt进行更加详细的配置,如FileNameFormat、SyncPolicy、FileRotationPolicy(可以设置在满足什么条件下,切出一个新的日志,如可以指定多长时间切出一个新的日志文件,可以指定一个日志文件大小达到设置值后,再写一个新日志文件),更多设置可以参考storm-hdfs,。 上面代码在打包的时候,需要注意,使用storm-starter自带的Maven打包配置,可能在将Topology部署运行的时候,会报错,可以使用maven-shade-plugin这个插件,如下配置所示:

<plugin>
				   <groupId>org.apache.maven.plugins</groupId>
				   <artifactId>maven-shade-plugin</artifactId>
				   <version>1.4</version>
				   <configuration>
					   <createDependencyReducedPom>true</createDependencyReducedPom>
				   </configuration>
				   <executions>
					   <execution>
						   <phase>package</phase>
						   <goals>
							   <goal>shade</goal>
						   </goals>
						   <configuration>
							   <transformers>
								   <transformer
										   implementation="org.apache.maven.plugins.shade.resource.ServicesResourceTransformer"/>
								   <transformer
										   implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
									   <mainClass></mainClass>
								   </transformer>
							   </transformers>
						   </configuration>
					   </execution>
				   </executions>
			   </plugin>

整合Kafka+Storm+HDFS

上面分别对整合Kafka+Storm和Storm+HDFS做了实践,可以将后者的Spout改成前者的Spout,从Kafka中消费消息,在Storm中可以做简单处理,然后将数据写入HDFS,最后可以在Hadoop平台上对数据进行离线分析处理。下面,写了一个简单的例子,从Kafka消费消息,然后经由Storm处理,写入到HDFS存储,代码如下所示:

package org.shirdrn.storm.examples;

import java.util.Arrays;
import java.util.Map;

import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.storm.hdfs.bolt.HdfsBolt;
import org.apache.storm.hdfs.bolt.format.DefaultFileNameFormat;
import org.apache.storm.hdfs.bolt.format.DelimitedRecordFormat;
import org.apache.storm.hdfs.bolt.format.FileNameFormat;
import org.apache.storm.hdfs.bolt.format.RecordFormat;
import org.apache.storm.hdfs.bolt.rotation.FileRotationPolicy;
import org.apache.storm.hdfs.bolt.rotation.TimedRotationPolicy;
import org.apache.storm.hdfs.bolt.rotation.TimedRotationPolicy.TimeUnit;
import org.apache.storm.hdfs.bolt.sync.CountSyncPolicy;
import org.apache.storm.hdfs.bolt.sync.SyncPolicy;

import storm.kafka.BrokerHosts;
import storm.kafka.KafkaSpout;
import storm.kafka.SpoutConfig;
import storm.kafka.StringScheme;
import storm.kafka.ZkHosts;
import backtype.storm.Config;
import backtype.storm.LocalCluster;
import backtype.storm.StormSubmitter;
import backtype.storm.generated.AlreadyAliveException;
import backtype.storm.generated.InvalidTopologyException;
import backtype.storm.spout.SchemeAsMultiScheme;
import backtype.storm.task.OutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.topology.TopologyBuilder;
import backtype.storm.topology.base.BaseRichBolt;
import backtype.storm.tuple.Fields;
import backtype.storm.tuple.Tuple;
import backtype.storm.tuple.Values;

public class DistributeWordTopology {
    
     public static class KafkaWordToUpperCase extends BaseRichBolt {

	private static final Log LOG = LogFactory.getLog(KafkaWordToUpperCase.class);
	private static final long serialVersionUID = -5207232012035109026L;
	private OutputCollector collector;
         
	@Override
	public void prepare(Map stormConf, TopologyContext context,
		OutputCollector collector) {
	     this.collector = collector;	    
	}

	@Override
	public void execute(Tuple input) {
	     String line = input.getString(0).trim();
	     LOG.info("RECV[kafka -> splitter] " + line);
	     if(!line.isEmpty()) {
		String upperLine = line.toUpperCase();
		LOG.info("EMIT[splitter -> counter] " + upperLine);
		collector.emit(input, new Values(upperLine, upperLine.length()));
	     }
	     collector.ack(input);
	}

	@Override
	public void declareOutputFields(OutputFieldsDeclarer declarer) {
	     declarer.declare(new Fields("line", "len"));         
	}
         
     }
    
     public static class RealtimeBolt extends BaseRichBolt {

	private static final Log LOG = LogFactory.getLog(KafkaWordToUpperCase.class);
	private static final long serialVersionUID = -4115132557403913367L;
	private OutputCollector collector;
         
	@Override
	public void prepare(Map stormConf, TopologyContext context,
		OutputCollector collector) {
	     this.collector = collector;	    
	}

	@Override
	public void execute(Tuple input) {
	     String line = input.getString(0).trim();
	     LOG.info("REALTIME: " + line);
	     collector.ack(input);
	}

	@Override
	public void declareOutputFields(OutputFieldsDeclarer declarer) {
	    
	}

     }

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

	// Configure Kafka
	String zks = "h1:2181,h2:2181,h3:2181";
	String topic = "my-replicated-topic5";
	String zkRoot = "/storm"; // default zookeeper root configuration for storm
	String id = "word";
	BrokerHosts brokerHosts = new ZkHosts(zks);
	SpoutConfig spoutConf = new SpoutConfig(brokerHosts, topic, zkRoot, id);
	spoutConf.scheme = new SchemeAsMultiScheme(new StringScheme());
	spoutConf.forceFromStart = true;
	spoutConf.zkServers = Arrays.asList(new String[] {"h1", "h2", "h3"});
	spoutConf.zkPort = 2181;
         
	// Configure HDFS bolt
	RecordFormat format = new DelimitedRecordFormat()
	        .withFieldDelimiter("\t"); // use "\t" instead of "," for field delimiter
	SyncPolicy syncPolicy = new CountSyncPolicy(1000); // sync the filesystem after every 1k tuples
	FileRotationPolicy rotationPolicy = new TimedRotationPolicy(1.0f, TimeUnit.MINUTES); // rotate files
	FileNameFormat fileNameFormat = new DefaultFileNameFormat()
	        .withPath("/storm/").withPrefix("app_").withExtension(".log"); // set file name format
	HdfsBolt hdfsBolt = new HdfsBolt()
	        .withFsUrl("hdfs://h1:8020")
	        .withFileNameFormat(fileNameFormat)
	        .withRecordFormat(format)
	        .withRotationPolicy(rotationPolicy)
	        .withSyncPolicy(syncPolicy);
         
	// configure & build topology
	TopologyBuilder builder = new TopologyBuilder();
	builder.setSpout("kafka-reader", new KafkaSpout(spoutConf), 5);
	builder.setBolt("to-upper", new KafkaWordToUpperCase(), 3).shuffleGrouping("kafka-reader");
	builder.setBolt("hdfs-bolt", hdfsBolt, 2).shuffleGrouping("to-upper");
	builder.setBolt("realtime", new RealtimeBolt(), 2).shuffleGrouping("to-upper");
         
	// submit topology
	Config conf = new Config();
	String name = DistributeWordTopology.class.getSimpleName();
	if (args != null && args.length > 0) {
	     String nimbus = args[0];
	     conf.put(Config.NIMBUS_HOST, nimbus);
	     conf.setNumWorkers(3);
	     StormSubmitter.submitTopologyWithProgressBar(name, conf, builder.createTopology());
	} else {
	     conf.setMaxTaskParallelism(3);
	     LocalCluster cluster = new LocalCluster();
	     cluster.submitTopology(name, conf, builder.createTopology());
	     Thread.sleep(60000);
	     cluster.shutdown();
	}
     }

}

上面代码中,名称为to-upper的Bolt将接收到的字符串行转换成大写以后,会将处理过的数据向后面的hdfs-bolt、realtime这两个Bolt各发一份拷贝,然后由这两个Bolt分别根据实际需要(实时/离线)单独处理。 打包后,在Storm集群上部署并运行这个Topology:

bin/storm jar ~/storm-examples-0.0.1-SNAPSHOT.jar org.shirdrn.storm.examples.DistributeWordTopology h1

可以通过Storm UI查看Topology运行情况,可以查看HDFS上生成的数据。


from :2014-10-25作者:简单之美 来源:简单之美

http://www.350351.com/jiagoucunchu/xiaoxixitong/400497.html

Logo

Kafka开源项目指南提供详尽教程,助开发者掌握其架构、配置和使用,实现高效数据流管理和实时处理。它高性能、可扩展,适合日志收集和实时数据处理,通过持久化保障数据安全,是企业大数据生态系统的核心。

更多推荐