[转载] Kafka+Storm+HDFS整合实践

2022-10-03 17:54:23

转载自http://www.tuicool.com/articles/NzyqAn

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。

码农公寓

相关文章