Kafka+Storm+HDFS整合实践
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
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
更多推荐
所有评论(0)