没有向ZK写offset数据
两天公司要学习kafka,结合之前的storm,做了一个简单的集成,之前也参考了网上的例子一些例子,发现或多或少都有一些问题。所以自己做了一个。
这个是网上其他人遇到的问题,给摘录一下,防止以后自己和大家出现:
这几天工作需要使用storm+kafka,基本场景是应用出现错误,发送日志到kafka的某个topic,storm订阅该topic,然后进行后续处理。场景非常简单,但是在学习过程中,遇到一个奇怪的异常情况:使用KafkaSpout读取topic数据时,没有向ZK写offset数据,致使每次都从头开始读取。纠结了两天,终于碰巧找到原因:应该使用BaseBasicBolt
作为bolt的父类,而不是BaseRichBolt
。
通过本文记录一下这种情况,后文中根据上述场景提供几个简单的例子。基础理论查看storm笔记:storm基本概念,或查看Storm 简介。
基本订阅
基本场景:订阅kafka的某个topic,然后在读取的消息前加上自定义的字符串,然后写回到kafka另外一个topic。
从Kafka读取数据的Spout使用storm.kafka.KafkaSpout
,向Kafka写数据的Bolt使用storm.kafka.bolt.KafkaBolt
。中间进行进行数据处理的Bolt定义为TopicMsgBolt
。闲言少叙,奉上代码:
public class TopicMsgTopology {
public static void main(String[] args) throws Exception {
// 配置Zookeeper地址
BrokerHosts brokerHosts = new ZkHosts("zk1:2181,zk2:2281,zk3:2381");
// 配置Kafka订阅的Topic,以及zookeeper中数据节点目录和名字
SpoutConfig spoutConfig = new SpoutConfig(brokerHosts, "msgTopic1", "/topology/root", "topicMsgTopology");
// 配置KafkaBolt中的kafka.broker.properties
Config conf = new Config();
Properties props = new Properties();
// 配置Kafka broker地址
props.put("metadata.broker.list", "dev2_55.wfj-search:9092");
// serializer.class为消息的序列化类
props.put("serializer.class", "kafka.serializer.StringEncoder");
conf.put("kafka.broker.properties", props);
// 配置KafkaBolt生成的topic
conf.put("topic", "msgTopic2");
spoutConfig.scheme = new SchemeAsMultiScheme(new MessageScheme());
TopologyBuilder builder = new TopologyBuilder();
builder.setSpout("msgKafkaSpout", new KafkaSpout(spoutConfig));
builder.setBolt("msgSentenceBolt", new TopicMsgBolt()).shuffleGrouping("msgKafkaSpout");
builder.setBolt("msgKafkaBolt", new KafkaBolt<String, Integer>()).shuffleGrouping("msgSentenceBolt");
if (args.length == 0) {
String topologyName = "kafkaTopicTopology";
LocalCluster cluster = new LocalCluster();
cluster.submitTopology(topologyName, conf, builder.createTopology());
Utils.sleep(100000);
cluster.killTopology(topologyName);
cluster.shutdown();
} else {
conf.setNumWorkers(1);
StormSubmitter.submitTopology(args[0], conf, builder.createTopology());
}
}
}
storm.kafka.ZkHosts
构造方法的参数是zookeeper标准配置地址的形式(ZooKeeper环境搭建可以查看ZooKeeper安装部署),zk1、zk2、zk3在本地配置了host,因为服务器使用的伪分布式模式,因此几个端口号不是默认的2181。
storm.kafka.SpoutConfig
构造方法第一个参数为上述的storm.kafka.ZkHosts
对象,第二个为待订阅的topic名称,第三个参数zkRoot为写读取topic时的偏移量offset数据的节点(zk node),第四个参数为该节点上的次级节点名(有个地方说这个是spout的id)。
backtype.storm.Config
对象是配置storm的topology(拓扑)所需要的基础配置。
backtype.storm.spout.SchemeAsMultiScheme
的构造方法输入的参数是订阅kafka数据的处理参数,这里的MessageScheme
是自定义的,代码如下:
public class MessageScheme implements Scheme {
private static final Logger logger = LoggerFactory.getLogger(MessageScheme.class);
@Override
public List<Object> deserialize(byte[] ser) {
try {
String msg = new String(ser, "UTF-8");
logger.info("get one message is {}", msg);
return new Values(msg);
} catch (UnsupportedEncodingException ignored) {
return null;
}
}
@Override
public Fields getOutputFields() {
return new Fields("msg");
}
}
MessageScheme
类中getOutputFields方法是KafkaSpout向后发送tuple(storm传输数据的最小结构)的名字,需要与接收数据的Bolt中统一(在这个例子中可以不统一,因为后面直接取第0条数据,但是在wordCount的那个例子中就需要统一了)。
TopicMsgBolt
类是从storm.kafka.KafkaSpout
接收数据的Bolt,对接收到的数据进行处理,然后向后传输给storm.kafka.bolt.KafkaBolt
。代码如下:
public class TopicMsgBolt extends BaseBasicBolt {
private static final Logger logger = LoggerFactory.getLogger(TopicMsgBolt.class);
@Override
public void execute(Tuple input, BasicOutputCollector collector) {
String word = (String) input.getValue(0);
String out = "Message got is '" + word + "'!";
logger.info("out={}", out);
collector.emit(new Values(out));
}
@Override
public void declareOutputFields(OutputFieldsDeclarer declarer) {
declarer.declare(new Fields("message"));
}
}
此处需要特别注意的是,要使用
backtype.storm.topology.base.BaseBasicBolt
对象作为父类,否则不会在zk记录偏移量offset数据。
需要编写的代码已完成,接下来就是在搭建好的storm、kafka中进行测试:
# 创建topic
./bin/kafka-topics.sh --create --zookeeper zk1:2181,zk2:2281,zk3:2381 --replication-factor 1 --partitions 1 --topic msgTopic1
./bin/kafka-topics.sh --create --zookeeper zk1:2181,zk2:2281,zk3:2381 --replication-factor 1 --partitions 1 --topic msgTopic2
接下来需要分别对msgTopic1、msgTopic2启动producer(生产者)与consumer(消费者):
# 对msgTopic1启动producer,用于发送数据
./bin/kafka-console-producer.sh --broker-list dev2_55.wfj-search:9092 --topic msgTopic1
# 对msgTopic2启动consumer,用于查看发送数据的处理结果
./bin/kafka-console-consumer.sh --zookeeper zk1:2181,zk2:2281,zk3:2381 --topic msgTopic2 --from-beginning
然后将打好的jar包上传到storm的nimbus(可以使用远程上传或先上传jar包到nimbus节点所在服务器,然后本地执行):
# ./bin/storm jar topology TopicMsgTopology.jar cn.howardliu.demo.storm.kafka.topicMsg.TopicMsgTopology TopicMsgTopology
待对应的worker启动好之后,就可以在msgTopic1的producer对应终端输入数据,然后在msgTopic2的consumer对应终端查看输出结果了。
有几点需要注意的:
- 必须先创建msgTopic1、msgTopic2两个topic;
- 定义的bolt必须使用BaseBasicBolt作为父类,不能够使用BaseRichBolt,否则无法记录偏移量;
- zookeeper最好使用至少三个节点的分布式模式或伪分布式模式,否则会出现一些异常情况;
- 在整个storm下,spout、bolt的id必须唯一,否则会出现异常。
TopicMsgBolt
类作为storm.kafka.bolt.KafkaBolt
前的最后一个Bolt,需要将输出数据名称定义为message,否则KafkaBolt无法接收数据。
wordCount
简单的输入输出做完了,来点复杂点儿的场景:从某个topic定于消息,然后根据空格分词,统计单词数量,然后将当前输入的单词数量推送到另一个topic。
首先规划需要用到的类:
- 从KafkaSpout接收数据并进行处理的
backtype.storm.spout.Scheme
子类; - 数据切分bolt:
SplitSentenceBolt
; - 计数bolt:
WordCountBolt
; - 报表bolt:
ReportBolt
; - topology定义:
WordCountTopology
; - 最后再加一个原样显示订阅数据的bolt:
SentenceBolt
。
backtype.storm.spout.Scheme
子类可以使用上面已经定义过的MessageScheme
,此处不再赘述。
SplitSentenceBolt
是对输入数据进行分割,简单的使用String类的split方法,然后将每个单词命名为“word”,向后传输,代码如下:
public class SplitSentenceBolt extends BaseBasicBolt {
@Override
public void declareOutputFields(OutputFieldsDeclarer outputFieldsDeclarer) {
outputFieldsDeclarer.declare(new Fields("word"));
}
@Override
public void execute(Tuple input, BasicOutputCollector collector) {
String sentence = input.getStringByField("msg");
String[] words = sentence.split(" ");
Arrays.asList(words).forEach(word -> collector.emit(new Values(word)));
}
}
SentenceBolt
是从KafkaSpout接收数据,然后直接输出。在拓扑图上就是从输入分叉,一个进入SplitSentenceBolt
,一个进入SentenceBolt
。这种结构可以应用在Lambda架构中,代码如下:
public class SentenceBolt extends BaseBasicBolt {
private static final Logger logger = LoggerFactory.getLogger(SentenceBolt.class);
@Override
public void execute(Tuple tuple, BasicOutputCollector basicOutputCollector) {
String msg = tuple.getStringByField("msg");
logger.info("get one message is {}", msg);
basicOutputCollector.emit(new Values(msg));
}
@Override
public void declareOutputFields(OutputFieldsDeclarer outputFieldsDeclarer) {
outputFieldsDeclarer.declare(new Fields("sentence"));
}
}
WordCountBolt
是对接收到的单词进行汇总统一,然后将单词“word”及其对应数量“count”向后传输,代码如下:
public class WordCountBolt extends BaseBasicBolt {
private Map<String, Long> counts = null;
@Override
public void prepare(Map stormConf, TopologyContext context) {
this.counts = new ConcurrentHashMap<>();
super.prepare(stormConf, context);
}
@Override
public void declareOutputFields(OutputFieldsDeclarer outputFieldsDeclarer) {
outputFieldsDeclarer.declare(new Fields("word", "count"));
}
@Override
public void execute(Tuple input, BasicOutputCollector collector) {
String word = input.getStringByField("word");
Long count = this.counts.get(word);
if (count == null) {
count = 0L;
}
count++;
this.counts.put(word, count);
collector.emit(new Values(word, count));
}
}
ReportBolt
是对接收到的单词及数量进行整理,拼成json格式,然后继续向后传输,代码如下:
public class ReportBolt extends BaseBasicBolt {
@Override
public void execute(Tuple input, BasicOutputCollector collector) {
String word = input.getStringByField("word");
Long count = input.getLongByField("count");
String reportMessage = "{'word': '" + word + "', 'count': '" + count + "'}";
collector.emit(new Values(reportMessage));
}
@Override
public void declareOutputFields(OutputFieldsDeclarer outputFieldsDeclarer) {
outputFieldsDeclarer.declare(new Fields("message"));
}
}
最后是定义topology(拓扑)WordCountTopology
,代码如下:
public class WordCountTopology {
private static final String KAFKA_SPOUT_ID = "kafkaSpout";
private static final String SENTENCE_BOLT_ID = "sentenceBolt";
private static final String SPLIT_BOLT_ID = "sentenceSplitBolt";
private static final String WORD_COUNT_BOLT_ID = "sentenceWordCountBolt";
private static final String REPORT_BOLT_ID = "reportBolt";
private static final String KAFKA_BOLT_ID = "kafkabolt";
private static final String CONSUME_TOPIC = "sentenceTopic";
private static final String PRODUCT_TOPIC = "wordCountTopic";
private static final String ZK_ROOT = "/topology/root";
private static final String ZK_ID = "wordCount";
private static final String DEFAULT_TOPOLOGY_NAME = "sentenceWordCountKafka";
public static void main(String[] args) throws Exception {
// 配置Zookeeper地址
BrokerHosts brokerHosts = new ZkHosts("zk1:2181,zk2:2281,zk3:2381");
// 配置Kafka订阅的Topic,以及zookeeper中数据节点目录和名字
SpoutConfig spoutConfig = new SpoutConfig(brokerHosts, CONSUME_TOPIC, ZK_ROOT, ZK_ID);
spoutConfig.scheme = new SchemeAsMultiScheme(new MessageScheme());
TopologyBuilder builder = new TopologyBuilder();
builder.setSpout(KAFKA_SPOUT_ID, new KafkaSpout(spoutConfig));
builder.setBolt(SENTENCE_BOLT_ID, new SentenceBolt()).shuffleGrouping(KAFKA_SPOUT_ID);
builder.setBolt(SPLIT_BOLT_ID, new SplitSentenceBolt()).shuffleGrouping(KAFKA_SPOUT_ID);
builder.setBolt(WORD_COUNT_BOLT_ID, new WordCountBolt()).fieldsGrouping(SPLIT_BOLT_ID, new Fields("word"));
builder.setBolt(REPORT_BOLT_ID, new ReportBolt()).shuffleGrouping(WORD_COUNT_BOLT_ID);
builder.setBolt(KAFKA_BOLT_ID, new KafkaBolt<String, Long>()).shuffleGrouping(REPORT_BOLT_ID);
Config config = new Config();
Map<String, String> map = new HashMap<>();
map.put("metadata.broker.list", "dev2_55.wfj-search:9092");// 配置Kafka broker地址
map.put("serializer.class", "kafka.serializer.StringEncoder");// serializer.class为消息的序列化类
config.put("kafka.broker.properties", map);// 配置KafkaBolt中的kafka.broker.properties
config.put("topic", PRODUCT_TOPIC);// 配置KafkaBolt生成的topic
if (args.length == 0) {
LocalCluster cluster = new LocalCluster();
cluster.submitTopology(DEFAULT_TOPOLOGY_NAME, config, builder.createTopology());
Utils.sleep(100000);
cluster.killTopology(DEFAULT_TOPOLOGY_NAME);
cluster.shutdown();
} else {
config.setNumWorkers(1);
StormSubmitter.submitTopology(args[0], config, builder.createTopology());
}
}
}
除了上面提过应该注意的地方,此处还需要注意,
storm.kafka.SpoutConfig
定义的zkRoot与id应该与第一个例子中不同(至少保证id不同,否则两个topology将使用一个节点记录偏移量)。
基本场景是应用出现错误,发送日志到kafka的某个topic,storm订阅该topic,然后进行后续处理。场景非常简单,但是在学习过程中,遇到一个奇怪的异常情况:使用KafkaSpout读取topic数据时,没有向ZK写offset数据,致使每次都从头开始读取。纠结了两天,终于碰巧找到原因:应该使用BaseRichBolt
。
基本订阅 :
基本场景:订阅kafka的某个topic,然后在读取的消息前加上自定义的字符串,然后写回到kafka另外一个topic。 从Kafka读取数据的Spout使用storm.kafka.KafkaSpout,向Kafka写数据的Bolt使用storm.kafka.bolt.KafkaBolt。中间进行进行数据处理的Bolt定义为TopicMsgBolt。
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 |
import backtype.storm.Config; import backtype.storm.LocalCluster; import backtype.storm.StormSubmitter; import backtype.storm.spout.SchemeAsMultiScheme; import backtype.storm.topology.IBasicBolt; import backtype.storm.topology.TopologyBuilder; import backtype.storm.utils.Utils; import storm.kafka.BrokerHosts; import storm.kafka.KafkaSpout; import storm.kafka.SpoutConfig; import storm.kafka.ZkHosts; import storm.kafka.bolt.KafkaBolt; import java.util.Properties; public class TopicMsgTopology { public static void main(String[] args) throws Exception { // 配置Zookeeper地址 BrokerHosts brokerHosts = new ZkHosts( "localhost:2181" ); // 配置Kafka订阅的Topic,以及zookeeper中数据节点目录和名字 SpoutConfig spoutConfig = new SpoutConfig(brokerHosts, "msgTopic1" , "/topology/root1" , "topicMsgTopology" ); // 配置KafkaBolt中的kafka.broker.properties Config conf = new Config(); Properties props = new Properties(); // 配置Kafka broker地址 props.put( "metadata.broker.list" , "localhost:9092" ); // serializer.class为消息的序列化类 props.put( "serializer.class" , "kafka.serializer.StringEncoder" ); conf.put( "kafka.broker.properties" , props); // 配置KafkaBolt生成的topic conf.put( "topic" , "msgTopic2" ); spoutConfig.scheme = new SchemeAsMultiScheme( new MessageScheme()); TopologyBuilder builder = new TopologyBuilder(); builder.setSpout( "msgKafkaSpout" , new KafkaSpout(spoutConfig)); builder.setBolt( "msgSentenceBolt" , (IBasicBolt) new TopicMsgBolt()).shuffleGrouping( "msgKafkaSpout" ); builder.setBolt( "msgKafkaBolt" , new KafkaBolt<String, Integer>()).shuffleGrouping( "msgSentenceBolt" ); if (args.length == 0 ) { String topologyName = "kafkaTopicTopology" ; LocalCluster cluster = new LocalCluster(); cluster.submitTopology(topologyName, conf, builder.createTopology()); Utils.sleep( 100000 ); cluster.killTopology(topologyName); cluster.shutdown(); } else { conf.setNumWorkers( 1 ); StormSubmitter.submitTopology(args[ 0 ], conf, builder.createTopology()); } } } |