大数据处理框架之Strom:容错机制

1、集群节点宕机
Nimbus服务器
　　单点故障，大部分时间是闲置的，在supervisor挂掉时会影响，所以宕机影响不大，重启即可
非Nimbus服务器
　　故障时，该节点上所有Task任务都会超时，Nimbus会将这些Task任务重新分配到其他服务器上运行

2、进程挂掉
Worker
　　挂掉时，Supervisor会重新启动这个进程。如果启动过程中仍然一直失败，并且无法向Nimbus发送心跳，Nimbus会将该Worker重新分配到其他服务器上
Supervisor
　　无状态（所有的状态信息都存放在Zookeeper中来管理）
　　快速失败（每当遇到任何异常情况，都会自动毁灭）
Nimbus
　　无状态（所有的状态信息都存放在Zookeeper中来管理）
　　快速失败（每当遇到任何异常情况，都会自动毁灭）

3、消息的完整性
从Spout中发出的Tuple，以及基于他所产生Tuple，由这些消息就构成了一棵tuple树，当这棵tuple树发送完成，并且树当中每一条消息都被正确处理，就表明spout发送消息被“完整处理”，即消息的完整性，storm使用Acker确保消息完整性，Acker是拓扑当中特殊的一些任务，负责跟踪每个Spout发出的Tuple的DAG（有向无环图）
Acker分为ack确认机制和fail失败处理机制，Spout作为数据源，当拓扑中bolt处理失败时该怎么办？Acker机制可以重发数据到bolt进行重新处理。

看下面的例子：

MessageSpout  ---->   split-bolt  ---->    write-bolt

MessageTopology

package bhz.topology;

import backtype.storm.Config;
import backtype.storm.LocalCluster;
import backtype.storm.topology.TopologyBuilder;
import bhz.bolt.SpliterBolt;
import bhz.bolt.WriterBolt;
import bhz.spout.MessageSpout;

public class MessageTopology {

    public static void main(String[] args) throws Exception {
        TopologyBuilder builder = new TopologyBuilder();
        builder.setSpout("spout", new MessageSpout());
        builder.setBolt("split-bolt", new SpliterBolt()).shuffleGrouping("spout");
        builder.setBolt("write-bolt", new WriterBolt()).shuffleGrouping("split-bolt");
        //本地配置
        Config config = new Config();
        config.setDebug(false);
        LocalCluster cluster = new LocalCluster();
        System.out.println(cluster);
        cluster.submitTopology("message", config, builder.createTopology());
        Thread.sleep(10000);
        cluster.killTopology("message");
        cluster.shutdown();
    }
}

MessageSpout

package bhz.spout;

import java.util.Map;

import backtype.storm.spout.SpoutOutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.IRichSpout;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.tuple.Fields;
import backtype.storm.tuple.Values;

public class MessageSpout implements IRichSpout {

    private static final long serialVersionUID = 1L;

    private int index = 0;

    private String[] subjects = new String[]{
            "groovy,oeacnbase",
            "openfire,restful",
            "flume,activiti",
            "hadoop,hbase",
            "spark,sqoop"
    };

    private SpoutOutputCollector collector;

    @Override
    public void open(Map conf, TopologyContext context, SpoutOutputCollector collector) {
        this.collector = collector;
    }

    @Override
    public void nextTuple() {
        if(index < subjects.length){
            String sub = subjects[index];
            //发送信息参数1 为数值， 参数2为msgId
            collector.emit(new Values(sub), index);
            index++;
        }
    }

    @Override
    public void declareOutputFields(OutputFieldsDeclarer declarer) {
        declarer.declare(new Fields("subjects"));
    }
    //当bolt 处理成功  ack确认 spout执行ack方法
    @Override
    public void ack(Object msgId) {
        System.out.println("【消息发送成功!!!】 (msgId = " + msgId +")");
    }
    //当bolt处理失败时，spout调用fail方法，进行重发处理
    @Override
    public void fail(Object msgId) {
        System.out.println("【消息发送失败!!!】  (msgId = " + msgId +")");
        System.out.println("【重发进行中...】");
        collector.emit(new Values(subjects[(Integer) msgId]), msgId);
        System.out.println("【重发成功!!!】");
    }

    @Override
    public void close() {

    }

    @Override
    public void activate() {

    }

    @Override
    public void deactivate() {

    }

    @Override
    public Map<String, Object> getComponentConfiguration() {
        return null;
    }

}

SpliterBolt

package bhz.bolt;

import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;

import backtype.storm.task.OutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.IRichBolt;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.tuple.Fields;
import backtype.storm.tuple.Tuple;
import backtype.storm.tuple.Values;

public class SpliterBolt implements IRichBolt {

    private static final long serialVersionUID = 1L;

    private OutputCollector collector;

    @Override
    public void prepare(Map config, TopologyContext context, OutputCollector collector) {
        this.collector = collector;
    }

    private boolean flag = false;

    @Override
    public void execute(Tuple tuple) {
        try {
            String subjects = tuple.getStringByField("subjects");

            if(!flag && subjects.equals("flume,activiti")){
                flag = true;
                int a = 1/0;
            }

            String[] words = subjects.split(",");
            //List<String> list = new ArrayList<String>();
            //int index = 0;
            for (String word : words) {
                //注意这里循环发送消息，要携带tuple对象，用于处理异常时重发策略
                collector.emit(tuple, new Values(word));
                //list.add(word);
                //index ++;
            }
            //collector.emit(tuple, new Values(list));
            collector.ack(tuple);//通知spout处理成功
        } catch (Exception e) {
            e.printStackTrace();
            collector.fail(tuple);//通知spout 处理失败
        }
    }

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

    @Override
    public void cleanup() {

    }

    @Override
    public Map<String, Object> getComponentConfiguration() {
        return null;
    }

}

WriterBolt

package bhz.bolt;

import java.io.FileWriter;
import java.io.IOException;
import java.util.List;
import java.util.Map;

import backtype.storm.task.OutputCollector;
import backtype.storm.task.TopologyContext;
import backtype.storm.topology.IRichBolt;
import backtype.storm.topology.OutputFieldsDeclarer;
import backtype.storm.tuple.Tuple;
import backtype.storm.tuple.Values;

public class WriterBolt implements IRichBolt {

    private static final long serialVersionUID = 1L;

    private FileWriter writer;

    private OutputCollector collector;

    @Override
    public void prepare(Map config, TopologyContext context, OutputCollector collector) {
        this.collector = collector;
        try {
            writer = new FileWriter("d://message.txt");
        } catch (IOException e) {
            e.printStackTrace();
        }
    }

    private boolean flag = false;

    @Override
    public void execute(Tuple tuple) {
        String word = tuple.getString(0);
//        List<String> list = (List<String>)tuple.getValueByField("word");
//        System.out.println("======================" + list);
        try {
            if(!flag && word.equals("hadoop")){
                flag = true;
                int a = 1/0;
            }
            writer.write(word);
            writer.write("\r\n");
            writer.flush();
        } catch (Exception e) {
            e.printStackTrace();
            collector.fail(tuple);//通知spout处理失败
        }
        collector.emit(tuple, new Values(word));
        collector.ack(tuple);//通知spout处理成功
    }

    @Override
    public void cleanup() {

    }

    @Override
    public void declareOutputFields(OutputFieldsDeclarer declarer) {

    }

    @Override
    public Map<String, Object> getComponentConfiguration() {
        return null;
    }

}

spout重发机制会带来一个问题：数据重复消费，看上面的例子当WriterBolt执行失败的时候，spout 将hadoop,hbase重发，那么hbase会被WriterBolt再执行一次，目前storm对此没有保障机制，按照业务设计的通用做法就是使用幂等性（比如使用唯一性ID），防止重复消费数据。