Storm---DirectGroup(直接分组)
以单词分割计数为例实现Storm的DirectGroup分组:
1、Spout实现
Spout是Storm数据源头,使用DirectGroup方式将Spout数据发送指定的Bolt,需注意:
1)、Spout消费的Bolt的Task(Task应为Storm的Executor的编号),在如下代码中Spout.open()初始化中拿到消费Task
2)、需使用SpoutOutputCollector.emitDirect()方法
3)、将Spout声明为直接流,即在Spout.declareOutputFields()声明
/**
* Fixed Cycle Spout
*
* @author hanhan.zhang
* */
public class FixedCycleSpout implements IRichSpout { private String _fieldName; private boolean _direct; // stream mark
private String _streamId; private int _index; // key = msgId, value = sending tuple
private Map<String, List<Object>> _pendingTuple; // send tuple
private List<Object> [] _sendTuple; private SpoutOutputCollector _collector;
private CountMetric _sendMetric;
private CountMetric _failMetric; // consume task set
private List<Integer> _consumeTaskIdList; public FixedCycleSpout(String _streamId, String _fieldName, boolean _direct, List<Object> ... _sendTuple) {
this._streamId = _streamId;
this._fieldName = _fieldName;
this._direct = _direct;
this._sendTuple = _sendTuple;
} @Override
public void open(Map conf, TopologyContext context, SpoutOutputCollector collector) {
this._index = 0;
_pendingTuple = Maps.newHashMap(); // register metric
this._sendMetric = context.registerMetric("cycle.spout.send.tuple.metric", new CountMetric(), 60);
this._failMetric = context.registerMetric("cycle.spout.fail.tuple.metric", new CountMetric(), 60);
this._collector = collector; // get consume task id
if (this._direct) {
this._consumeTaskIdList = Lists.newLinkedList();
Map<String, Map<String, Grouping>> consumeTargets = context.getThisTargets();
if (consumeTargets != null && !consumeTargets.isEmpty()) {
// streamId = this._streamId
consumeTargets.forEach((streamId, target) -> {
if (target != null && !target.isEmpty()) {
// componentId = consume target component Id
target.forEach((componentId, group) -> {
if (group.is_set_direct()) {
this._consumeTaskIdList.addAll(context.getComponentTasks(componentId));
}
});
}
});
}
}
} @Override
public void close() { } @Override
public void activate() { } @Override
public void deactivate() { } @Override
public void nextTuple() {
this._sendMetric.incr();
if (this._index == _sendTuple.length) {
this._index = 0;
}
String msgId = UUID.randomUUID().toString();
List<Object> tuple = this._sendTuple[this._index++];
sendTuple(msgId, tuple);
} @Override
public void ack(Object msgId) {
String msgIdStr = (String) msgId;
System.out.println("ack tuple with msgId " + msgIdStr);
this._pendingTuple.remove(msgIdStr);
} @Override
public void fail(Object msgId) {
this._failMetric.incr();
String msgIdStr = (String) msgId;
System.out.println("fail tuple with msgId " + msgIdStr);
sendTuple(msgIdStr, this._pendingTuple.get(msgIdStr));
} @Override
public void declareOutputFields(OutputFieldsDeclarer declarer) {
declarer.declareStream(this._streamId, this._direct, new Fields(_fieldName));
} @Override
public Map<String, Object> getComponentConfiguration() {
return null;
} protected void sendTuple(String msgId, List<Object> tuple) {
this._pendingTuple.put(msgId, tuple);
if (this._direct) {
if (this._consumeTaskIdList == null || this._consumeTaskIdList.isEmpty()) {
throw new IllegalStateException("direct task is empty !");
}
this._consumeTaskIdList.forEach(taskId ->
this._collector.emitDirect(taskId, this._streamId, tuple, msgId));
} else {
this._collector.emit(tuple, msgId);
}
}
}
2、Bolt实现
/**
* Sentence Split Bolt
*
* @author hanhan.zhang
* */
public class SentenceSplitBolt implements IRichBolt { private OutputCollector _collector; private CountMetric _ackMetric; private CountMetric _failMetric; private String _separator; private int _taskId; private boolean _direct; private String _streamId; public SentenceSplitBolt(String _streamId, boolean _direct) {
this._streamId = _streamId;
this._direct = _direct;
} /**
* @param context
* 1: Register Metric
* 2: Next Bolt Message
* @param collector (thread-safe)
* 1: Emit Tuple
* 2: Ack/Fail Tuple
* */
public void prepare(Map stormConf, TopologyContext context, OutputCollector collector) {
this._collector = collector;
// register metric for monitor
this._ackMetric = context.registerMetric("sentence.split.ack.metric", new CountMetric(), 60);
this._failMetric = context.registerMetric("sentence.split.fail.metric", new CountMetric(), 60);
this._taskId = context.getThisTaskId(); this._separator = (String) stormConf.get(Const.SEPARATOR);
} @Override
public void execute(Tuple input) {
try {
String sentence = input.getString(0);
if (Strings.isNullOrEmpty(sentence)) {
return;
}
String []fields = sentence.split(_separator);
for (String field : fields) {
if (this._direct) {
this._collector.emitDirect(this._taskId, _streamId, input, new Values(field, 1));
} else {
this._collector.emit(this._streamId, input, new Values(field, 1));
}
}
this._collector.ack(input);
this._ackMetric.incr();
} catch (Exception e) {
this._collector.fail(input);
this._failMetric.incr();
}
} @Override
public void cleanup() { } @Override
public void declareOutputFields(OutputFieldsDeclarer declarer) {
declarer.declareStream(this._streamId, this._direct, new Fields("word", "count"));
} @Override
public Map<String, Object> getComponentConfiguration() {
return null;
}
} /**
* Word Sum Bolt
*
* @author hanhan.zhang
* */
public class WordSumBolt extends BaseRichBolt { private OutputCollector _collector; private int _taskId; private Cache<String, AtomicInteger> _wordCache; @Override
public void prepare(Map stormConf, TopologyContext context, OutputCollector collector) {
this._collector = collector;
this._taskId = context.getThisTaskId();
this._wordCache = CacheBuilder.newBuilder()
.maximumSize(1024)
.expireAfterWrite(3, TimeUnit.SECONDS)
.removalListener((removalNotification) -> {
String key = (String) removalNotification.getKey();
AtomicInteger sum = (AtomicInteger) removalNotification.getValue();
System.out.println("word sum result : [" + key + "," + sum.get() + "]");
})
.build();
} @Override
public void execute(Tuple input) {
try {
String word = input.getString(0);
int count = input.getInteger(1);
if (Strings.isEmpty(word)) {
return;
}
AtomicInteger counter = this._wordCache.getIfPresent(word);
if (counter == null) {
this._wordCache.put(word, new AtomicInteger(count));
} else {
counter.addAndGet(count);
}
this._collector.ack(input);
} catch (Exception e) {
this._collector.fail(input);
}
} @Override
public void declareOutputFields(OutputFieldsDeclarer declarer) { }
}
3、Storm运行
/**
* Tuple Split-Flow Topology
*
* @author hanhan.zhang
* */
public class FlowTopology { public static void main(String[] args) { // send tuple
List<Object> []tuple = new List[] {new Values("the cow jumped over the moon"),
new Values("the man went to the store and bought some candy"),
new Values("four score and seven years ago"),
new Values("how many apples can you eat")}; //stream name
String spoutStreamId = "topology.flow.cycle.spout.stream";
String splitStreamId = "topology.flow.split.bolt.stream"; // spout
FixedCycleSpout cycleSpout = new FixedCycleSpout(spoutStreamId, "sentence", true, tuple); // bolt
SentenceSplitBolt splitBolt = new SentenceSplitBolt(splitStreamId, false);
WordSumBolt sumBolt = new WordSumBolt(); TopologyBuilder topologyBuilder = new TopologyBuilder();
topologyBuilder.setSpout ("sentence.cycle.spout", cycleSpout, 1); topologyBuilder.setBolt("sentence.split.bolt", splitBolt, 1)
.directGrouping("sentence.cycle.spout", spoutStreamId); topologyBuilder.setBolt("word.sum.bolt", sumBolt, 3)
.fieldsGrouping("sentence.split.bolt", splitStreamId, new Fields("word")); Config config = new Config();
config.put(Const.SEPARATOR, " "); LocalCluster localCluster = new LocalCluster();
localCluster.submitTopology("flowTopology", config, topologyBuilder.createTopology()); } }
Storm---DirectGroup(直接分组)的更多相关文章
- 简单聊聊Storm的流分组策略
简单聊聊Storm的流分组策略 首先我要强调的是,Storm的分组策略对结果有着直接的影响,不同的分组的结果一定是不一样的.其次,不同的分组策略对资源的利用也是有着非常大的不同,本文主要讲一讲loca ...
- Storm Topology及分组原理
Storm的通信机制,需要满足如下一些条件以满足Storm的语义. 1.建立数据传输的缓冲区.在通信连接没有建立之前把发送的数据缓存起来.数据发送方可以在连接建立之前发送消息,而不需要等连接建立起来, ...
- Storm Grouping —— 流分组策略
Storm Grouping: Shuffle Grouping :随机分组,尽量均匀分布到下游Bolt中 将流分组定义为混排.这种混排分组意味着来自Spout的输入将混排,或随机分发给此Bolt中的 ...
- storm的流分组
用的是ShuffleGrouping分组方式,并行度设置为3 这是跑下来的结果 参考代码StormTopologyShufferGrouping.java package yehua.storm; i ...
- 大数据量场景下storm自定义分组与Hbase预分区完美结合大幅度节省内存空间
前言:在系统中向hbase中插入数据时,常常通过设置region的预分区来防止大数据量插入的热点问题,提高数据插入的效率,同时可以减少当数据猛增时由于Region split带来的资源消耗.大量的预分 ...
- storm自定义分组与Hbase预分区结合节省内存消耗
Hbas预分区 在系统中向hbase中插入数据时,常常通过设置region的预分区来防止大数据量插入的热点问题,提高数据插入的效率,同时可以减少当数据猛增时由于Region split带来的资源消耗. ...
- storm Tutorial 的解读 + 个人理解
参考链接: Tutorial storm Tutorial 中文解读+分析 导读.摘要: .hadoop有master与slave,Storm与之对应的节点是什么? .Storm控制节点上面运行一个后 ...
- [转载] 使用 Twitter Storm 处理实时的大数据
转载自http://www.ibm.com/developerworks/cn/opensource/os-twitterstorm/ 流式处理大数据简介 Storm 是一个开源的.大数据处理系统,与 ...
- Storm日志分析调研及其实时架构
1.Storm第一个Demo 2.Windows下基于eclipse的Storm应用开发与调试 3.Storm实例+mysql数据库保存 4.Storm原理介绍 5. flume+kafka+stor ...
- Storm知识点
1. 离线计算是什么? 离线计算:批量获取数据.批量传输数据.周期性批量计算数据.数据展示 代表技术:Sqoop批量导入数据.HDFS批量存储数据.MapReduce批量计算数据.Hive批量计算数据 ...
随机推荐
- 解决Windows10与Ubuntu系统时间不一致问题
前言: 安装完windows与Ubuntu双系统之后会发现windows与Ubuntu时间不一致.这是硬件时间都一样的情况下,Ubuntu使用的是UST,Windows使用的是CST.要解决该问题就要 ...
- ubuntu下dpkg/apt的一些常用命令
apt-get update 更新源(硬盘缓存的包的版本等)apt-get upgrade 更新包(更新系统中已安装的包,依据缓存包的版本更新)apt-get dist-upgrade 升级系统apt ...
- python 字符集转换-灰常慢
代码 def toUni (text): str = text try: charstyle = chardet.detect(text) # print 'confidence: ', charst ...
- 洛谷——P1748 H数
P1748 H数 题目背景 无 题目描述 所谓H数,是指只含有2,3,5,7这些质因数的数,如630是H数,而22不是.现在要求输出第n个H数,为了方便起见将H[1]定为1.已知n不超过10000,最 ...
- Codeforces 538 F. A Heap of Heaps
\(>Codeforces \space 538 F. A Heap of Heaps<\) 题目大意 :给出 \(n\) 个点,编号为 \(1 - n\) ,每个点有点权,将这些点构建成 ...
- JZYZOJ1524 [haoi2012]外星人 欧拉函数
http://172.20.6.3/Problem_Show.asp?id=1524 大概可以算一个结论吧,欧拉函数在迭代的时候,每次迭代之后消去一个2,每个非2的质因子迭代一次又(相当于)生成一个2 ...
- BZOJ 4260 Codechef REBXOR(字典树)
[题目链接] http://www.lydsy.com/JudgeOnline/problem.php?id=4260 [题目大意] 给出一个数列,请找出两段连续且不相交的数段,使得其分别异或和的和 ...
- 【线段树】POJ3225-Help with Intervals
---恢复内容开始--- [题目大意] (直接引用ACM神犇概括,貌似是notonlysucess?) U:把区间[l,r]覆盖成1 I:把[-∞,l)(r,∞]覆盖成0 D:把区间[l,r]覆盖成0 ...
- maven打包出现找不到java.lang包的问题
修改pom文件中的bootclasspth路径为${java.home}/lib/rt.jar;${java.home}/lib/jce.jar windows为分号 linux为冒号
- Java高级架构师(一)第34节:Nginx的Http模块部分的指令
默认长链接的数目在100个 默认长链接的超时时间,一般在75S.