Hadoop MapReduce: context.write changes values

Question

我是Hadoop的新手，正在编写MapReduce作业，我遇到了一个问题，它似乎reducers context.write方法正在将正确的值更改为不正确的值。

MapReduce作业应该做什么？

• 计算总字数(int wordCount)
• 计算不同单词的数量(int counter_dist)
• 计算以"z"或"Z"开头的单词数(int counter_startZ)
• 计算出现少于 4 次的单词数(int counter_less4)

所有这些都必须在单个MapReduce作业中完成。

正在分析的文本文件

Hello how zou zou zou zou how are you

正确输出：
wordCount = 9
counter_dist = 5
counter_startZ = 4counter_less4 = 4

映射器类

public class WordCountMapper extends Mapper <Object, Text, Text, IntWritable> {
private final static IntWritable one = new IntWritable(1);
private Text word = new Text();
@Override
public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
StringTokenizer itr = new StringTokenizer(value.toString());
while (itr.hasMoreTokens()) {
String hasKey = itr.nextToken();
word.set(hasKey);
context.write(word, one);
}
}
}

化简类
为了调试我的代码，我打印了很多语句来检查每个点的值。标准输出代码如下。

public class WordCountReducer extends Reducer <Text, IntWritable, Text, IntWritable> {
int wordCount = 0; // Total number of words
int counter_dist = 0; // Number of distinct words in the corpus
int counter_startZ = 0; // Number of words that start with letter Z
int counter_less4 = 0; // Number of words that appear less than 4 
@Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException {
int repeatedWords = 0;
System.out.println("###Reduce method starts");
System.out.println("Values: wordCount:" + wordCount + " counter_dist:" + counter_dist + " counter_startZ:" + counter_startZ + " counter_less4:" + counter_less4 + " (start)");
for (IntWritable val : values){
System.out.println("Key: " + key.toString());
repeatedWords++;
wordCount += val.get();
if(key.toString().startsWith("z") || key.toString().startsWith("Z")){
counter_startZ++;
}
System.out.println("Values: wordCount:" + wordCount + " counter_dist:" + counter_dist + " counter_startZ:" + counter_startZ + " counter_less4:" + counter_less4 + " (end of loop)");
}
counter_dist++;
if(repeatedWords < 4){
counter_less4++;
}
System.out.println("Values: wordCount:" + wordCount + " counter_dist:" + counter_dist + " counter_startZ:" + counter_startZ + " counter_less4:" + counter_less4 + " repeatedWords:" + repeatedWords + " (end)");
System.out.println("###Reduce method endsn");
}

@Override
public void cleanup(Context context) throws IOException, InterruptedException{
System.out.println("###CLEANUP: wordCount: " + wordCount);
System.out.println("###CLEANUP: counter_dist: " + counter_dist);
System.out.println("###CLEANUP: counter_startZ: " + counter_startZ);
System.out.println("###CLEANUP: counter_less4: " + counter_less4);
context.write(new Text("Total words: "), new IntWritable(wordCount));
context.write(new Text("Distinct words: "), new IntWritable(counter_dist));
context.write(new Text("Starts with Z: "), new IntWritable(counter_startZ));
context.write(new Text("Appears less than 4 times:"), new IntWritable(counter_less4));
}

}

标准日志，我用于调试

###Reduce method starts
Values: wordCount:0 counter_dist:0 counter_startZ:0 counter_less4:0 (start)
Key: Hello
Values: wordCount:1 counter_dist:0 counter_startZ:0 counter_less4:0 (end of loop)
Values: wordCount:1 counter_dist:1 counter_startZ:0 counter_less4:1 repeatedWords:1 (end)
###Reduce method ends
###Reduce method starts
Values: wordCount:1 counter_dist:1 counter_startZ:0 counter_less4:1 (start)
Key: are
Values: wordCount:2 counter_dist:1 counter_startZ:0 counter_less4:1 (end of loop)
Values: wordCount:2 counter_dist:2 counter_startZ:0 counter_less4:2 repeatedWords:1 (end)
###Reduce method ends
###Reduce method starts
Values: wordCount:2 counter_dist:2 counter_startZ:0 counter_less4:2 (start)
Key: how
Values: wordCount:3 counter_dist:2 counter_startZ:0 counter_less4:2 (end of loop)
Key: how
Values: wordCount:4 counter_dist:2 counter_startZ:0 counter_less4:2 (end of loop)
Values: wordCount:4 counter_dist:3 counter_startZ:0 counter_less4:3 repeatedWords:2 (end)
###Reduce method ends
###Reduce method starts
Values: wordCount:4 counter_dist:3 counter_startZ:0 counter_less4:3 (start)
Key: you
Values: wordCount:5 counter_dist:3 counter_startZ:0 counter_less4:3 (end of loop)
Values: wordCount:5 counter_dist:4 counter_startZ:0 counter_less4:4 repeatedWords:1 (end)
###Reduce method ends
###Reduce method starts
Values: wordCount:5 counter_dist:4 counter_startZ:0 counter_less4:4 (start)
Key: zou
Values: wordCount:6 counter_dist:4 counter_startZ:1 counter_less4:4 (end of loop)
Key: zou
Values: wordCount:7 counter_dist:4 counter_startZ:2 counter_less4:4 (end of loop)
Key: zou
Values: wordCount:8 counter_dist:4 counter_startZ:3 counter_less4:4 (end of loop)
Key: zou
Values: wordCount:9 counter_dist:4 counter_startZ:4 counter_less4:4 (end of loop)
Values: wordCount:9 counter_dist:5 counter_startZ:4 counter_less4:4 repeatedWords:4 (end)
###Reduce method ends
###CLEANUP: wordCount: 9
###CLEANUP: counter_dist: 5
###CLEANUP: counter_startZ: 4
###CLEANUP: counter_less4: 4

从日志来看，所有值似乎都是正确的，并且一切正常。但是，当我在HDFS中打开输出目录并读取"part-r-00000"文件时，写入其中的context.write的输出完全不同。

Total words: 22
Distinct words: 4
Starts with Z: 0
Appears less than 4 times: 4

Answer 1

您永远不能依赖cleanup()方法来实现关键的程序逻辑。每次 JVM 被扯掉时都会调用cleanup()方法。因此，根据生成和杀死的 JVM(您无法预测)的数量，您的逻辑变得不稳定。

将initialization和写入上下文都移动到reduce方法中。

即

int wordCount = 0; // Total number of words
int counter_dist = 0; // Number of distinct words in the corpus
int counter_startZ = 0; // Number of words that start with letter Z
int counter_less4 = 0; // Number of words that appear less than 4 

和

context.write(new Text("Total words: "), new IntWritable(wordCount));
context.write(new Text("Distinct words: "), new IntWritable(counter_dist));
context.write(new Text("Starts with Z: "), new IntWritable(counter_startZ));
context.write(new Text("Appears less than 4 times:"), new IntWritable(counter_less4));

---

编辑：根据OP评论，似乎整个逻辑是有缺陷的。

下面是实现所需结果的代码。请注意，我没有实现setup()或cleanup();因为根本不需要。

使用计数器计算您要查找的内容。MapReduce完成后，只需获取驱动程序类中的计数器即可。

例如，在映射器中可以计算单词数和以"z"或"Z"开头的单词数

public class WordCountMapper extends Mapper <Object, Text, Text, IntWritable> {
private final static IntWritable one = new IntWritable(1);
private Text word = new Text();
@Override
public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
StringTokenizer itr = new StringTokenizer(value.toString());
while (itr.hasMoreTokens()) {
String hasKey = itr.nextToken();
word.set(hasKey);
context.getCounter("my_counters", "TOTAL_WORDS").increment(1);
if(hasKey.toUpperCase().startsWith("Z")){
context.getCounter("my_counters", "Z_WORDS").increment(1);
}
context.write(word, one);
}
}
}

不同单词和words appearing less than 4 times的数量可以在减速器计数器中计数。

public class WordCountReducer extends Reducer <Text, IntWritable, Text, IntWritable> {
@Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException {
int wordCount= 0;
context.getCounter("my_counters", "DISTINCT_WORDS").increment(1);
for (IntWritable val : values){
wordCount += val.get();
}
if(wordCount < 4{
context.getCounter("my_counters", "WORDS_LESS_THAN_4").increment(1);
}
}
}

在驱动程序类中获取计数器。以下代码位于您提交作业的行之后

CounterGroup group = job.getCounters().getGroup("my_counters");
for (Counter counter : group) {
System.out.println(counter.getName() + "=" + counter.getValue());
}