《hadoop权威指南》里的例子:有许多气象记录,需要找出每年的最高温度,通过secondary sort实现。(这个任务其实没有必要使用secondary sort,这里只是为了演示)
map-reduce的shuffle阶段,只会根据key进行排序,而同一个key的value是无序的,所以要把年份和温度都放在key里面。map的输出:key为year-temperature pair,value为null。job的partitioner设为根据年份进行hash(将同一年份的数据发送到同一个reducer)。job的sortComparator根据年份和温度进行比较。job的groupingComparator只根据年份进行比较。完整代码在此。
reducer的代码很简单,只有一行:将key写出即可。输出就是每个年份以及当年的最高气温。
void reduce(IntPair key, Iterable<NullWritable> values, Context context) {
context.write(key, NullWritable.get());
}
groupingComparator的作用是什么?书中对此语焉不详,只说将相同年份的数据分到一组。网上的解释也大多是错误的。
误解:reducer会根据groupingComparator对数据进行排序。略一思考就会发现这不合逻辑,如果reduce需要排序的话,那么map阶段的排序就没有必要了,反正到了reducer这里还会被打乱。
hadoop官方文档对groupingComparator的解释是:controls which keys are grouped together for a single call to Reducer.reduce()。那么到底是如何control的呢?看下面的例子。样本数据如下,第一列是key,第二列是value。
1 12 4 41 5 52 8 82 1 11 1 13 8 83 8 81 5 51 5 53 4 43 4 42
比较一下三种操作的输出:
不指定groupingComparator: 1:12,11,13, 4:41,43,42, 5:52,51,53, 8:82,83,81, 指定Less5GroupComparator,即将小于5的key分为一组,其余一组: 1:12,11,13,41,43,42, 5:52,51,53,82,83,81, 指定Sum9GroupComparator,即相加等于9的key分为一组: 1:12,11,13, 4:41,43,42,52,51,53, 8:82,83,81,
代码如下:
import java.io.IOException;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.io.WritableComparable;
import org.apache.hadoop.io.WritableComparator;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
public class GroupingComparatorExample {
public static class MyMapper
extends Mapper
{
public void map(LongWritable key, Text value, Context context)
throws IOException, InterruptedException {
String[] fields = value.toString().split(" ");
if(fields.length==2) {
int k = Integer.parseInt(fields[0]);
int v = Integer.parseInt(fields[1]);
context.write(new IntWritable(k), new IntWritable(v));
}
}
}
public static class MyReducer
extends Reducer
{
public void reduce(IntWritable key, Iterable
values, Context context)
throws IOException, InterruptedException {
System.out.print(key+":");
for(IntWritable v : values)
System.out.print(v+",");
System.out.println();
}
}
public static class Sum9GroupComparator extends WritableComparator {
protected Sum9GroupComparator() {
super(IntWritable.class, true);
}
public int compare(WritableComparable w1, WritableComparable w2) {
IntWritable i1 = (IntWritable) w1;
IntWritable i2 = (IntWritable) w2;
if(i1.get()+i2.get() == 9) return 0;
return i1.compareTo(i2);
}
}
public static class Less5GroupComparator extends WritableComparator {
protected Less5GroupComparator() {
super(IntWritable.class, true);
}
public int compare(WritableComparable w1, WritableComparable w2) {
int i1 = ((IntWritable) w1).get();
int i2 = ((IntWritable) w2).get();
if(i1<5 && i2<5) return 0;
if(i1>=5 && i2>=5) return 0;
return Integer.compare(i1, i2);
}
}
public static void main(String[] args) throws Exception {
if (args.length != 2) {
System.err.println("Usage: GroupingComparatorExample
从第一组输出可以看到:数据已经按key排序,同一个key里的value是无序的。另外可以推测,mapper使用的排序算法是稳定的,因为同一个key的values的顺序与输入相同。
第二组输出已经正确地将数据分成了两组,并且每一组的key是该组里面key的第一个值。(即key为1和5,而不是4和8)
第三组,key为4和5的已经正确地归为了一组,但是1和8呢。实际上goupingComparator并不进行任何排序操作,只是依次取出reducer收到的key-value对,然后比较当前key与前一个key,如果比较的结果为0,就认为是同一个group,对同一个group里的数据进行一次reduce调用。
org.apache.hadoop.mapreduce.Reducer的实现已经明白无误地说明了这一点:
//以下代码来自hadoop-2.4.1
//org.apache.hadoop.mapreduce.Reducer.run()主要代码如下
public void run(Context context) throws IOException, InterruptedException {
setup(context);
try {
while (context.nextKey()) {
reduce(context.getCurrentKey(), context.getValues(), context);
}
} finally {
cleanup(context);
}
}
//context.getValues()将会使用如下的迭代器进行遍历
//ReduceContextImpl.ValueIterator
protected class ValueIterator implements ReduceContext.ValueIterator
{
public boolean hasNext() {
return firstValue || nextKeyIsSame;
}
public VALUEIN next() {
// if this is the first record, we don't need to advance
if (firstValue) {
firstValue = false;
return value;
}
// if this isn't the first record and the next key is different, they
// can't advance it here.
if (!nextKeyIsSame) {
throw new NoSuchElementException("iterate past last value");
}
// otherwise, go to the next key/value pair
nextKeyValue();
return value;
}
}
//这里的comparator就是job.setGroupingComparator()时设置的
//org.apache.hadoop.mapreduce.task.ReduceContextImpl.nextKeyValue()
public boolean nextKeyValue() throws IOException, InterruptedException {
//...
hasMore = input.next();
if (hasMore) {
nextKey = input.getKey();
nextKeyIsSame = comparator.compare(currentRawKey.getBytes(), 0, currentRawKey.getLength(),
nextKey.getData(), nextKey.getPosition(), nextKey.getLength() - nextKey.getPosition()
) == 0;
} else {
nextKeyIsSame = false;
}
inputValueCounter.increment(1);
}
另外,如果不设置groupingComparator的话,使用的就是map阶段排序用的comparator。
// org.apache.hadoop.mapreduce.Job
public RawComparator getOutputValueGroupingComparator() {
Class<? extends RawComparator> theClass = getClass(
JobContext.GROUP_COMPARATOR_CLASS, null, RawComparator.class);
if (theClass == null) {
return getOutputKeyComparator();
}
return ReflectionUtils.newInstance(theClass, this);
}