Hadoop如何实现关联计算

选择Hadoop，低成本和高扩展性是主要原因，但但它的开发效率实在无法让人满意。
以关联计算为例。
假设：HDFS上有2个文件，分别是客户信息和订单信息，customerID是它们之间的关联字段。如何进行关联计算，以便将客户名称添加到订单列表中？
一般方法是：输入2个源文件。根据文件名在Map中处理每条数据，如果是Order，则在foreign key上加标记”O”，形成combined key；如果是Customer则做标记”C”。Map之后的数据按照key分区，再按照combined key分组排序。最后在reduce中合并结果再输出。
实现代码：
public static class JMapper extends Mapper<LongWritable, Text, TextPair, Text> {
//mark every row with "O" or "C" according to file name
@Override
protected void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {
String pathName = ((FileSplit) context.getInputSplit()).getPath().toString();
if (pathName.contains("order.txt")) {//identify order by file name
String values[] = value.toString().split("\t");
TextPair tp = new TextPair(new Text(values[1]), new Text("O"));//mark with "O"
context.write(tp, new Text(values[0] + "\t" + values[2]));
}
if (pathName.contains("customer.txt")) {//identify customer by file name
String values[] = value.toString().split("\t");
TextPair tp = new TextPair(new Text(values[0]), new Text("C"));//mark with "C"
context.write(tp, new Text(values[1]));
}
}
}
public static class JPartitioner extends Partitioner<TextPair, Text> {
//partition by key, i.e. customerID
@Override
public int getPartition(TextPair key, Text value, int numParititon) {
return Math.abs(key.getFirst().hashCode() * 127) % numParititon;
}
}
public static class JComparator extends WritableComparator {
//group by muti-key
public JComparator() {
super(TextPair.class, true);
}
@SuppressWarnings("unchecked")
public int compare(WritableComparable a, WritableComparable b) {
TextPair t1 = (TextPair) a;
TextPair t2 = (TextPair) b;
return t1.getFirst().compareTo(t2.getFirst());
}
}
public static class JReduce extends Reducer<TextPair, Text, Text, Text> {
//merge and output
protected void reduce(TextPair key, Iterable<Text> values, Context context) throws IOException,InterruptedException {
Text pid = key.getFirst();
String desc = values.iterator().next().toString();
while (values.iterator().hasNext()) {
context.write(pid, new Text(values.iterator().next().toString() + "\t" + desc));
}
}
}
public class TextPair implements WritableComparable<TextPair> {
//make muti-key
private Text first;
private Text second;
public TextPair() {
set(new Text(), new Text());
}
public TextPair(String first, String second) {
set(new Text(first), new Text(second));
}
public TextPair(Text first, Text second) {
set(first, second);
}
public void set(Text first, Text second) {
this.first = first;
this.second = second;
}
public Text getFirst() {
return first;
}
public Text getSecond() {
return second;
}
public void write(DataOutput out) throws IOException {
first.write(out);
second.write(out);
}
public void readFields(DataInput in) throws IOException {
first.readFields(in);
second.readFields(in);
}
public int compareTo(TextPair tp) {
int cmp = first.compareTo(tp.first);
if (cmp != 0) {
return cmp;
}
return second.compareTo(tp.second);
}
}
public static void main(String agrs[]) throws IOException, InterruptedException, ClassNotFoundException {
//job entrance
Configuration conf = new Configuration();
GenericOptionsParser parser = new GenericOptionsParser(conf, agrs);
String[] otherArgs = parser.getRemainingArgs();
if (agrs.length < 3) {
System.err.println("Usage: J <in_path_one> <in_path_two> <output>");
System.exit(2);
}
Job job = new Job(conf, "J");
job.setJarByClass(J.class);//Join class
job.setMapperClass(JMapper.class);//Map class
job.setMapOutputKeyClass(TextPair.class);//Map output key class
job.setMapOutputValueClass(Text.class);//Map output value class
job.setPartitionerClass(JPartitioner.class);//partition class
job.setGroupingComparatorClass(JComparator.class);//condition group class after partition
job.setReducerClass(Example_Join_01_Reduce.class);//reduce class
job.setOutputKeyClass(Text.class);//reduce output key class
job.setOutputValueClass(Text.class);//reduce ouput value class
FileInputFormat.addInputPath(job, new Path(otherArgs[0]));//one of source files
FileInputFormat.addInputPath(job, new Path(otherArgs[1]));//another file
FileOutputFormat.setOutputPath(job, new Path(otherArgs[2]));//output path
System.exit(job.waitForCompletion(true) ? 0 : 1);//run untill job ends
}

不能直接使用原始数据，而是要搞一堆代码处理标记，并绕过MapReduce原本的架构，最后从底层设计并计算数据之间的关联关系。这还是最简单的关联计算，如果用MapReduce进行多表关联或逻辑更复杂的关联计算，复杂度会呈几何级数递增。