Question

我正在尝试使用复合键在mapreduce中进行二次排序：

String natural-key =程序名称
Long key-for-sorting =自1970年以来以毫米为单位的时间

问题在于排序后我会根据整个复合键获得大量的reducer

通过调试，我已经验证了哈希码和比较功能是正确的。从调试记录每个块来自不同的reducer的位置，它表明分组或分区没有成功。来自调试日志：

14/12/14 00:55:12 INFO popularitweet.EtanReducer: key=the voice
14/12/14 00:55:12 INFO popularitweet.EtanReducer: the voice: Thu Dec 11 17:51:03 +0000 2014
14/12/14 00:55:12 INFO popularitweet.EtanReducer: the voice: Thu Dec 11 17:51:03 +0000 2014
14/12/14 00:55:12 INFO popularitweet.EtanReducer: key the voice ended



14/12/14 00:55:12 INFO popularitweet.EtanReducer: key=top gear
14/12/14 00:55:12 INFO popularitweet.EtanReducer: top gear: Thu Dec 11 17:51:04 +0000 2014
14/12/14 00:55:12 INFO popularitweet.EtanReducer: key top gear ended



14/12/14 00:55:12 INFO popularitweet.EtanReducer: key=american horror story
14/12/14 00:55:12 INFO popularitweet.EtanReducer: american horror story: Thu Dec 11 17:51:04 +0000 2014
14/12/14 00:55:12 INFO popularitweet.EtanReducer: key american horror story ended



14/12/14 00:55:12 INFO popularitweet.EtanReducer: key=the voice
14/12/14 00:55:12 INFO popularitweet.EtanReducer: the voice: Thu Dec 11 17:51:04 +0000 2014
14/12/14 00:55:12 INFO popularitweet.EtanReducer: key the voice ended

正如您所看到的，语音被发送到两个不同的Reducer，但时间戳不同。任何帮助，将不胜感激。复合键是以下类：

public class ProgramKey implements WritableComparable<ProgramKey> {
private String program;
private Long timestamp;

public ProgramKey() {
}

public ProgramKey(String program, Long timestamp) {
    this.program = program;
    this.timestamp = timestamp;
}

@Override
public int compareTo(ProgramKey o) {
    int result = program.compareTo(o.program);
    if (result == 0) {
        result = timestamp.compareTo(o.timestamp);
    }
    return result;
}

@Override
public void write(DataOutput dataOutput) throws IOException {
    WritableUtils.writeString(dataOutput, program);
    dataOutput.writeLong(timestamp);
}

@Override
public void readFields(DataInput dataInput) throws IOException {
    program = WritableUtils.readString(dataInput);
    timestamp = dataInput.readLong();
}

我实现的Partitioner，GroupingComparator和SortingComparator是：

public class ProgramKeyPartitioner extends Partitioner<ProgramKey, TweetObject> {

@Override
public int getPartition(ProgramKey programKey, TweetObject tweetObject, int numPartitions) {
    int hash = programKey.getProgram().hashCode();
    int partition = hash % numPartitions;
    return partition;
}

}

public class ProgramKeyGroupingComparator extends WritableComparator {
protected ProgramKeyGroupingComparator() {
    super(ProgramKey.class, true);
}

@Override
public int compare(WritableComparable a, WritableComparable b) {
    ProgramKey k1 = (ProgramKey) a;
    ProgramKey k2 = (ProgramKey) b;
    return k1.getProgram().compareTo(k2.getProgram());
}

}

public class TimeStampComparator extends WritableComparator {
protected TimeStampComparator() {
    super(ProgramKey.class, true);
}

@Override
public int compare(WritableComparable a, WritableComparable b) {
    ProgramKey ts1 = (ProgramKey)a;
    ProgramKey ts2 = (ProgramKey)a;

    int result = ts1.getProgram().compareTo(ts2.getProgram());
    if (result == 0) {
        result = ts1.getTimestamp().compareTo(ts2.getTimestamp());
    }
    return result;
}

}

    public static void main(String[] args) throws IOException,
        InterruptedException, ClassNotFoundException {



    // Create configuration
    Configuration conf = new Configuration();

    // Create job
    Job job = new Job(conf, "test1");
    job.setJarByClass(EtanMapReduce.class);

    // Set partitioner keyComparator and groupComparator
    job.setPartitionerClass(ProgramKeyPartitioner.class);
    job.setGroupingComparatorClass(ProgramKeyGroupingComparator.class);
    job.setSortComparatorClass(TimeStampComparator.class);

    // Setup MapReduce
    job.setMapperClass(EtanMapper.class);
    job.setMapOutputKeyClass(ProgramKey.class);
    job.setMapOutputValueClass(TweetObject.class);
    job.setReducerClass(EtanReducer.class);

    // Specify key / value
    job.setOutputKeyClass(Text.class);
    job.setOutputValueClass(TweetObject.class);

    // Input
    FileInputFormat.addInputPath(job, inputPath);
    job.setInputFormatClass(TextInputFormat.class);

    // Output
    FileOutputFormat.setOutputPath(job, outputDir);
    job.setOutputFormatClass(TextOutputFormat.class);

    // Delete output if exists
    FileSystem hdfs = FileSystem.get(conf);
    if (hdfs.exists(outputDir))
        hdfs.delete(outputDir, true);

    // Execute job
    logger.info("starting job");
    int code = job.waitForCompletion(true) ? 0 : 1;
    System.exit(code);

}

Answer 1

编辑...

你的TimeStampComparator似乎有一个错字......你将ts2设置为何时应该设置为b：

ProgramKey ts1 = (ProgramKey)a;
ProgramKey ts2 = (ProgramKey)a;

应该是：

ProgramKey ts1 = (ProgramKey)a;
ProgramKey ts2 = (ProgramKey)b;

这将导致键/值对的排序不正确，并使分组比较器对键/值对进行排序的假设无效。

还要检查原始程序名称是否为UTF-8，因为这是WritableUtils假设的名称。您系统的默认代码页是否也是UTF-8？

mapreduce辅助排序不起作用

1 个答案: