不能对序列文件使用hadoop getmerge，因为它会将它们合并为二进制文件，而不是序列文件（因此合并后的文件中会有很多头文件…）。
因此，您可以像@donald miner建议的那样，使用单个reducer编写一个小型hadoop作业，或者使用 SequenceFile.Reader 以及 SeuquenceFile.Writer .
我选择了第二个选项，下面是我的代码：

package ru.mail.go.webbase.markov.hadoop.utils;

import java.io.IOException;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.HBaseConfiguration;
import org.apache.hadoop.io.SequenceFile;
import org.apache.hadoop.io.Writable;
import org.apache.hadoop.util.ReflectionUtils;

public class SequenceFilesUtils {
    private static final Configuration conf = HBaseConfiguration.create();

public static <K, V> void merge(Path fromDirectory, Path toFile, Class<K> keyClass, Class<V> valueClass) throws IOException {
    FileSystem fs = FileSystem.get(conf);

    if (!fs.isDirectory(fromDirectory)) {
        throw new IllegalArgumentException("'" + fromDirectory.toString() + "' is not a directory");
    }

    SequenceFile.Writer writer = SequenceFile.createWriter(
            conf,
            SequenceFile.Writer.file(toFile),
            SequenceFile.Writer.keyClass(keyClass),
            SequenceFile.Writer.valueClass(valueClass)
            );

    for (FileStatus status : fs.listStatus(fromDirectory)) {
        if (status.isDirectory()) {
            System.out.println("Skip directory '" + status.getPath().getName() + "'");
            continue;
        }

        Path file = status.getPath();

        if (file.getName().startsWith("_")) {
            System.out.println("Skip \"_\"-file '" + file.getName() + "'"); //There are files such "_SUCCESS"-named in jobs' ouput folders 
            continue;
        }

        System.out.println("Merging '" + file.getName() + "'");

        SequenceFile.Reader reader = new SequenceFile.Reader(conf, SequenceFile.Reader.file(file));
        Writable key = (Writable) ReflectionUtils.newInstance(reader.getKeyClass(), conf);
        Writable value = (Writable) ReflectionUtils.newInstance(reader.getValueClass(), conf);

        while (reader.next(key, value)) {
            writer.append(key, value);
        }

        reader.close();
    }

    writer.close();
}
}

这是我的测试：

public class SequenceFilesUtilsTest {
private static final String OUT_PATH = "./UNIVERSE/SequenceFilesUtilsTest/";

@Before
public void initEnviroment() throws IOException {
    TestUtils.createDirectory(OUT_PATH);
    TestUtils.createDirectory(OUT_PATH + "/in");
}

@Test
public void test() throws Exception {
    Configuration conf = HBaseConfiguration.create();

    Path inPath1 = new Path("file://" + new File(OUT_PATH).getAbsolutePath() + "/in/in1.seq");
    System.out.println("Saving first part to '" + inPath1 + "'");
    SequenceFile.Writer writer1 = SequenceFile.createWriter(
            conf,
            SequenceFile.Writer.file(inPath1),
            SequenceFile.Writer.keyClass(LongWritable.class),
            SequenceFile.Writer.valueClass(Text.class)
            );
    writer1.append(new LongWritable(101), new Text("FIRST1"));
    writer1.append(new LongWritable(102), new Text("FIRST2"));
    writer1.append(new LongWritable(103), new Text("FIRST3"));
    writer1.append(new LongWritable(104), new Text("FIRST4"));
    writer1.close();

    Path inPath2 = new Path("file://" + new File(OUT_PATH).getAbsolutePath() + "/in/in2.seq");
    System.out.println("Saving second part to '" + inPath2 + "'");
    SequenceFile.Writer writer2 = SequenceFile.createWriter(
            conf,
            SequenceFile.Writer.file(inPath2),
            SequenceFile.Writer.keyClass(LongWritable.class),
            SequenceFile.Writer.valueClass(Text.class)
            );
    writer2.append(new LongWritable(201), new Text("SND1"));
    writer2.append(new LongWritable(202), new Text("SND2"));
    writer2.append(new LongWritable(203), new Text("SND3"));
    writer2.close();

    SequenceFilesUtils.merge(
            new Path("file://" + new File(OUT_PATH).getAbsolutePath() + "/in"),
            new Path("file://" + new File(OUT_PATH).getAbsolutePath() + "/merged.seq"),
            LongWritable.class,
            Text.class);

    Path mergedPath = new Path("file://" + new File(OUT_PATH).getAbsolutePath() + "/merged.seq");
    SequenceFile.Reader reader = new SequenceFile.Reader(conf, SequenceFile.Reader.file(mergedPath));
    LongWritable key = (LongWritable) ReflectionUtils.newInstance(reader.getKeyClass(), conf);
    Text value = (Text) ReflectionUtils.newInstance(reader.getValueClass(), conf);

    reader.next(key, value);
    Assert.assertEquals(101, key.get());
    Assert.assertEquals("FIRST1", value.toString());

    reader.next(key, value);
    Assert.assertEquals(102, key.get());
    Assert.assertEquals("FIRST2", value.toString());

    reader.next(key, value);
    Assert.assertEquals(103, key.get());
    Assert.assertEquals("FIRST3", value.toString());

    reader.next(key, value);
    Assert.assertEquals(104, key.get());
    Assert.assertEquals("FIRST4", value.toString());

    reader.next(key, value);
    Assert.assertEquals(201, key.get());
    Assert.assertEquals("SND1", value.toString());

    reader.next(key, value);
    Assert.assertEquals(202, key.get());
    Assert.assertEquals("SND2", value.toString());

    reader.next(key, value);
    Assert.assertEquals(203, key.get());
    Assert.assertEquals("SND3", value.toString());

    reader.close();
}
}

如果要处理大量的序列文件，我建议编写一个使用 Mapper 作为你的Map绘制者 Reducer 作为你的减速机。对于i/o格式，使用 SequenceFileInputFormat 以及 SequenceFileOutputFormat . 将减速器的数量设置为1。这些都是您在驱动程序/主代码的配置和作业对象中设置的内容。请参见如何设置输出格式、如何设置输入格式、如何设置Map器以及如何设置缩减器。
请注意 Mapper 以及 Reducer 就是对数据不做任何处理--只是传递它。这就是为什么不在这里编写map函数或reduce函数。
这将要做的是加载序列文件，对Map器中的数据不做任何处理，将所有记录洗牌到reducer，然后将它们全部输出到一个文件中。这样做的副作用是对输出序列文件中的键进行排序。

你考虑过搬家吗？我写它是为了处理某些sequencefile杂务，包括sequencefile合并。
在您的情况下，您可以运行：

forqlift seq2seq --file new_combined_file.seq  \
    original_file1.seq  original_file2.seq original_file3.seq ...

是的，福克利夫特 seq2seq 工具标记为“实验性”。。。但在我的内部测试中效果很好。

如果要将多个文件合并到单个文件中，则有两个AN：

母语

getmerge

用法： hadoop fs -getmerge <src> <localdst> 将源目录和目标文件作为输入，并将src中的文件连接到目标本地文件中。可以选择将addnl设置为允许在每个文件的末尾添加换行符。

java应用编程接口

org.apache.hadoop.fs.FileUtil.copyMerge(FileSystem srcFS, Path srcDir, FileSystem dstFS, Path dstFile, boolean deleteSource, Configuration conf, String addString);

将目录中的所有文件复制到一个输出文件（合并）

复制到hdfs

put

用法： hadoop dfs -put <localsrc> ... <dst> 将单个src或多个src从本地文件系统复制到目标文件系统。还从stdin读取输入并写入目标文件系统。

copyFromLocal

用法： hadoop dfs -copyFromLocal <localsrc> URI 与put命令类似，只是源代码仅限于本地文件引用。