Java InputFormat.createRecordReader方法代码示例

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test
public void testReinit() throws Exception {
  // Test that a split containing multiple files works correctly,
  // with the child RecordReader getting its initialize() method
  // called a second time.
  TaskAttemptID taskId = new TaskAttemptID("jt", 0, TaskType.MAP, 0, 0);
  Configuration conf = new Configuration();
  TaskAttemptContext context = new TaskAttemptContextImpl(conf, taskId);

  // This will create a CombineFileRecordReader that itself contains a
  // DummyRecordReader.
  InputFormat inputFormat = new ChildRRInputFormat();

  Path [] files = { new Path("file1"), new Path("file2") };
  long [] lengths = { 1, 1 };

  CombineFileSplit split = new CombineFileSplit(files, lengths);
  RecordReader rr = inputFormat.createRecordReader(split, context);
  assertTrue("Unexpected RR type!", rr instanceof CombineFileRecordReader);

  // first initialize() call comes from MapTask. We'll do it here.
  rr.initialize(split, context);

  // First value is first filename.
  assertTrue(rr.nextKeyValue());
  assertEquals("file1", rr.getCurrentValue().toString());

  // The inner RR will return false, because it only emits one (k, v) pair.
  // But there's another sub-split to process. This returns true to us.
  assertTrue(rr.nextKeyValue());
  
  // And the 2nd rr will have its initialize method called correctly.
  assertEquals("file2", rr.getCurrentValue().toString());
  
  // But after both child RR's have returned their singleton (k, v), this
  // should also return false.
  assertFalse(rr.nextKeyValue());
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
private static List<Text> readSplit(InputFormat<LongWritable,Text> format,
  InputSplit split, Job job) throws IOException, InterruptedException {
  List<Text> result = new ArrayList<Text>();
  Configuration conf = job.getConfiguration();
  TaskAttemptContext context = MapReduceTestUtil.
    createDummyMapTaskAttemptContext(conf);
  RecordReader<LongWritable, Text> reader = format.createRecordReader(split,
    MapReduceTestUtil.createDummyMapTaskAttemptContext(conf));
  MapContext<LongWritable,Text,LongWritable,Text> mcontext =
    new MapContextImpl<LongWritable,Text,LongWritable,Text>(conf,
    context.getTaskAttemptID(), reader, null, null,
    MapReduceTestUtil.createDummyReporter(),
    split);
  reader.initialize(split, mcontext);
  while (reader.nextKeyValue()) {
    result.add(new Text(reader.getCurrentValue()));
  }
  return result;
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
private int countRecords(int numSplits) 
    throws IOException, InterruptedException {
  InputFormat<Text, BytesWritable> format =
    new SequenceFileInputFilter<Text, BytesWritable>();
  if (numSplits == 0) {
    numSplits =
      random.nextInt(MAX_LENGTH / (SequenceFile.SYNC_INTERVAL / 20)) + 1;
  }
  FileInputFormat.setMaxInputSplitSize(job, 
    fs.getFileStatus(inFile).getLen() / numSplits);
  TaskAttemptContext context = MapReduceTestUtil.
    createDummyMapTaskAttemptContext(job.getConfiguration());
  // check each split
  int count = 0;
  for (InputSplit split : format.getSplits(job)) {
    RecordReader<Text, BytesWritable> reader =
      format.createRecordReader(split, context);
    MapContext<Text, BytesWritable, Text, BytesWritable> mcontext = 
      new MapContextImpl<Text, BytesWritable, Text, BytesWritable>(
      job.getConfiguration(), 
      context.getTaskAttemptID(), reader, null, null, 
      MapReduceTestUtil.createDummyReporter(), split);
    reader.initialize(split, mcontext);
    try {
      while (reader.nextKeyValue()) {
        LOG.info("Accept record " + reader.getCurrentKey().toString());
        count++;
      }
    } finally {
      reader.close();
    }
  }
  return count;
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test
public void testRecordReaderInit() throws InterruptedException, IOException {
  // Test that we properly initialize the child recordreader when
  // CombineFileInputFormat and CombineFileRecordReader are used.

  TaskAttemptID taskId = new TaskAttemptID("jt", 0, TaskType.MAP, 0, 0);
  Configuration conf1 = new Configuration();
  conf1.set(DUMMY_KEY, "STATE1");
  TaskAttemptContext context1 = new TaskAttemptContextImpl(conf1, taskId);

  // This will create a CombineFileRecordReader that itself contains a
  // DummyRecordReader.
  InputFormat inputFormat = new ChildRRInputFormat();

  Path [] files = { new Path("file1") };
  long [] lengths = { 1 };

  CombineFileSplit split = new CombineFileSplit(files, lengths);

  RecordReader rr = inputFormat.createRecordReader(split, context1);
  assertTrue("Unexpected RR type!", rr instanceof CombineFileRecordReader);

  // Verify that the initial configuration is the one being used.
  // Right after construction the dummy key should have value "STATE1"
  assertEquals("Invalid initial dummy key value", "STATE1",
    rr.getCurrentKey().toString());

  // Switch the active context for the RecordReader...
  Configuration conf2 = new Configuration();
  conf2.set(DUMMY_KEY, "STATE2");
  TaskAttemptContext context2 = new TaskAttemptContextImpl(conf2, taskId);
  rr.initialize(split, context2);

  // And verify that the new context is updated into the child record reader.
  assertEquals("Invalid secondary dummy key value", "STATE2",
    rr.getCurrentKey().toString());
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
/**
 * From each split sampled, take the first numSamples / numSplits records.
 */
@SuppressWarnings("unchecked") // ArrayList::toArray doesn't preserve type
public K[] getSample(InputFormat<K,V> inf, Job job) 
    throws IOException, InterruptedException {
  List<InputSplit> splits = inf.getSplits(job);
  ArrayList<K> samples = new ArrayList<K>(numSamples);
  int splitsToSample = Math.min(maxSplitsSampled, splits.size());
  int samplesPerSplit = numSamples / splitsToSample;
  long records = 0;
  for (int i = 0; i < splitsToSample; ++i) {
    TaskAttemptContext samplingContext = new TaskAttemptContextImpl(
        job.getConfiguration(), new TaskAttemptID());
    RecordReader<K,V> reader = inf.createRecordReader(
        splits.get(i), samplingContext);
    reader.initialize(splits.get(i), samplingContext);
    while (reader.nextKeyValue()) {
      samples.add(ReflectionUtils.copy(job.getConfiguration(),
                                       reader.getCurrentKey(), null));
      ++records;
      if ((i+1) * samplesPerSplit <= records) {
        break;
      }
    }
    reader.close();
  }
  return (K[])samples.toArray();
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
/**
 * Constructs the DelegatingRecordReader.
 * 
 * @param split TaggegInputSplit object
 * @param context TaskAttemptContext object
 *  
 * @throws IOException
 * @throws InterruptedException
 */
@SuppressWarnings("unchecked")
public DelegatingRecordReader(InputSplit split, TaskAttemptContext context)
    throws IOException, InterruptedException {
  // Find the InputFormat and then the RecordReader from the
  // TaggedInputSplit.
  TaggedInputSplit taggedInputSplit = (TaggedInputSplit) split;
  InputFormat<K, V> inputFormat = (InputFormat<K, V>) ReflectionUtils
      .newInstance(taggedInputSplit.getInputFormatClass(), context
          .getConfiguration());
  originalRR = inputFormat.createRecordReader(taggedInputSplit
      .getInputSplit(), context);
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Override
public void initialize(final InputSplit inputSplit, final TaskAttemptContext taskAttemptContext) throws IOException, InterruptedException {
    final Configuration configuration = taskAttemptContext.getConfiguration();
    final InputFormat<NullWritable, VertexWritable> inputFormat = ReflectionUtils.newInstance(configuration.getClass(Constants.GREMLIN_HADOOP_GRAPH_READER, InputFormat.class, InputFormat.class), configuration);
    if (!(inputFormat instanceof GraphFilterAware) && configuration.get(Constants.GREMLIN_HADOOP_GRAPH_FILTER, null) != null)
        this.graphFilter = VertexProgramHelper.deserialize(ConfUtil.makeApacheConfiguration(configuration), Constants.GREMLIN_HADOOP_GRAPH_FILTER);
    this.recordReader = inputFormat.createRecordReader(inputSplit, taskAttemptContext);
    this.recordReader.initialize(inputSplit, taskAttemptContext);
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
/**
 * For each split sampled, emit when the ratio of the number of records
 * retained to the total record count is less than the specified
 * frequency.
 */
@SuppressWarnings("unchecked") // ArrayList::toArray doesn't preserve type
public K[] getSample(InputFormat<K,V> inf, Job job) 
    throws IOException, InterruptedException {
  List<InputSplit> splits = inf.getSplits(job);
  ArrayList<K> samples = new ArrayList<K>();
  int splitsToSample = Math.min(maxSplitsSampled, splits.size());
  long records = 0;
  long kept = 0;
  for (int i = 0; i < splitsToSample; ++i) {
    TaskAttemptContext samplingContext = new TaskAttemptContextImpl(
        job.getConfiguration(), new TaskAttemptID());
    RecordReader<K,V> reader = inf.createRecordReader(
        splits.get(i), samplingContext);
    reader.initialize(splits.get(i), samplingContext);
    while (reader.nextKeyValue()) {
      ++records;
      if ((double) kept / records < freq) {
        samples.add(ReflectionUtils.copy(job.getConfiguration(),
                             reader.getCurrentKey(), null));
        ++kept;
      }
    }
    reader.close();
  }
  return (K[])samples.toArray();
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test(enabled = true, dependsOnMethods = { "testWritePersonData" })
public void testReadPersonData() throws Exception {
  long sumage = 0L;
  long reccnt = 0L;
  File folder = new File(m_workdir.toString());
  File[] listfiles = folder.listFiles();
  for (int idx = 0; idx < listfiles.length; ++idx) {
    if (listfiles[idx].isFile()
        && listfiles[idx].getName().startsWith(MneConfigHelper.getBaseOutputName(m_conf, null))
        && listfiles[idx].getName().endsWith(MneConfigHelper.DEFAULT_FILE_EXTENSION)) {
      System.out.println(String.format("Verifying : %s", listfiles[idx].getName()));
      FileSplit split = new FileSplit(
          new Path(m_workdir, listfiles[idx].getName()), 0, 0L, new String[0]);
      InputFormat<NullWritable, MneDurableInputValue<Person<Long>>> inputFormat =
          new MneInputFormat<MneDurableInputValue<Person<Long>>, Person<Long>>();
      RecordReader<NullWritable, MneDurableInputValue<Person<Long>>> reader =
          inputFormat.createRecordReader(split, m_tacontext);
      MneDurableInputValue<Person<Long>> personval = null;
      while (reader.nextKeyValue()) {
        personval = reader.getCurrentValue();
        AssertJUnit.assertTrue(personval.getValue().getAge() < 51);
        sumage += personval.getValue().getAge();
        ++reccnt;
      }
      reader.close();
    }
  }
  AssertJUnit.assertEquals(m_reccnt, reccnt);
  AssertJUnit.assertEquals(m_sumage, sumage);
  System.out.println(String.format("The checksum of ages is %d", sumage));
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test(enabled = true, dependsOnMethods = { "testWriteLongData" })
public void testReadLongData() throws Exception {
  long sum = 0L;
  long reccnt = 0L;
  File folder = new File(m_workdir.toString());
  File[] listfiles = folder.listFiles();
  for (int idx = 0; idx < listfiles.length; ++idx) {
    if (listfiles[idx].isFile()
        && listfiles[idx].getName().startsWith(MneConfigHelper.getBaseOutputName(m_conf, null))
        && listfiles[idx].getName().endsWith(MneConfigHelper.DEFAULT_FILE_EXTENSION)) {
      System.out.println(String.format("Verifying : %s", listfiles[idx].getName()));
      FileSplit split = new FileSplit(
          new Path(m_workdir, listfiles[idx].getName()), 0, 0L, new String[0]);
      InputFormat<NullWritable, MneDurableInputValue<Long>> inputFormat =
          new MneInputFormat<MneDurableInputValue<Long>, Long>();
      RecordReader<NullWritable, MneDurableInputValue<Long>> reader =
          inputFormat.createRecordReader(split, m_tacontext);
      MneDurableInputValue<Long> mdval = null;
      while (reader.nextKeyValue()) {
        mdval = reader.getCurrentValue();
        sum += mdval.getValue();
        ++reccnt;
      }
      reader.close();
    }
  }
  AssertJUnit.assertEquals(m_sum, sum);
  AssertJUnit.assertEquals(m_reccnt, reccnt);
  System.out.println(String.format("The checksum of long data is %d", sum));
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
public void testBinary() throws IOException, InterruptedException {
  Job job = Job.getInstance();
  FileSystem fs = FileSystem.getLocal(job.getConfiguration());
  Path dir = new Path(System.getProperty("test.build.data",".") + "/mapred");
  Path file = new Path(dir, "testbinary.seq");
  Random r = new Random();
  long seed = r.nextLong();
  r.setSeed(seed);

  fs.delete(dir, true);
  FileInputFormat.setInputPaths(job, dir);

  Text tkey = new Text();
  Text tval = new Text();

  SequenceFile.Writer writer = new SequenceFile.Writer(fs,
    job.getConfiguration(), file, Text.class, Text.class);
  try {
    for (int i = 0; i < RECORDS; ++i) {
      tkey.set(Integer.toString(r.nextInt(), 36));
      tval.set(Long.toString(r.nextLong(), 36));
      writer.append(tkey, tval);
    }
  } finally {
    writer.close();
  }
  TaskAttemptContext context = MapReduceTestUtil.
    createDummyMapTaskAttemptContext(job.getConfiguration());
  InputFormat<BytesWritable,BytesWritable> bformat =
    new SequenceFileAsBinaryInputFormat();

  int count = 0;
  r.setSeed(seed);
  BytesWritable bkey = new BytesWritable();
  BytesWritable bval = new BytesWritable();
  Text cmpkey = new Text();
  Text cmpval = new Text();
  DataInputBuffer buf = new DataInputBuffer();
  FileInputFormat.setInputPaths(job, file);
  for (InputSplit split : bformat.getSplits(job)) {
    RecordReader<BytesWritable, BytesWritable> reader =
          bformat.createRecordReader(split, context);
    MapContext<BytesWritable, BytesWritable, BytesWritable, BytesWritable> 
      mcontext = new MapContextImpl<BytesWritable, BytesWritable,
        BytesWritable, BytesWritable>(job.getConfiguration(), 
        context.getTaskAttemptID(), reader, null, null, 
        MapReduceTestUtil.createDummyReporter(), 
        split);
    reader.initialize(split, mcontext);
    try {
      while (reader.nextKeyValue()) {
        bkey = reader.getCurrentKey();
        bval = reader.getCurrentValue();
        tkey.set(Integer.toString(r.nextInt(), 36));
        tval.set(Long.toString(r.nextLong(), 36));
        buf.reset(bkey.getBytes(), bkey.getLength());
        cmpkey.readFields(buf);
        buf.reset(bval.getBytes(), bval.getLength());
        cmpval.readFields(buf);
        assertTrue(
          "Keys don't match: " + "*" + cmpkey.toString() + ":" +
          tkey.toString() + "*",
          cmpkey.toString().equals(tkey.toString()));
        assertTrue(
          "Vals don't match: " + "*" + cmpval.toString() + ":" +
          tval.toString() + "*",
          cmpval.toString().equals(tval.toString()));
        ++count;
      }
    } finally {
      reader.close();
    }
  }
  assertEquals("Some records not found", RECORDS, count);
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test(timeout=10000)
public void testFormat() throws IOException, InterruptedException {
  Job job = Job.getInstance(conf);

  Random random = new Random();
  long seed = random.nextLong();
  random.setSeed(seed);

  localFs.delete(workDir, true);
  FileInputFormat.setInputPaths(job, workDir);

  final int length = 10000;
  final int numFiles = 10;

  // create files with a variety of lengths
  createFiles(length, numFiles, random, job);

  TaskAttemptContext context = MapReduceTestUtil.
    createDummyMapTaskAttemptContext(job.getConfiguration());
  // create a combine split for the files
  InputFormat<IntWritable,BytesWritable> format =
    new CombineSequenceFileInputFormat<IntWritable,BytesWritable>();
  for (int i = 0; i < 3; i++) {
    int numSplits =
      random.nextInt(length/(SequenceFile.SYNC_INTERVAL/20)) + 1;
    LOG.info("splitting: requesting = " + numSplits);
    List<InputSplit> splits = format.getSplits(job);
    LOG.info("splitting: got =        " + splits.size());

    // we should have a single split as the length is comfortably smaller than
    // the block size
    assertEquals("We got more than one splits!", 1, splits.size());
    InputSplit split = splits.get(0);
    assertEquals("It should be CombineFileSplit",
      CombineFileSplit.class, split.getClass());

    // check the split
    BitSet bits = new BitSet(length);
    RecordReader<IntWritable,BytesWritable> reader =
      format.createRecordReader(split, context);
    MapContext<IntWritable,BytesWritable,IntWritable,BytesWritable> mcontext =
      new MapContextImpl<IntWritable,BytesWritable,IntWritable,BytesWritable>(job.getConfiguration(),
      context.getTaskAttemptID(), reader, null, null,
      MapReduceTestUtil.createDummyReporter(), split);
    reader.initialize(split, mcontext);
    assertEquals("reader class is CombineFileRecordReader.",
      CombineFileRecordReader.class, reader.getClass());

    try {
      while (reader.nextKeyValue()) {
        IntWritable key = reader.getCurrentKey();
        BytesWritable value = reader.getCurrentValue();
        assertNotNull("Value should not be null.", value);
        final int k = key.get();
        LOG.debug("read " + k);
        assertFalse("Key in multiple partitions.", bits.get(k));
        bits.set(k);
      }
    } finally {
      reader.close();
    }
    assertEquals("Some keys in no partition.", length, bits.cardinality());
  }
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
private static void validateFileSplits(final List<FileSplit> fileSplits, final Configuration configuration,
                                       final Class<? extends InputFormat<NullWritable, VertexWritable>> inputFormatClass,
                                       final Optional<Class<? extends OutputFormat<NullWritable, VertexWritable>>> outFormatClass) throws Exception {

    final InputFormat inputFormat = ReflectionUtils.newInstance(inputFormatClass, configuration);
    final TaskAttemptContext job = new TaskAttemptContextImpl(configuration, new TaskAttemptID(UUID.randomUUID().toString(), 0, TaskType.MAP, 0, 0));

    int vertexCount = 0;
    int outEdgeCount = 0;
    int inEdgeCount = 0;

    final OutputFormat<NullWritable, VertexWritable> outputFormat = outFormatClass.isPresent() ? ReflectionUtils.newInstance(outFormatClass.get(), configuration) : null;
    final RecordWriter<NullWritable, VertexWritable> writer = null == outputFormat ? null : outputFormat.getRecordWriter(job);

    boolean foundKeyValue = false;
    for (final FileSplit split : fileSplits) {
        logger.info("\treading file split {}", split.getPath().getName() + " ({}", split.getStart() + "..." + (split.getStart() + split.getLength()), "{} {} bytes)");
        final RecordReader reader = inputFormat.createRecordReader(split, job);

        float lastProgress = -1f;
        while (reader.nextKeyValue()) {
            //System.out.println("" + reader.getProgress() + "> " + reader.getCurrentKey() + ": " + reader.getCurrentValue());
            final float progress = reader.getProgress();
            assertTrue(progress >= lastProgress);
            assertEquals(NullWritable.class, reader.getCurrentKey().getClass());
            final VertexWritable vertexWritable = (VertexWritable) reader.getCurrentValue();
            if (null != writer) writer.write(NullWritable.get(), vertexWritable);
            vertexCount++;
            outEdgeCount = outEdgeCount + (int) IteratorUtils.count(vertexWritable.get().edges(Direction.OUT));
            inEdgeCount = inEdgeCount + (int) IteratorUtils.count(vertexWritable.get().edges(Direction.IN));
            //
            final Vertex vertex = vertexWritable.get();
            assertEquals(Integer.class, vertex.id().getClass());
            if (vertex.value("name").equals("SUGAR MAGNOLIA")) {
                foundKeyValue = true;
                assertEquals(92, IteratorUtils.count(vertex.edges(Direction.OUT)));
                assertEquals(77, IteratorUtils.count(vertex.edges(Direction.IN)));
            }
            lastProgress = progress;
        }
    }

    assertEquals(8049, outEdgeCount);
    assertEquals(8049, inEdgeCount);
    assertEquals(outEdgeCount, inEdgeCount);
    assertEquals(808, vertexCount);
    assertTrue(foundKeyValue);

    if (null != writer) {
        writer.close(new TaskAttemptContextImpl(configuration, job.getTaskAttemptID()));
        for (int i = 1; i < 10; i++) {
            final File outputDirectory = new File(new URL(configuration.get("mapreduce.output.fileoutputformat.outputdir")).toURI());
            final List<FileSplit> splits = generateFileSplits(new File(outputDirectory.getAbsoluteFile() + "/_temporary/0/_temporary/" + job.getTaskAttemptID().getTaskID().toString().replace("task", "attempt") + "_0" + "/part-m-00000"), i);
            validateFileSplits(splits, configuration, inputFormatClass, Optional.empty());
        }
    }
}
 

import org.apache.hadoop.mapreduce.InputFormat; //导入方法依赖的package包/类
@Test(enabled = true, dependsOnMethods = { "testWriteChunkData" })
public void testReadChunkData() throws Exception {
  List<String> partfns = new ArrayList<String>();
  long reccnt = 0L;
  long tsize = 0L;
  Checksum cs = new CRC32();
  cs.reset();
  File folder = new File(m_workdir.toString());
  File[] listfiles = folder.listFiles();
  for (int idx = 0; idx < listfiles.length; ++idx) {
    if (listfiles[idx].isFile()
        && listfiles[idx].getName().startsWith(MneConfigHelper.getBaseOutputName(m_conf, null))
        && listfiles[idx].getName().endsWith(MneConfigHelper.DEFAULT_FILE_EXTENSION)) {
      partfns.add(listfiles[idx].getName());
    }
  }
  Collections.sort(partfns); // keep the order for checksum
  for (int idx = 0; idx < partfns.size(); ++idx) {
    System.out.println(String.format("Verifying : %s", partfns.get(idx)));
    FileSplit split = new FileSplit(
        new Path(m_workdir, partfns.get(idx)), 0, 0L, new String[0]);
    InputFormat<NullWritable, MneDurableInputValue<DurableChunk<?>>> inputFormat =
        new MneInputFormat<MneDurableInputValue<DurableChunk<?>>, DurableChunk<?>>();
    RecordReader<NullWritable, MneDurableInputValue<DurableChunk<?>>> reader =
        inputFormat.createRecordReader(split, m_tacontext);
    MneDurableInputValue<DurableChunk<?>> dchkval = null;
    while (reader.nextKeyValue()) {
      dchkval = reader.getCurrentValue();
      byte b;
      for (int j = 0; j < dchkval.getValue().getSize(); ++j) {
        b = unsafe.getByte(dchkval.getValue().get() + j);
        cs.update(b);
      }
      tsize += dchkval.getValue().getSize();
      ++reccnt;
    }
    reader.close();
  }
  AssertJUnit.assertEquals(m_reccnt, reccnt);
  AssertJUnit.assertEquals(m_totalsize, tsize);
  AssertJUnit.assertEquals(m_checksum, cs.getValue());
  System.out.println(String.format("The checksum of chunk is %d", m_checksum));
}