Examples with TaskAttemptContextImpl org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl used on opensource projects

Example 1 with TaskAttemptContextImpl

use of org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in project hadoop by apache.

the class TestRecovery method writeOutput.

private void writeOutput(TaskAttempt attempt, Configuration conf) throws Exception {
    TaskAttemptContext tContext = new TaskAttemptContextImpl(conf, TypeConverter.fromYarn(attempt.getID()));
    TextOutputFormat<?, ?> theOutputFormat = new TextOutputFormat();
    RecordWriter theRecordWriter = theOutputFormat.getRecordWriter(tContext);
    NullWritable nullWritable = NullWritable.get();
    try {
        theRecordWriter.write(key1, val1);
        theRecordWriter.write(null, nullWritable);
        theRecordWriter.write(null, val1);
        theRecordWriter.write(nullWritable, val2);
        theRecordWriter.write(key2, nullWritable);
        theRecordWriter.write(key1, null);
        theRecordWriter.write(null, null);
        theRecordWriter.write(key2, val2);
    } finally {
        theRecordWriter.close(tContext);
    }
    OutputFormat outputFormat = ReflectionUtils.newInstance(tContext.getOutputFormatClass(), conf);
    OutputCommitter committer = outputFormat.getOutputCommitter(tContext);
    committer.commitTask(tContext);
}

Example 2 with TaskAttemptContextImpl

use of org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in project hadoop by apache.

the class TeraInputFormat method writePartitionFile.

/**
   * Use the input splits to take samples of the input and generate sample
   * keys. By default reads 100,000 keys from 10 locations in the input, sorts
   * them and picks N-1 keys to generate N equally sized partitions.
   * @param job the job to sample
   * @param partFile where to write the output file to
   * @throws Throwable if something goes wrong
   */
public static void writePartitionFile(final JobContext job, Path partFile) throws Throwable {
    long t1 = System.currentTimeMillis();
    Configuration conf = job.getConfiguration();
    final TeraInputFormat inFormat = new TeraInputFormat();
    final TextSampler sampler = new TextSampler();
    int partitions = job.getNumReduceTasks();
    long sampleSize = conf.getLong(TeraSortConfigKeys.SAMPLE_SIZE.key(), TeraSortConfigKeys.DEFAULT_SAMPLE_SIZE);
    final List<InputSplit> splits = inFormat.getSplits(job);
    long t2 = System.currentTimeMillis();
    System.out.println("Computing input splits took " + (t2 - t1) + "ms");
    int samples = Math.min(conf.getInt(TeraSortConfigKeys.NUM_PARTITIONS.key(), TeraSortConfigKeys.DEFAULT_NUM_PARTITIONS), splits.size());
    System.out.println("Sampling " + samples + " splits of " + splits.size());
    final long recordsPerSample = sampleSize / samples;
    final int sampleStep = splits.size() / samples;
    Thread[] samplerReader = new Thread[samples];
    SamplerThreadGroup threadGroup = new SamplerThreadGroup("Sampler Reader Thread Group");
    // take N samples from different parts of the input
    for (int i = 0; i < samples; ++i) {
        final int idx = i;
        samplerReader[i] = new Thread(threadGroup, "Sampler Reader " + idx) {

            {
                setDaemon(true);
            }

            public void run() {
                long records = 0;
                try {
                    TaskAttemptContext context = new TaskAttemptContextImpl(job.getConfiguration(), new TaskAttemptID());
                    RecordReader<Text, Text> reader = inFormat.createRecordReader(splits.get(sampleStep * idx), context);
                    reader.initialize(splits.get(sampleStep * idx), context);
                    while (reader.nextKeyValue()) {
                        sampler.addKey(new Text(reader.getCurrentKey()));
                        records += 1;
                        if (recordsPerSample <= records) {
                            break;
                        }
                    }
                } catch (IOException ie) {
                    System.err.println("Got an exception while reading splits " + StringUtils.stringifyException(ie));
                    throw new RuntimeException(ie);
                } catch (InterruptedException e) {
                }
            }
        };
        samplerReader[i].start();
    }
    FileSystem outFs = partFile.getFileSystem(conf);
    DataOutputStream writer = outFs.create(partFile, true, 64 * 1024, (short) 10, outFs.getDefaultBlockSize(partFile));
    for (int i = 0; i < samples; i++) {
        try {
            samplerReader[i].join();
            if (threadGroup.getThrowable() != null) {
                throw threadGroup.getThrowable();
            }
        } catch (InterruptedException e) {
        }
    }
    for (Text split : sampler.createPartitions(partitions)) {
        split.write(writer);
    }
    writer.close();
    long t3 = System.currentTimeMillis();
    System.out.println("Computing parititions took " + (t3 - t2) + "ms");
}

Example 3 with TaskAttemptContextImpl

use of org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in project hadoop by apache.

the class TestLineRecordReader method testMultipleClose.

@Test
public void testMultipleClose() throws IOException {
    URL testFileUrl = getClass().getClassLoader().getResource("recordSpanningMultipleSplits.txt.bz2");
    assertNotNull("Cannot find recordSpanningMultipleSplits.txt.bz2", testFileUrl);
    File testFile = new File(testFileUrl.getFile());
    Path testFilePath = new Path(testFile.getAbsolutePath());
    long testFileSize = testFile.length();
    Configuration conf = new Configuration();
    conf.setInt(org.apache.hadoop.mapreduce.lib.input.LineRecordReader.MAX_LINE_LENGTH, Integer.MAX_VALUE);
    TaskAttemptContext context = new TaskAttemptContextImpl(conf, new TaskAttemptID());
    // read the data and check whether BOM is skipped
    FileSplit split = new FileSplit(testFilePath, 0, testFileSize, null);
    LineRecordReader reader = new LineRecordReader();
    reader.initialize(split, context);
    //noinspection StatementWithEmptyBody
    while (reader.nextKeyValue()) ;
    reader.close();
    reader.close();
    BZip2Codec codec = new BZip2Codec();
    codec.setConf(conf);
    Set<Decompressor> decompressors = new HashSet<Decompressor>();
    for (int i = 0; i < 10; ++i) {
        decompressors.add(CodecPool.getDecompressor(codec));
    }
    assertEquals(10, decompressors.size());
}

Example 4 with TaskAttemptContextImpl

use of org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in project hadoop by apache.

the class TestLineRecordReader method testUncompressedInputDefaultDelimiterPosValue.

@Test
public void testUncompressedInputDefaultDelimiterPosValue() throws Exception {
    Configuration conf = new Configuration();
    String inputData = "1234567890\r\n12\r\n345";
    Path inputFile = createInputFile(conf, inputData);
    conf.setInt("io.file.buffer.size", 10);
    conf.setInt(org.apache.hadoop.mapreduce.lib.input.LineRecordReader.MAX_LINE_LENGTH, Integer.MAX_VALUE);
    FileSplit split = new FileSplit(inputFile, 0, 15, (String[]) null);
    TaskAttemptContext context = new TaskAttemptContextImpl(conf, new TaskAttemptID());
    LineRecordReader reader = new LineRecordReader(null);
    reader.initialize(split, context);
    LongWritable key;
    Text value;
    reader.nextKeyValue();
    key = reader.getCurrentKey();
    value = reader.getCurrentValue();
    // Get first record:"1234567890"
    assertEquals(10, value.getLength());
    assertEquals(0, key.get());
    reader.nextKeyValue();
    // Get second record:"12"
    assertEquals(2, value.getLength());
    // Key should be 12 right after "1234567890\r\n"
    assertEquals(12, key.get());
    assertFalse(reader.nextKeyValue());
    // Key should be 16 right after "1234567890\r\n12\r\n"
    assertEquals(16, key.get());
    split = new FileSplit(inputFile, 15, 4, (String[]) null);
    reader = new LineRecordReader(null);
    reader.initialize(split, context);
    // The second split dropped the first record "\n"
    reader.nextKeyValue();
    key = reader.getCurrentKey();
    value = reader.getCurrentValue();
    // Get third record:"345"
    assertEquals(3, value.getLength());
    // Key should be 16 right after "1234567890\r\n12\r\n"
    assertEquals(16, key.get());
    assertFalse(reader.nextKeyValue());
    // Key should be 19 right after "1234567890\r\n12\r\n345"
    assertEquals(19, key.get());
    inputData = "123456789\r\r\n";
    inputFile = createInputFile(conf, inputData);
    split = new FileSplit(inputFile, 0, 12, (String[]) null);
    reader = new LineRecordReader(null);
    reader.initialize(split, context);
    reader.nextKeyValue();
    key = reader.getCurrentKey();
    value = reader.getCurrentValue();
    // Get first record:"123456789"
    assertEquals(9, value.getLength());
    assertEquals(0, key.get());
    reader.nextKeyValue();
    // Get second record:""
    assertEquals(0, value.getLength());
    // Key should be 10 right after "123456789\r"
    assertEquals(10, key.get());
    assertFalse(reader.nextKeyValue());
    // Key should be 12 right after "123456789\r\r\n"
    assertEquals(12, key.get());
}

Example 5 with TaskAttemptContextImpl

use of org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in project hadoop by apache.

the class TestLineRecordReader method readRecords.

// Use the LineRecordReader to read records from the file
public ArrayList<String> readRecords(URL testFileUrl, int splitSize) throws IOException {
    // Set up context
    File testFile = new File(testFileUrl.getFile());
    long testFileSize = testFile.length();
    Path testFilePath = new Path(testFile.getAbsolutePath());
    Configuration conf = new Configuration();
    conf.setInt("io.file.buffer.size", 1);
    TaskAttemptContext context = new TaskAttemptContextImpl(conf, new TaskAttemptID());
    // Gather the records returned by the record reader
    ArrayList<String> records = new ArrayList<String>();
    long offset = 0;
    while (offset < testFileSize) {
        FileSplit split = new FileSplit(testFilePath, offset, splitSize, null);
        LineRecordReader reader = new LineRecordReader();
        reader.initialize(split, context);
        while (reader.nextKeyValue()) {
            records.add(reader.getCurrentValue().toString());
        }
        offset += splitSize;
    }
    return records;
}