Examples with InputSplit org.apache.hadoop.mapreduce.InputSplit used on opensource projects

Example 71 with InputSplit

use of org.apache.hadoop.mapreduce.InputSplit in project ignite by apache.

the class HadoopV2Splitter method splitJob.

/**
     * @param ctx Job context.
     * @return Collection of mapped splits.
     * @throws IgniteCheckedException If mapping failed.
     */
public static Collection<HadoopInputSplit> splitJob(JobContext ctx) throws IgniteCheckedException {
    try {
        InputFormat<?, ?> format = ReflectionUtils.newInstance(ctx.getInputFormatClass(), ctx.getConfiguration());
        assert format != null;
        List<InputSplit> splits = format.getSplits(ctx);
        Collection<HadoopInputSplit> res = new ArrayList<>(splits.size());
        int id = 0;
        for (InputSplit nativeSplit : splits) {
            if (nativeSplit instanceof FileSplit) {
                FileSplit s = (FileSplit) nativeSplit;
                res.add(new HadoopFileBlock(s.getLocations(), s.getPath().toUri(), s.getStart(), s.getLength()));
            } else
                res.add(HadoopUtils.wrapSplit(id, nativeSplit, nativeSplit.getLocations()));
            id++;
        }
        return res;
    } catch (IOException | ClassNotFoundException e) {
        throw new IgniteCheckedException(e);
    } catch (InterruptedException e) {
        Thread.currentThread().interrupt();
        throw new IgniteInterruptedCheckedException(e);
    }
}

Example 72 with InputSplit

use of org.apache.hadoop.mapreduce.InputSplit in project jena by apache.

the class AbstractBlankNodeTests method blank_node_divergence_02.

/**
     * Test that starts with two blank nodes with the same identity in a single
     * file, splits them over two files and shows that they diverge in the
     * subsequent job when the JENA-820 workaround is not enabled
     * 
     * @throws IOException
     * @throws InterruptedException
     */
@Test
public void blank_node_divergence_02() throws IOException, InterruptedException {
    Assume.assumeTrue("Requires ParserProfile be respected", this.respectsParserProfile());
    Assume.assumeFalse("Requires that Blank Node identity not be preserved", this.preservesBlankNodeIdentity());
    // Temporary files
    File a = File.createTempFile("bnode_divergence", getInitialInputExtension());
    File intermediateOutputDir = Files.createTempDirectory("bnode_divergence", new FileAttribute[0]).toFile();
    try {
        // Prepare the input data
        // Two mentions of the same blank node in the same file
        List<T> tuples = new ArrayList<>();
        Node bnode = NodeFactory.createBlankNode();
        Node pred = NodeFactory.createURI("http://example.org/predicate");
        tuples.add(createTuple(bnode, pred, NodeFactory.createLiteral("first")));
        tuples.add(createTuple(bnode, pred, NodeFactory.createLiteral("second")));
        writeTuples(a, tuples);
        // Set up fake job which will process the file as a single split
        Configuration config = new Configuration(true);
        InputFormat<LongWritable, TValue> inputFormat = createInitialInputFormat();
        Job job = Job.getInstance(config);
        job.setInputFormatClass(inputFormat.getClass());
        NLineInputFormat.setNumLinesPerSplit(job, 100);
        FileInputFormat.setInputPaths(job, new Path(a.getAbsolutePath()));
        FileOutputFormat.setOutputPath(job, new Path(intermediateOutputDir.getAbsolutePath()));
        JobContext context = new JobContextImpl(job.getConfiguration(), job.getJobID());
        // Get the splits
        List<InputSplit> splits = inputFormat.getSplits(context);
        Assert.assertEquals(1, splits.size());
        for (InputSplit split : splits) {
            // Initialize the input reading
            TaskAttemptContext inputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), createAttemptID(1, 1, 1));
            RecordReader<LongWritable, TValue> reader = inputFormat.createRecordReader(split, inputTaskContext);
            reader.initialize(split, inputTaskContext);
            // Copy the input to the output - each triple goes to a separate
            // output file
            // This is how we force multiple files to be produced
            int taskID = 1;
            while (reader.nextKeyValue()) {
                // Prepare the output writing
                OutputFormat<LongWritable, TValue> outputFormat = createIntermediateOutputFormat();
                TaskAttemptContext outputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), createAttemptID(1, ++taskID, 1));
                RecordWriter<LongWritable, TValue> writer = outputFormat.getRecordWriter(outputTaskContext);
                writer.write(reader.getCurrentKey(), reader.getCurrentValue());
                writer.close(outputTaskContext);
            }
        }
        // Promote outputs from temporary status
        promoteInputs(intermediateOutputDir);
        // Now we need to create a subsequent job that reads the
        // intermediate outputs
        // As described in JENA-820 at this point the blank nodes are
        // consistent, however when we read them from different files they
        // by default get treated as different nodes and so the blank nodes
        // diverge which is incorrect and undesirable behaviour in
        // multi-stage pipelines. However it is the default behaviour
        // because when we start from external inputs we want them to be
        // file scoped.
        LOGGER.debug("Intermediate output directory is {}", intermediateOutputDir.getAbsolutePath());
        job = Job.getInstance(config);
        inputFormat = createIntermediateInputFormat();
        job.setInputFormatClass(inputFormat.getClass());
        FileInputFormat.setInputPaths(job, new Path(intermediateOutputDir.getAbsolutePath()));
        // Make sure JENA-820 flag is disabled
        job.getConfiguration().setBoolean(RdfIOConstants.GLOBAL_BNODE_IDENTITY, false);
        context = new JobContextImpl(job.getConfiguration(), job.getJobID());
        // Get the splits
        splits = inputFormat.getSplits(context);
        Assert.assertEquals(2, splits.size());
        // Expect to end up with a single blank node
        Set<Node> nodes = new HashSet<Node>();
        for (InputSplit split : splits) {
            TaskAttemptContext inputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), new TaskAttemptID());
            RecordReader<LongWritable, TValue> reader = inputFormat.createRecordReader(split, inputTaskContext);
            reader.initialize(split, inputTaskContext);
            while (reader.nextKeyValue()) {
                nodes.add(getSubject(reader.getCurrentValue().get()));
            }
        }
        // Nodes should have diverged
        Assert.assertEquals(2, nodes.size());
    } finally {
        a.delete();
        deleteDirectory(intermediateOutputDir);
    }
}

Example 73 with InputSplit

use of org.apache.hadoop.mapreduce.InputSplit in project jena by apache.

the class AbstractBlankNodeTests method blank_node_identity_01.

/**
     * Test that starts with two blank nodes in two different files and checks
     * that writing them to a single file does not conflate them
     * 
     * @throws IOException
     * @throws InterruptedException
     */
@Test
public void blank_node_identity_01() throws IOException, InterruptedException {
    Assume.assumeTrue("Requires ParserProfile be respected", this.respectsParserProfile());
    Assume.assumeFalse("Requires that Blank Node identity not be preserved", this.preservesBlankNodeIdentity());
    // Temporary files
    File a = File.createTempFile("bnode_identity", getInitialInputExtension());
    File b = File.createTempFile("bnode_identity", getInitialInputExtension());
    File intermediateOutputDir = Files.createTempDirectory("bnode_identity", new FileAttribute[0]).toFile();
    try {
        // Prepare the input data
        // Different blank nodes in different files
        List<T> tuples = new ArrayList<>();
        Node bnode1 = NodeFactory.createBlankNode();
        Node bnode2 = NodeFactory.createBlankNode();
        Node pred = NodeFactory.createURI("http://example.org/predicate");
        tuples.add(createTuple(bnode1, pred, NodeFactory.createLiteral("first")));
        writeTuples(a, tuples);
        tuples.clear();
        tuples.add(createTuple(bnode2, pred, NodeFactory.createLiteral("second")));
        writeTuples(b, tuples);
        // Set up fake job which will process the two files
        Configuration config = new Configuration(true);
        InputFormat<LongWritable, TValue> inputFormat = createInitialInputFormat();
        Job job = Job.getInstance(config);
        job.setInputFormatClass(inputFormat.getClass());
        NLineInputFormat.setNumLinesPerSplit(job, 100);
        FileInputFormat.setInputPaths(job, new Path(a.getAbsolutePath()), new Path(b.getAbsolutePath()));
        FileOutputFormat.setOutputPath(job, new Path(intermediateOutputDir.getAbsolutePath()));
        JobContext context = new JobContextImpl(job.getConfiguration(), job.getJobID());
        // Get the splits
        List<InputSplit> splits = inputFormat.getSplits(context);
        Assert.assertEquals(2, splits.size());
        // Prepare the output writing - putting all output to a single file
        OutputFormat<LongWritable, TValue> outputFormat = createIntermediateOutputFormat();
        TaskAttemptContext outputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), createAttemptID(1, 2, 1));
        RecordWriter<LongWritable, TValue> writer = outputFormat.getRecordWriter(outputTaskContext);
        for (InputSplit split : splits) {
            // Initialize the input reading
            TaskAttemptContext inputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), createAttemptID(1, 1, 1));
            RecordReader<LongWritable, TValue> reader = inputFormat.createRecordReader(split, inputTaskContext);
            reader.initialize(split, inputTaskContext);
            // output
            while (reader.nextKeyValue()) {
                writer.write(reader.getCurrentKey(), reader.getCurrentValue());
            }
        }
        writer.close(outputTaskContext);
        // Promote outputs from temporary status
        promoteInputs(intermediateOutputDir);
        // Now we need to create a subsequent job that reads the
        // intermediate outputs
        // The Blank nodes should have been given separate identities so we
        // should not be conflating them, this is the opposite problem to
        // that described in JENA-820
        LOGGER.debug("Intermediate output directory is {}", intermediateOutputDir.getAbsolutePath());
        job = Job.getInstance(config);
        inputFormat = createIntermediateInputFormat();
        job.setInputFormatClass(inputFormat.getClass());
        NLineInputFormat.setNumLinesPerSplit(job, 100);
        FileInputFormat.setInputPaths(job, new Path(intermediateOutputDir.getAbsolutePath()));
        context = new JobContextImpl(job.getConfiguration(), job.getJobID());
        // Get the splits
        splits = inputFormat.getSplits(context);
        Assert.assertEquals(1, splits.size());
        // Expect to end up with a single blank node
        Set<Node> nodes = new HashSet<Node>();
        for (InputSplit split : splits) {
            TaskAttemptContext inputTaskContext = new TaskAttemptContextImpl(job.getConfiguration(), new TaskAttemptID());
            RecordReader<LongWritable, TValue> reader = inputFormat.createRecordReader(split, inputTaskContext);
            reader.initialize(split, inputTaskContext);
            while (reader.nextKeyValue()) {
                nodes.add(getSubject(reader.getCurrentValue().get()));
            }
        }
        // Nodes must not have converged
        Assert.assertEquals(2, nodes.size());
    } finally {
        a.delete();
        b.delete();
        deleteDirectory(intermediateOutputDir);
    }
}

Example 74 with InputSplit

use of org.apache.hadoop.mapreduce.InputSplit in project jena by apache.

the class AbstractNodeTupleInputFormatTests method testSingleInput.

/**
     * Runs a test with a single input
     * 
     * @param config
     *            Configuration
     * @param input
     *            Input
     * @param expectedTuples
     *            Expected tuples
     * @throws IOException
     * @throws InterruptedException
     */
protected final void testSingleInput(Configuration config, File input, int expectedSplits, int expectedTuples) throws IOException, InterruptedException {
    // Set up fake job
    InputFormat<LongWritable, T> inputFormat = this.getInputFormat();
    Job job = Job.getInstance(config);
    job.setInputFormatClass(inputFormat.getClass());
    this.addInputPath(input, job.getConfiguration(), job);
    JobContext context = new JobContextImpl(job.getConfiguration(), job.getJobID());
    Assert.assertEquals(1, FileInputFormat.getInputPaths(context).length);
    NLineInputFormat.setNumLinesPerSplit(job, LARGE_SIZE);
    // Check splits
    List<InputSplit> splits = inputFormat.getSplits(context);
    Assert.assertEquals(expectedSplits, splits.size());
    // Check tuples
    for (InputSplit split : splits) {
        TaskAttemptContext taskContext = new TaskAttemptContextImpl(job.getConfiguration(), new TaskAttemptID());
        RecordReader<LongWritable, T> reader = inputFormat.createRecordReader(split, taskContext);
        reader.initialize(split, taskContext);
        this.checkTuples(reader, expectedTuples);
    }
}

Example 75 with InputSplit

use of org.apache.hadoop.mapreduce.InputSplit in project jena by apache.

the class AbstractNLineFileInputFormat method getSplits.

/**
     * Logically splits the set of input files for the job, splits N lines of
     * the input as one split.
     * 
     * @see FileInputFormat#getSplits(JobContext)
     */
@Override
public final List<InputSplit> getSplits(JobContext job) throws IOException {
    boolean debug = LOGGER.isDebugEnabled();
    if (debug && FileInputFormat.getInputDirRecursive(job)) {
        LOGGER.debug("Recursive searching for input data is enabled");
    }
    List<InputSplit> splits = new ArrayList<InputSplit>();
    int numLinesPerSplit = NLineInputFormat.getNumLinesPerSplit(job);
    for (FileStatus status : listStatus(job)) {
        if (debug) {
            LOGGER.debug("Determining how to split input file/directory {}", status.getPath());
        }
        splits.addAll(NLineInputFormat.getSplitsForFile(status, job.getConfiguration(), numLinesPerSplit));
    }
    return splits;
}