Search in sources :

Example 46 with TaskAttemptContext

use of org.apache.hadoop.mapreduce.TaskAttemptContext 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);
    }
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) Node(org.apache.jena.graph.Node) ArrayList(java.util.ArrayList) TaskAttemptContextImpl(org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl) LongWritable(org.apache.hadoop.io.LongWritable) JobContext(org.apache.hadoop.mapreduce.JobContext) Job(org.apache.hadoop.mapreduce.Job) InputSplit(org.apache.hadoop.mapreduce.InputSplit) HashSet(java.util.HashSet) Path(org.apache.hadoop.fs.Path) JobContextImpl(org.apache.hadoop.mapreduce.task.JobContextImpl) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) File(java.io.File) FileAttribute(java.nio.file.attribute.FileAttribute) Test(org.junit.Test)

Example 47 with TaskAttemptContext

use of org.apache.hadoop.mapreduce.TaskAttemptContext 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);
    }
}
Also used : Path(org.apache.hadoop.fs.Path) JobContextImpl(org.apache.hadoop.mapreduce.task.JobContextImpl) Configuration(org.apache.hadoop.conf.Configuration) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) Node(org.apache.jena.graph.Node) ArrayList(java.util.ArrayList) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) TaskAttemptContextImpl(org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl) LongWritable(org.apache.hadoop.io.LongWritable) JobContext(org.apache.hadoop.mapreduce.JobContext) Job(org.apache.hadoop.mapreduce.Job) File(java.io.File) InputSplit(org.apache.hadoop.mapreduce.InputSplit) FileAttribute(java.nio.file.attribute.FileAttribute) HashSet(java.util.HashSet) Test(org.junit.Test)

Example 48 with TaskAttemptContext

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

the class AbstractNodeTupleOutputFormatTests method testOutput.

/**
     * Tests output
     * 
     * @param f
     *            File to output to
     * @param num
     *            Number of tuples to output
     * @throws IOException
     * @throws InterruptedException
     */
protected final void testOutput(File f, int num) throws IOException, InterruptedException {
    // Prepare configuration
    Configuration config = this.prepareConfiguration();
    // Set up fake job
    OutputFormat<NullWritable, T> outputFormat = this.getOutputFormat();
    Job job = Job.getInstance(config);
    job.setOutputFormatClass(outputFormat.getClass());
    this.addOutputPath(f, job.getConfiguration(), job);
    JobContext context = new JobContextImpl(job.getConfiguration(), job.getJobID());
    Assert.assertNotNull(FileOutputFormat.getOutputPath(context));
    // Output the data
    TaskAttemptID id = new TaskAttemptID("outputTest", 1, TaskType.MAP, 1, 1);
    TaskAttemptContext taskContext = new TaskAttemptContextImpl(job.getConfiguration(), id);
    RecordWriter<NullWritable, T> writer = outputFormat.getRecordWriter(taskContext);
    Iterator<T> tuples = this.generateTuples(num);
    while (tuples.hasNext()) {
        writer.write(NullWritable.get(), tuples.next());
    }
    writer.close(taskContext);
    // Check output
    File outputFile = this.findOutputFile(this.folder.getRoot(), context);
    Assert.assertNotNull(outputFile);
    this.checkTuples(outputFile, num);
}
Also used : JobContextImpl(org.apache.hadoop.mapreduce.task.JobContextImpl) Configuration(org.apache.hadoop.conf.Configuration) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) NullWritable(org.apache.hadoop.io.NullWritable) TaskAttemptContextImpl(org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl) JobContext(org.apache.hadoop.mapreduce.JobContext) Job(org.apache.hadoop.mapreduce.Job) File(java.io.File)

Example 49 with TaskAttemptContext

use of org.apache.hadoop.mapreduce.TaskAttemptContext 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);
    }
}
Also used : JobContextImpl(org.apache.hadoop.mapreduce.task.JobContextImpl) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) TaskAttemptContextImpl(org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) LongWritable(org.apache.hadoop.io.LongWritable) JobContext(org.apache.hadoop.mapreduce.JobContext) Job(org.apache.hadoop.mapreduce.Job) InputSplit(org.apache.hadoop.mapreduce.InputSplit)

Example 50 with TaskAttemptContext

use of org.apache.hadoop.mapreduce.TaskAttemptContext in project eiger by wlloyd.

the class ColumnFamilyInputFormat method getSplits.

//
// Old Hadoop API
//
public org.apache.hadoop.mapred.InputSplit[] getSplits(JobConf jobConf, int numSplits) throws IOException {
    TaskAttemptContext tac = new TaskAttemptContext(jobConf, new TaskAttemptID());
    List<org.apache.hadoop.mapreduce.InputSplit> newInputSplits = this.getSplits(tac);
    org.apache.hadoop.mapred.InputSplit[] oldInputSplits = new org.apache.hadoop.mapred.InputSplit[newInputSplits.size()];
    for (int i = 0; i < newInputSplits.size(); i++) oldInputSplits[i] = (ColumnFamilySplit) newInputSplits.get(i);
    return oldInputSplits;
}
Also used : TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) InputSplit(org.apache.hadoop.mapreduce.InputSplit)

Aggregations

TaskAttemptContext (org.apache.hadoop.mapreduce.TaskAttemptContext)110 Configuration (org.apache.hadoop.conf.Configuration)58 Job (org.apache.hadoop.mapreduce.Job)44 Path (org.apache.hadoop.fs.Path)39 TaskAttemptContextImpl (org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl)38 InputSplit (org.apache.hadoop.mapreduce.InputSplit)36 Test (org.junit.Test)35 TaskAttemptID (org.apache.hadoop.mapreduce.TaskAttemptID)33 JobContext (org.apache.hadoop.mapreduce.JobContext)28 IOException (java.io.IOException)27 File (java.io.File)22 LongWritable (org.apache.hadoop.io.LongWritable)22 JobContextImpl (org.apache.hadoop.mapreduce.task.JobContextImpl)21 RecordWriter (org.apache.hadoop.mapreduce.RecordWriter)19 MapContextImpl (org.apache.hadoop.mapreduce.task.MapContextImpl)17 FileSystem (org.apache.hadoop.fs.FileSystem)16 OutputCommitter (org.apache.hadoop.mapreduce.OutputCommitter)12 ArrayList (java.util.ArrayList)11 BytesWritable (org.apache.hadoop.io.BytesWritable)10 MapFile (org.apache.hadoop.io.MapFile)10