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

Example 11 with JobContextImpl

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

the class TestCopyCommitter method testDeleteMissingFlatInterleavedFiles.

@Test
public void testDeleteMissingFlatInterleavedFiles() {
    TaskAttemptContext taskAttemptContext = getTaskAttemptContext(config);
    JobContext jobContext = new JobContextImpl(taskAttemptContext.getConfiguration(), taskAttemptContext.getTaskAttemptID().getJobID());
    Configuration conf = jobContext.getConfiguration();
    String sourceBase;
    String targetBase;
    FileSystem fs = null;
    try {
        OutputCommitter committer = new CopyCommitter(null, taskAttemptContext);
        fs = FileSystem.get(conf);
        sourceBase = "/tmp1/" + String.valueOf(rand.nextLong());
        targetBase = "/tmp1/" + String.valueOf(rand.nextLong());
        TestDistCpUtils.createFile(fs, sourceBase + "/1");
        TestDistCpUtils.createFile(fs, sourceBase + "/3");
        TestDistCpUtils.createFile(fs, sourceBase + "/4");
        TestDistCpUtils.createFile(fs, sourceBase + "/5");
        TestDistCpUtils.createFile(fs, sourceBase + "/7");
        TestDistCpUtils.createFile(fs, sourceBase + "/8");
        TestDistCpUtils.createFile(fs, sourceBase + "/9");
        TestDistCpUtils.createFile(fs, targetBase + "/2");
        TestDistCpUtils.createFile(fs, targetBase + "/4");
        TestDistCpUtils.createFile(fs, targetBase + "/5");
        TestDistCpUtils.createFile(fs, targetBase + "/7");
        TestDistCpUtils.createFile(fs, targetBase + "/9");
        TestDistCpUtils.createFile(fs, targetBase + "/A");
        DistCpOptions options = new DistCpOptions(Arrays.asList(new Path(sourceBase)), new Path("/out"));
        options.setSyncFolder(true);
        options.setDeleteMissing(true);
        options.appendToConf(conf);
        CopyListing listing = new GlobbedCopyListing(conf, CREDENTIALS);
        Path listingFile = new Path("/tmp1/" + String.valueOf(rand.nextLong()));
        listing.buildListing(listingFile, options);
        conf.set(DistCpConstants.CONF_LABEL_TARGET_WORK_PATH, targetBase);
        conf.set(DistCpConstants.CONF_LABEL_TARGET_FINAL_PATH, targetBase);
        committer.commitJob(jobContext);
        if (!TestDistCpUtils.checkIfFoldersAreInSync(fs, targetBase, sourceBase)) {
            Assert.fail("Source and target folders are not in sync");
        }
        Assert.assertEquals(fs.listStatus(new Path(targetBase)).length, 4);
        //Test for idempotent commit
        committer.commitJob(jobContext);
        if (!TestDistCpUtils.checkIfFoldersAreInSync(fs, targetBase, sourceBase)) {
            Assert.fail("Source and target folders are not in sync");
        }
        Assert.assertEquals(fs.listStatus(new Path(targetBase)).length, 4);
    } catch (IOException e) {
        LOG.error("Exception encountered while testing for delete missing", e);
        Assert.fail("Delete missing failure");
    } finally {
        TestDistCpUtils.delete(fs, "/tmp1");
        conf.set(DistCpConstants.CONF_LABEL_DELETE_MISSING, "false");
    }
}

Example 12 with JobContextImpl

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

the class TestCopyCommitter method testAtomicCommitMissingFinal.

@Test
public void testAtomicCommitMissingFinal() {
    TaskAttemptContext taskAttemptContext = getTaskAttemptContext(config);
    JobContext jobContext = new JobContextImpl(taskAttemptContext.getConfiguration(), taskAttemptContext.getTaskAttemptID().getJobID());
    Configuration conf = jobContext.getConfiguration();
    String workPath = "/tmp1/" + String.valueOf(rand.nextLong());
    String finalPath = "/tmp1/" + String.valueOf(rand.nextLong());
    FileSystem fs = null;
    try {
        OutputCommitter committer = new CopyCommitter(null, taskAttemptContext);
        fs = FileSystem.get(conf);
        fs.mkdirs(new Path(workPath));
        conf.set(DistCpConstants.CONF_LABEL_TARGET_WORK_PATH, workPath);
        conf.set(DistCpConstants.CONF_LABEL_TARGET_FINAL_PATH, finalPath);
        conf.setBoolean(DistCpConstants.CONF_LABEL_ATOMIC_COPY, true);
        Assert.assertTrue(fs.exists(new Path(workPath)));
        Assert.assertFalse(fs.exists(new Path(finalPath)));
        committer.commitJob(jobContext);
        Assert.assertFalse(fs.exists(new Path(workPath)));
        Assert.assertTrue(fs.exists(new Path(finalPath)));
        //Test for idempotent commit
        committer.commitJob(jobContext);
        Assert.assertFalse(fs.exists(new Path(workPath)));
        Assert.assertTrue(fs.exists(new Path(finalPath)));
    } catch (IOException e) {
        LOG.error("Exception encountered while testing for preserve status", e);
        Assert.fail("Atomic commit failure");
    } finally {
        TestDistCpUtils.delete(fs, workPath);
        TestDistCpUtils.delete(fs, finalPath);
        conf.setBoolean(DistCpConstants.CONF_LABEL_ATOMIC_COPY, false);
    }
}

Example 13 with JobContextImpl

use of org.apache.hadoop.mapreduce.task.JobContextImpl in project gora by apache.

the class GoraMapReduceUtils method createJobContext.

public static JobContext createJobContext(Configuration conf, String inputPath) throws IOException {
    if (inputPath != null) {
        Job job = Job.getInstance(conf);
        FileInputFormat.addInputPath(job, new Path(inputPath));
        return new JobContextImpl(job.getConfiguration(), null);
    }
    return new JobContextImpl(conf, null);
}

Example 14 with JobContextImpl

use of org.apache.hadoop.mapreduce.task.JobContextImpl 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 15 with JobContextImpl

use of org.apache.hadoop.mapreduce.task.JobContextImpl 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);
    }
}