Examples with PartitionOutput co.cask.cdap.api.dataset.lib.PartitionOutput used on opensource projects

Example 6 with PartitionOutput

use of co.cask.cdap.api.dataset.lib.PartitionOutput in project cdap by caskdata.

the class PartitionRollbackTestRun method testPFSRollback.

/*
   * This tests all the following cases:
   *
   *  1. addPartition(location) fails because partition already exists
   *  2. addPartition(location) fails because Hive partition already exists
   *  3. addPartition(location) succeeds but transaction fails
   *  4. partitionOutput.addPartition() fails because partition already exists
   *  5. partitionOutput.addPartition() fails because Hive partition already exists
   *  6. partitionOutput.addPartition() succeeds but transaction fails
   *  7. mapreduce writing partition fails because location already exists
   *  8. mapreduce writing partition fails because partition already exists
   *  9. mapreduce writing partition fails because Hive partition already exists
   *  10. mapreduce writing dynamic partition fails because location already exists
   *  11. mapreduce writing dynamic partition fails because partition already exists
   *  12. mapreduce writing dynamic partition fails because Hive partition already exists
   *  13. multi-output mapreduce writing partition fails because location already exists
   *  13a. first output fails, other output must rollback 0 and 5
   *  13b. second output fails, first output must rollback 0 and 5
   *  14. multi-output mapreduce writing partition fails because partition already exists
   *  14a. first output fails, other output must rollback partition 5
   *  14b. second output fails, first output must rollback partition 5
   *  15. multi-output mapreduce writing partition fails because Hive partition already exists
   *  15a. first output fails, other output must rollback partitions 0 and 5
   *  15b. second output fails, first output must rollback partitions 0 and 5
   *
   * For all these cases, we validate that existing files and partitions are preserved, and newly
   * added files and partitions are rolled back.
   */
@Test
public void testPFSRollback() throws Exception {
    ApplicationManager appManager = deployApplication(AppWritingToPartitioned.class);
    MapReduceManager mrManager = appManager.getMapReduceManager(MAPREDUCE);
    int numRuns = 0;
    Validator pfsValidator = new Validator(PFS);
    Validator otherValidator = new Validator(OTHER);
    final UnitTestManager.UnitTestDatasetManager<PartitionedFileSet> pfsManager = pfsValidator.getPfsManager();
    final PartitionedFileSet pfs = pfsManager.get();
    final PartitionedFileSet other = otherValidator.getPfsManager().get();
    final String path3 = pfsValidator.getRelativePath3();
    // 1. addPartition(location) fails because partition already exists
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                pfs.addPartition(KEY_1, path3);
            }
        });
        Assert.fail("Expected tx to fail because partition for number=1 already exists");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    // 2. addPartition(location) fails because Hive partition already exists
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                pfs.addPartition(KEY_4, path3);
            }
        });
        Assert.fail("Expected tx to fail because hive partition for number=1 already exists");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    // 3. addPartition(location) succeeds but transaction fails
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                pfs.addPartition(KEY_3, path3);
                throw new RuntimeException("fail the tx");
            }
        });
        Assert.fail("Expected tx to fail because it threw a runtime exception");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    // 4. partitionOutput.addPartition() fails because partition already exists
    final PartitionOutput output2x = pfs.getPartitionOutput(KEY_2);
    final Location location2x = output2x.getLocation();
    try (Writer writer = new OutputStreamWriter(location2x.append("file").getOutputStream())) {
        writer.write("2x,2x\n");
    }
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                output2x.addPartition();
            }
        });
        Assert.fail("Expected tx to fail because partition for number=2 already exists");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    Assert.assertFalse(location2x.exists());
    // 5. partitionOutput.addPartition() fails because Hive partition already exists
    final PartitionOutput output4x = pfs.getPartitionOutput(KEY_4);
    final Location location4x = output4x.getLocation();
    try (Writer writer = new OutputStreamWriter(location4x.append("file").getOutputStream())) {
        writer.write("4x,4x\n");
    }
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                output4x.addPartition();
            }
        });
        Assert.fail("Expected tx to fail because hive partition for number=4 already exists");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    Assert.assertFalse(location4x.exists());
    // 6. partitionOutput.addPartition() succeeds but transaction fails
    final PartitionOutput output5x = pfs.getPartitionOutput(KEY_5);
    final Location location5x = output5x.getLocation();
    try (Writer writer = new OutputStreamWriter(location5x.append("file").getOutputStream())) {
        writer.write("5x,5x\n");
    }
    try {
        pfsManager.execute(new Runnable() {

            @Override
            public void run() {
                output5x.addPartition();
                throw new RuntimeException("fail the tx");
            }
        });
        Assert.fail("Expected tx to fail because it threw a runtime exception");
    } catch (TransactionFailureException e) {
    // expected
    }
    pfsValidator.validate();
    Assert.assertFalse(location5x.exists());
    // 7. mapreduce writing partition fails because location already exists
    mrManager.start(ImmutableMap.of(PFS_OUT, "1", "input.text", "1x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    // 8. mapreduce writing partition fails because partition already exists
    mrManager.start(ImmutableMap.of(PFS_OUT, "2", "input.text", "2x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_2).getLocation().exists());
    // 9. mapreduce writing partition fails because Hive partition already exists
    mrManager.start(ImmutableMap.of(PFS_OUT, "4", "input.text", "4x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_4).getLocation().exists());
    // 10. mapreduce writing dynamic partition fails because location already exists
    mrManager.start(ImmutableMap.of("input.text", "3x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
    // 11. mapreduce writing dynamic partition fails because partition already exists
    mrManager.start(ImmutableMap.of("input.text", "2x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_2).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
    // 12. mapreduce writing dynamic partition fails because Hive partition already exists
    mrManager.start(ImmutableMap.of("input.text", "0x 4x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_0).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_4).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
    // 13. multi-output mapreduce writing partition fails because location already exists
    // 13a. first output fails, other output must rollback 0 and 5
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, PFS_OUT, "1", "input.text", "0x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(other.getPartitionOutput(KEY_0).getLocation().exists());
    Assert.assertFalse(other.getPartitionOutput(KEY_5).getLocation().exists());
    // 13b. second output fails, first output must rollback 0 and 5
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, OTHER_OUT, "1", "input.text", "0x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_0).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
    // 14. multi-output mapreduce writing partition fails because partition already exists
    // 14a. first output fails, other output must rollback partition 5
    // TODO: bring this back when CDAP-8766 is fixed
    /*
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, PFS_OUT, "2", OTHER_OUT, "5", "input.text", "2x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(other.getPartitionOutput(KEY_5).getLocation().exists());
    */
    // 14b. second output fails, first output must rollback partition 5
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, PFS_OUT, "5", OTHER_OUT, "2", "input.text", "2x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
// 15. multi-output mapreduce writing partition fails because Hive partition already exists
// 15a. first output fails, other output must rollback partitions 0 and 5
// TODO: bring this back when CDAP-8766 is fixed
/*
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, PFS_OUT, "4", "input.text", "0x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(pfs.getPartitionOutput(KEY_4).getLocation().exists());
    Assert.assertFalse(other.getPartitionOutput(KEY_0).getLocation().exists());
    Assert.assertFalse(other.getPartitionOutput(KEY_5).getLocation().exists());
    // 15b. second output fails, first output must rollback partitions 0 and 5
    mrManager.start(ImmutableMap.of("output.datasets", BOTH, OTHER_OUT, "4", "input.text", "0x 5x"));
    mrManager.waitForRuns(ProgramRunStatus.FAILED, ++numRuns, 2, TimeUnit.MINUTES);
    pfsValidator.validate();
    otherValidator.validate();
    Assert.assertFalse(other.getPartitionOutput(KEY_4).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_0).getLocation().exists());
    Assert.assertFalse(pfs.getPartitionOutput(KEY_5).getLocation().exists());
    */
}

Example 7 with PartitionOutput

use of co.cask.cdap.api.dataset.lib.PartitionOutput in project cdap by caskdata.

the class SparkFileSetTestRun method testSparkWithPartitionedFileSet.

private void testSparkWithPartitionedFileSet(ApplicationManager applicationManager, String sparkProgram) throws Exception {
    DataSetManager<PartitionedFileSet> pfsManager = getDataset("pfs");
    PartitionedFileSet pfs = pfsManager.get();
    PartitionOutput partitionOutput = pfs.getPartitionOutput(PartitionKey.builder().addStringField("x", "nn").build());
    Location location = partitionOutput.getLocation();
    prepareFileInput(location);
    partitionOutput.addPartition();
    pfsManager.flush();
    Map<String, String> inputArgs = new HashMap<>();
    PartitionedFileSetArguments.setInputPartitionFilter(inputArgs, PartitionFilter.builder().addRangeCondition("x", "na", "nx").build());
    Map<String, String> outputArgs = new HashMap<>();
    final PartitionKey outputKey = PartitionKey.builder().addStringField("x", "xx").build();
    PartitionedFileSetArguments.setOutputPartitionKey(outputArgs, outputKey);
    Map<String, String> args = new HashMap<>();
    args.putAll(RuntimeArguments.addScope(Scope.DATASET, "pfs", inputArgs));
    args.putAll(RuntimeArguments.addScope(Scope.DATASET, "pfs", outputArgs));
    args.put("input", "pfs");
    args.put("output", "pfs");
    SparkManager sparkManager = applicationManager.getSparkManager(sparkProgram).start(args);
    sparkManager.waitForRun(ProgramRunStatus.COMPLETED, 10, TimeUnit.MINUTES);
    pfsManager.flush();
    PartitionDetail partition = pfs.getPartition(outputKey);
    Assert.assertNotNull(partition);
    validateFileOutput(partition.getLocation());
    // Cleanup after test completed
    location.delete(true);
    partition.getLocation().delete(true);
    pfs.dropPartition(partitionOutput.getPartitionKey());
    pfs.dropPartition(partition.getPartitionKey());
    pfsManager.flush();
}

Example 8 with PartitionOutput

use of co.cask.cdap.api.dataset.lib.PartitionOutput in project cdap by caskdata.

the class SparkFileSetTestRun method addTimePartition.

private void addTimePartition(DataSetManager<TimePartitionedFileSet> tpfsManager, long inputTime) throws IOException, TransactionFailureException, InterruptedException {
    TimePartitionedFileSet tpfs = tpfsManager.get();
    PartitionOutput partitionOutput = tpfs.getPartitionOutput(inputTime);
    Location location = partitionOutput.getLocation();
    prepareFileInput(location);
    partitionOutput.addPartition();
    tpfsManager.flush();
}

Example 9 with PartitionOutput

use of co.cask.cdap.api.dataset.lib.PartitionOutput in project cdap by caskdata.

the class PartitionedFileSetTest method testUpdateMetadata.

@Test
public void testUpdateMetadata() throws Exception {
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    dsFrameworkUtil.newTransactionExecutor((TransactionAware) dataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            PartitionOutput partitionOutput = dataset.getPartitionOutput(PARTITION_KEY);
            ImmutableMap<String, String> originalEntries = ImmutableMap.of("key1", "value1");
            partitionOutput.setMetadata(originalEntries);
            partitionOutput.addPartition();
            ImmutableMap<String, String> updatedMetadata = ImmutableMap.of("key2", "value2");
            dataset.addMetadata(PARTITION_KEY, updatedMetadata);
            PartitionDetail partitionDetail = dataset.getPartition(PARTITION_KEY);
            Assert.assertNotNull(partitionDetail);
            HashMap<String, String> combinedEntries = Maps.newHashMap();
            combinedEntries.putAll(originalEntries);
            combinedEntries.putAll(updatedMetadata);
            Assert.assertEquals(combinedEntries, partitionDetail.getMetadata().asMap());
            // adding an entry, for a key that already exists will throw an Exception
            try {
                dataset.addMetadata(PARTITION_KEY, "key2", "value3");
                Assert.fail("Expected not to be able to update an existing metadata entry");
            } catch (DataSetException expected) {
            }
            PartitionKey nonexistentPartitionKey = PartitionKey.builder().addIntField("i", 42).addLongField("l", 17L).addStringField("s", "nonexistent").build();
            try {
                // adding an entry, for a key that already exists will throw an Exception
                dataset.addMetadata(nonexistentPartitionKey, "key2", "value3");
                Assert.fail("Expected not to be able to add metadata for a nonexistent partition");
            } catch (DataSetException expected) {
            }
        }
    });
}

Example 10 with PartitionOutput

use of co.cask.cdap.api.dataset.lib.PartitionOutput in project cdap by caskdata.

the class PartitionedFileSetTest method testRollbackOnTransactionAbort.

@Test
public void testRollbackOnTransactionAbort() throws Exception {
    PartitionedFileSet pfs = dsFrameworkUtil.getInstance(pfsInstance);
    TransactionContext txContext = new TransactionContext(txClient, (TransactionAware) pfs);
    txContext.start();
    PartitionOutput output = pfs.getPartitionOutput(PARTITION_KEY);
    Location outputLocation = output.getLocation().append("file");
    Assert.assertFalse(outputLocation.exists());
    // this will create the file
    outputLocation.getOutputStream().close();
    Assert.assertTrue(outputLocation.exists());
    output.addPartition();
    Assert.assertNotNull(pfs.getPartition(PARTITION_KEY));
    Assert.assertTrue(pfs.getPartition(PARTITION_KEY).getLocation().exists());
    txContext.abort();
    // because the previous transaction aborted, the partition as well as the file will not exist
    txContext.start();
    Assert.assertNull(pfs.getPartition(PARTITION_KEY));
    Assert.assertFalse(outputLocation.exists());
    txContext.finish();
}