Examples with DataWriter org.apache.spark.sql.sources.v2.writer.DataWriter used on opensource projects

Example 1 with DataWriter

use of org.apache.spark.sql.sources.v2.writer.DataWriter in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testDataSourceWriterInternal.

private void testDataSourceWriterInternal(Map<String, String> extraMetadata, Map<String, String> expectedExtraMetadata, boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    String instantTime = "001";
    // init writer
    HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(extraMetadata), populateMetaFields, false);
    DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
    String[] partitionPaths = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS;
    List<String> partitionPathsAbs = new ArrayList<>();
    for (String partitionPath : partitionPaths) {
        partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
    }
    int size = 10 + RANDOM.nextInt(1000);
    int batches = 2;
    Dataset<Row> totalInputRows = null;
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
        if (totalInputRows == null) {
            totalInputRows = inputRows;
        } else {
            totalInputRows = totalInputRows.union(inputRows);
        }
    }
    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify output
    assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    // verify extra metadata
    Option<HoodieCommitMetadata> commitMetadataOption = HoodieClientTestUtils.getCommitMetadataForLatestInstant(metaClient);
    assertTrue(commitMetadataOption.isPresent());
    Map<String, String> actualExtraMetadata = new HashMap<>();
    commitMetadataOption.get().getExtraMetadata().entrySet().stream().filter(entry -> !entry.getKey().equals(HoodieCommitMetadata.SCHEMA_KEY)).forEach(entry -> actualExtraMetadata.put(entry.getKey(), entry.getValue()));
    assertEquals(actualExtraMetadata, expectedExtraMetadata);
}

Example 2 with DataWriter

use of org.apache.spark.sql.sources.v2.writer.DataWriter in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testAbort.

/**
 * Tests that DataSourceWriter.abort() will abort the written records of interest write and commit batch1 write and abort batch2 Read of entire dataset should show only records from batch1.
 * commit batch1
 * abort batch2
 * verify only records from batch1 is available to read
 */
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testAbort(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    String instantTime0 = "00" + 0;
    // init writer
    HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
    DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
    List<String> partitionPaths = Arrays.asList(HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS);
    List<String> partitionPathsAbs = new ArrayList<>();
    for (String partitionPath : partitionPaths) {
        partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
    }
    int size = 10 + RANDOM.nextInt(100);
    int batches = 1;
    Dataset<Row> totalInputRows = null;
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
        if (totalInputRows == null) {
            totalInputRows = inputRows;
        } else {
            totalInputRows = totalInputRows.union(inputRows);
        }
    }
    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    // commit 1st batch
    dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify rows
    assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    // 2nd batch. abort in the end
    String instantTime1 = "00" + 1;
    dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
    writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(1, RANDOM.nextLong(), RANDOM.nextLong());
    for (int j = 0; j < batches; j++) {
        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
        writeRows(inputRows, writer);
    }
    commitMetadata = (HoodieWriterCommitMessage) writer.commit();
    commitMessages = new ArrayList<>();
    commitMessages.add(commitMetadata);
    // commit 1st batch
    dataSourceInternalWriter.abort(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
    metaClient.reloadActiveTimeline();
    result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
    // verify rows
    // only rows from first batch should be present
    assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
}

Example 3 with DataWriter

use of org.apache.spark.sql.sources.v2.writer.DataWriter in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testMultipleDataSourceWrites.

@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testMultipleDataSourceWrites(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    int partitionCounter = 0;
    // execute N rounds
    for (int i = 0; i < 2; i++) {
        String instantTime = "00" + i;
        // init writer
        HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
        List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
        Dataset<Row> totalInputRows = null;
        DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
        int size = 10 + RANDOM.nextInt(1000);
        // one batch per partition
        int batches = 2;
        for (int j = 0; j < batches; j++) {
            String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
            Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
            writeRows(inputRows, writer);
            if (totalInputRows == null) {
                totalInputRows = inputRows;
            } else {
                totalInputRows = totalInputRows.union(inputRows);
            }
        }
        HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
        commitMessages.add(commitMetadata);
        dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
        metaClient.reloadActiveTimeline();
        Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime, populateMetaFields);
        // verify output
        assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
        assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    }
}

Example 4 with DataWriter

use of org.apache.spark.sql.sources.v2.writer.DataWriter in project hudi by apache.

the class TestHoodieDataSourceInternalWriter method testLargeWrites.

// takes up lot of running time with CI.
@Disabled
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testLargeWrites(boolean populateMetaFields) throws Exception {
    // init config and table
    HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
    int partitionCounter = 0;
    // execute N rounds
    for (int i = 0; i < 3; i++) {
        String instantTime = "00" + i;
        // init writer
        HoodieDataSourceInternalWriter dataSourceInternalWriter = new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
        List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
        Dataset<Row> totalInputRows = null;
        DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
        int size = 10000 + RANDOM.nextInt(10000);
        // one batch per partition
        int batches = 3;
        for (int j = 0; j < batches; j++) {
            String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
            Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
            writeRows(inputRows, writer);
            if (totalInputRows == null) {
                totalInputRows = inputRows;
            } else {
                totalInputRows = totalInputRows.union(inputRows);
            }
        }
        HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
        commitMessages.add(commitMetadata);
        dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
        metaClient.reloadActiveTimeline();
        Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime, populateMetaFields);
        // verify output
        assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
        assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
    }
}