Examples with InputBatch org.apache.hudi.utilities.sources.InputBatch used on opensource projects

Example 1 with InputBatch

use of org.apache.hudi.utilities.sources.InputBatch in project hudi by apache.

the class TestAbstractDebeziumSource method testDebeziumEvents.

@ParameterizedTest
@MethodSource("testArguments")
public void testDebeziumEvents(Operation operation) throws Exception {
    String sourceClass = getSourceClass();
    // topic setup.
    testUtils.createTopic(TEST_TOPIC_NAME, 2);
    TypedProperties props = createPropsForJsonSource();
    SchemaProvider schemaProvider = new MockSchemaRegistryProvider(props, jsc, this);
    SourceFormatAdapter debeziumSource = new SourceFormatAdapter(UtilHelpers.createSource(sourceClass, props, jsc, sparkSession, schemaProvider, metrics));
    testUtils.sendMessages(TEST_TOPIC_NAME, new String[] { generateDebeziumEvent(operation).toString() });
    InputBatch<Dataset<Row>> fetch = debeziumSource.fetchNewDataInRowFormat(Option.empty(), 10);
    assertEquals(1, fetch.getBatch().get().count());
    // Ensure the before fields are picked for DELETE CDC Events,
    // and after fields are picked for INSERT and UPDATE CDC Events.
    final String fieldPrefix = (operation.equals(Operation.DELETE)) ? "before_" : "after_";
    assertTrue(fetch.getBatch().get().select("type").collectAsList().stream().allMatch(r -> r.getString(0).startsWith(fieldPrefix)));
    assertTrue(fetch.getBatch().get().select("type").collectAsList().stream().allMatch(r -> r.getString(0).startsWith(fieldPrefix)));
    // Validate DB specific meta fields
    validateMetaFields(fetch.getBatch().get());
}

Example 2 with InputBatch

use of org.apache.hudi.utilities.sources.InputBatch in project hudi by apache.

the class DistributedTestDataSource method fetchNewData.

@Override
protected InputBatch<JavaRDD<GenericRecord>> fetchNewData(Option<String> lastCkptStr, long sourceLimit) {
    int nextCommitNum = lastCkptStr.map(s -> Integer.parseInt(s) + 1).orElse(0);
    String instantTime = String.format("%05d", nextCommitNum);
    LOG.info("Source Limit is set to " + sourceLimit);
    // No new data.
    if (sourceLimit <= 0) {
        return new InputBatch<>(Option.empty(), instantTime);
    }
    TypedProperties newProps = new TypedProperties();
    newProps.putAll(props);
    // Set the maxUniqueRecords per partition for TestDataSource
    int maxUniqueRecords = props.getInteger(SourceConfigs.MAX_UNIQUE_RECORDS_PROP, SourceConfigs.DEFAULT_MAX_UNIQUE_RECORDS);
    String maxUniqueRecordsPerPartition = String.valueOf(Math.max(1, maxUniqueRecords / numTestSourcePartitions));
    newProps.setProperty(SourceConfigs.MAX_UNIQUE_RECORDS_PROP, maxUniqueRecordsPerPartition);
    int perPartitionSourceLimit = Math.max(1, (int) (sourceLimit / numTestSourcePartitions));
    JavaRDD<GenericRecord> avroRDD = sparkContext.parallelize(IntStream.range(0, numTestSourcePartitions).boxed().collect(Collectors.toList()), numTestSourcePartitions).mapPartitionsWithIndex((p, idx) -> {
        LOG.info("Initializing source with newProps=" + newProps);
        if (!dataGeneratorMap.containsKey(p)) {
            initDataGen(newProps, p);
        }
        return fetchNextBatch(newProps, perPartitionSourceLimit, instantTime, p).iterator();
    }, true);
    return new InputBatch<>(Option.of(avroRDD), instantTime);
}

Example 3 with InputBatch

use of org.apache.hudi.utilities.sources.InputBatch in project hudi by apache.

the class DeltaSync method fetchFromSource.

private Pair<SchemaProvider, Pair<String, JavaRDD<HoodieRecord>>> fetchFromSource(Option<String> resumeCheckpointStr) {
    final Option<JavaRDD<GenericRecord>> avroRDDOptional;
    final String checkpointStr;
    SchemaProvider schemaProvider;
    if (transformer.isPresent()) {
        // Transformation is needed. Fetch New rows in Row Format, apply transformation and then convert them
        // to generic records for writing
        InputBatch<Dataset<Row>> dataAndCheckpoint = formatAdapter.fetchNewDataInRowFormat(resumeCheckpointStr, cfg.sourceLimit);
        Option<Dataset<Row>> transformed = dataAndCheckpoint.getBatch().map(data -> transformer.get().apply(jssc, sparkSession, data, props));
        checkpointStr = dataAndCheckpoint.getCheckpointForNextBatch();
        boolean reconcileSchema = props.getBoolean(DataSourceWriteOptions.RECONCILE_SCHEMA().key());
        if (this.userProvidedSchemaProvider != null && this.userProvidedSchemaProvider.getTargetSchema() != null) {
            // If the target schema is specified through Avro schema,
            // pass in the schema for the Row-to-Avro conversion
            // to avoid nullability mismatch between Avro schema and Row schema
            avroRDDOptional = transformed.map(t -> HoodieSparkUtils.createRdd(t, HOODIE_RECORD_STRUCT_NAME, HOODIE_RECORD_NAMESPACE, reconcileSchema, Option.of(this.userProvidedSchemaProvider.getTargetSchema())).toJavaRDD());
            schemaProvider = this.userProvidedSchemaProvider;
        } else {
            // Use Transformed Row's schema if not overridden. If target schema is not specified
            // default to RowBasedSchemaProvider
            schemaProvider = transformed.map(r -> {
                // determine the targetSchemaProvider. use latestTableSchema if reconcileSchema is enabled.
                SchemaProvider targetSchemaProvider = null;
                if (reconcileSchema) {
                    targetSchemaProvider = UtilHelpers.createLatestSchemaProvider(r.schema(), jssc, fs, cfg.targetBasePath);
                } else {
                    targetSchemaProvider = UtilHelpers.createRowBasedSchemaProvider(r.schema(), props, jssc);
                }
                return (SchemaProvider) new DelegatingSchemaProvider(props, jssc, dataAndCheckpoint.getSchemaProvider(), targetSchemaProvider);
            }).orElse(dataAndCheckpoint.getSchemaProvider());
            avroRDDOptional = transformed.map(t -> HoodieSparkUtils.createRdd(t, HOODIE_RECORD_STRUCT_NAME, HOODIE_RECORD_NAMESPACE, reconcileSchema, Option.ofNullable(schemaProvider.getTargetSchema())).toJavaRDD());
        }
    } else {
        // Pull the data from the source & prepare the write
        InputBatch<JavaRDD<GenericRecord>> dataAndCheckpoint = formatAdapter.fetchNewDataInAvroFormat(resumeCheckpointStr, cfg.sourceLimit);
        avroRDDOptional = dataAndCheckpoint.getBatch();
        checkpointStr = dataAndCheckpoint.getCheckpointForNextBatch();
        schemaProvider = dataAndCheckpoint.getSchemaProvider();
    }
    if (!cfg.allowCommitOnNoCheckpointChange && Objects.equals(checkpointStr, resumeCheckpointStr.orElse(null))) {
        LOG.info("No new data, source checkpoint has not changed. Nothing to commit. Old checkpoint=(" + resumeCheckpointStr + "). New Checkpoint=(" + checkpointStr + ")");
        String commitActionType = CommitUtils.getCommitActionType(cfg.operation, HoodieTableType.valueOf(cfg.tableType));
        hoodieMetrics.updateMetricsForEmptyData(commitActionType);
        return null;
    }
    jssc.setJobGroup(this.getClass().getSimpleName(), "Checking if input is empty");
    if ((!avroRDDOptional.isPresent()) || (avroRDDOptional.get().isEmpty())) {
        LOG.info("No new data, perform empty commit.");
        return Pair.of(schemaProvider, Pair.of(checkpointStr, jssc.emptyRDD()));
    }
    boolean shouldCombine = cfg.filterDupes || cfg.operation.equals(WriteOperationType.UPSERT);
    JavaRDD<GenericRecord> avroRDD = avroRDDOptional.get();
    JavaRDD<HoodieRecord> records = avroRDD.map(gr -> {
        HoodieRecordPayload payload = shouldCombine ? DataSourceUtils.createPayload(cfg.payloadClassName, gr, (Comparable) HoodieAvroUtils.getNestedFieldVal(gr, cfg.sourceOrderingField, false, props.getBoolean(KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.key(), Boolean.parseBoolean(KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.defaultValue())))) : DataSourceUtils.createPayload(cfg.payloadClassName, gr);
        return new HoodieAvroRecord<>(keyGenerator.getKey(gr), payload);
    });
    return Pair.of(schemaProvider, Pair.of(checkpointStr, records));
}