org.apache.spark.sql.sources.v2.reader.InputPartition Java Examples

/**
 * This is called in the Spark Driver when data is to be materialized into {@link ColumnarBatch}
 */
@Override
public List<InputPartition<ColumnarBatch>> planBatchInputPartitions() {
  Preconditions.checkState(enableBatchRead(), "Batched reads not enabled");
  Preconditions.checkState(batchSize > 0, "Invalid batch size");
  String tableSchemaString = SchemaParser.toJson(table.schema());
  String expectedSchemaString = SchemaParser.toJson(lazySchema());
  String nameMappingString = table.properties().get(DEFAULT_NAME_MAPPING);

  List<InputPartition<ColumnarBatch>> readTasks = Lists.newArrayList();
  for (CombinedScanTask task : tasks()) {
    readTasks.add(new ReadTask<>(
        task, tableSchemaString, expectedSchemaString, nameMappingString, io, encryptionManager, caseSensitive,
        localityPreferred, new BatchReaderFactory(batchSize)));
  }
  LOG.info("Batching input partitions with {} tasks.", readTasks.size());

  return readTasks;
}
 

/**
 * This is called in the Spark Driver when data is to be materialized into {@link InternalRow}
 */
@Override
public List<InputPartition<InternalRow>> planInputPartitions() {
  String tableSchemaString = SchemaParser.toJson(table.schema());
  String expectedSchemaString = SchemaParser.toJson(lazySchema());
  String nameMappingString = table.properties().get(DEFAULT_NAME_MAPPING);

  List<InputPartition<InternalRow>> readTasks = Lists.newArrayList();
  for (CombinedScanTask task : tasks()) {
    readTasks.add(new ReadTask<>(
        task, tableSchemaString, expectedSchemaString, nameMappingString, io, encryptionManager, caseSensitive,
        localityPreferred, InternalRowReaderFactory.INSTANCE));
  }

  return readTasks;
}
 

@Test
public void testUnpartitionedIDFilters() {
  DataSourceOptions options = new DataSourceOptions(ImmutableMap.of(
      "path", unpartitioned.toString())
  );

  IcebergSource source = new IcebergSource();

  for (int i = 0; i < 10; i += 1) {
    DataSourceReader reader = source.createReader(options);

    pushFilters(reader, EqualTo.apply("id", i));

    List<InputPartition<InternalRow>> tasks = reader.planInputPartitions();
    Assert.assertEquals("Should only create one task for a small file", 1, tasks.size());

    // validate row filtering
    assertEqualsSafe(SCHEMA.asStruct(), expected(i),
        read(unpartitioned.toString(), "id = " + i));
  }
}
 

@Test
public void testUnpartitionedTimestampFilter() {
  DataSourceOptions options = new DataSourceOptions(ImmutableMap.of(
      "path", unpartitioned.toString())
  );

  IcebergSource source = new IcebergSource();

  DataSourceReader reader = source.createReader(options);

  pushFilters(reader, LessThan.apply("ts", "2017-12-22T00:00:00+00:00"));

  List<InputPartition<InternalRow>> tasks = reader.planInputPartitions();
  Assert.assertEquals("Should only create one task for a small file", 1, tasks.size());

  assertEqualsSafe(SCHEMA.asStruct(), expected(5, 6, 7, 8, 9),
      read(unpartitioned.toString(), "ts < cast('2017-12-22 00:00:00+00:00' as timestamp)"));
}
 

private List<InputPartition<ColumnarBatch>> planBatchInputPartitionsParallel() {

    try (FlightClient client = clientFactory.apply()) {
      FlightInfo info = client.getInfo(FlightDescriptor.command(sql.getBytes()));
      return planBatchInputPartitionsSerial(info);
    } catch (InterruptedException e) {
      throw new RuntimeException(e);
    }
  }
 

private List<InputPartition<ColumnarBatch>> planBatchInputPartitionsSerial(FlightInfo info) {
  LOGGER.warn("planning partitions for endpoints {}", Joiner.on(", ").join(info.getEndpoints().stream().map(e -> e.getLocations().get(0).getUri().toString()).collect(Collectors.toList())));
  List<InputPartition<ColumnarBatch>> batches = info.getEndpoints().stream().map(endpoint -> {
    Location location = (endpoint.getLocations().isEmpty()) ?
      Location.forGrpcInsecure(defaultLocation.getUri().getHost(), defaultLocation.getUri().getPort()) :
      endpoint.getLocations().get(0);
    FactoryOptions options = dataSourceOptions.value().copy(location, endpoint.getTicket().getBytes());
    LOGGER.warn("X1 {}", dataSourceOptions.value());
    return new FlightDataReaderFactory(lazySparkContext().broadcast(options));
  }).collect(Collectors.toList());
  LOGGER.info("Created {} batches from arrow endpoints", batches.size());
  return batches;
}
 

@Test
public void testUnpartitionedCaseInsensitiveIDFilters() {
  DataSourceOptions options = new DataSourceOptions(ImmutableMap.of(
      "path", unpartitioned.toString())
  );

  // set spark.sql.caseSensitive to false
  String caseSensitivityBeforeTest = TestFilteredScan.spark.conf().get("spark.sql.caseSensitive");
  TestFilteredScan.spark.conf().set("spark.sql.caseSensitive", "false");

  try {
    IcebergSource source = new IcebergSource();

    for (int i = 0; i < 10; i += 1) {
      DataSourceReader reader = source.createReader(options);

      pushFilters(reader, EqualTo.apply("ID", i)); // note lower(ID) == lower(id), so there must be a match

      List<InputPartition<InternalRow>> tasks = reader.planInputPartitions();
      Assert.assertEquals("Should only create one task for a small file", 1, tasks.size());

      // validate row filtering
      assertEqualsSafe(SCHEMA.asStruct(), expected(i),
          read(unpartitioned.toString(), "id = " + i));
    }
  } finally {
    // return global conf to previous state
    TestFilteredScan.spark.conf().set("spark.sql.caseSensitive", caseSensitivityBeforeTest);
  }
}
 

@Test
public void testBucketPartitionedIDFilters() {
  File location = buildPartitionedTable("bucketed_by_id", BUCKET_BY_ID, "bucket4", "id");

  DataSourceOptions options = new DataSourceOptions(ImmutableMap.of(
      "path", location.toString())
  );

  IcebergSource source = new IcebergSource();
  DataSourceReader unfiltered = source.createReader(options);
  Assert.assertEquals("Unfiltered table should created 4 read tasks",
      4, unfiltered.planInputPartitions().size());

  for (int i = 0; i < 10; i += 1) {
    DataSourceReader reader = source.createReader(options);

    pushFilters(reader, EqualTo.apply("id", i));

    List<InputPartition<InternalRow>> tasks = reader.planInputPartitions();

    // validate predicate push-down
    Assert.assertEquals("Should create one task for a single bucket", 1, tasks.size());

    // validate row filtering
    assertEqualsSafe(SCHEMA.asStruct(), expected(i), read(location.toString(), "id = " + i));
  }
}
 

@Override
public List<InputPartition<ColumnarBatch>> planBatchInputPartitions() {
  return planBatchInputPartitionsParallel();
}