Java Code Examples for org.apache.parquet.hadoop.metadata.BlockMetaData#getColumns()

private static boolean dictionaryPredicatesMatch(Predicate parquetPredicate, BlockMetaData blockMetadata, ParquetDataSource dataSource, Map<List<String>, RichColumnDescriptor> descriptorsByPath, TupleDomain<ColumnDescriptor> parquetTupleDomain)
{
    for (ColumnChunkMetaData columnMetaData : blockMetadata.getColumns()) {
        RichColumnDescriptor descriptor = descriptorsByPath.get(Arrays.asList(columnMetaData.getPath().toArray()));
        if (descriptor != null) {
            if (isOnlyDictionaryEncodingPages(columnMetaData) && isColumnPredicate(descriptor, parquetTupleDomain)) {
                byte[] buffer = new byte[toIntExact(columnMetaData.getTotalSize())];
                dataSource.readFully(columnMetaData.getStartingPos(), buffer);
                //  Early abort, predicate already filters block so no more dictionaries need be read
                if (!parquetPredicate.matches(new DictionaryDescriptor(descriptor, readDictionaryPage(buffer, columnMetaData.getCodec())))) {
                    return false;
                }
            }
        }
    }
    return true;
}
 

private void checkDeltaByteArrayProblem(FileMetaData meta, Configuration conf, BlockMetaData block) {
  // splitting files?
  if (conf.getBoolean(ParquetInputFormat.SPLIT_FILES, true)) {
    // this is okay if not using DELTA_BYTE_ARRAY with the bug
    Set<Encoding> encodings = new HashSet<Encoding>();
    for (ColumnChunkMetaData column : block.getColumns()) {
      encodings.addAll(column.getEncodings());
    }
    for (Encoding encoding : encodings) {
      if (CorruptDeltaByteArrays.requiresSequentialReads(meta.getCreatedBy(), encoding)) {
        throw new ParquetDecodingException("Cannot read data due to " +
            "PARQUET-246: to read safely, set " + SPLIT_FILES + " to false");
      }
    }
  }
}
 

private boolean eval(MessageType fileSchema, BlockMetaData rowGroup) {
  if (rowGroup.getRowCount() <= 0) {
    return ROWS_CANNOT_MATCH;
  }

  this.stats = Maps.newHashMap();
  this.valueCounts = Maps.newHashMap();
  this.conversions = Maps.newHashMap();
  for (ColumnChunkMetaData col : rowGroup.getColumns()) {
    PrimitiveType colType = fileSchema.getType(col.getPath().toArray()).asPrimitiveType();
    if (colType.getId() != null) {
      int id = colType.getId().intValue();
      stats.put(id, col.getStatistics());
      valueCounts.put(id, col.getValueCount());
      conversions.put(id, ParquetConversions.converterFromParquet(colType));
    }
  }

  return ExpressionVisitors.visitEvaluator(expr, this);
}
 

private boolean eval(MessageType fileSchema, BlockMetaData rowGroup) {
  if (rowGroup.getRowCount() <= 0) {
    return ROWS_CANNOT_MATCH;
  }

  this.stats = Maps.newHashMap();
  this.valueCounts = Maps.newHashMap();
  this.conversions = Maps.newHashMap();
  for (ColumnChunkMetaData col : rowGroup.getColumns()) {
    PrimitiveType colType = fileSchema.getType(col.getPath().toArray()).asPrimitiveType();
    if (colType.getId() != null) {
      int id = colType.getId().intValue();
      stats.put(id, col.getStatistics());
      valueCounts.put(id, col.getValueCount());
      conversions.put(id, converterFromParquet(colType));
    }
  }

  return ExpressionVisitors.visit(expr, this);
}
 

public void processBlocks(TransParquetFileReader reader, ParquetFileWriter writer, ParquetMetadata meta, MessageType schema,
                           String createdBy, CompressionCodecName codecName) throws IOException {
  int blockIndex = 0;
  PageReadStore store = reader.readNextRowGroup();
  while (store != null) {
    writer.startBlock(store.getRowCount());
    BlockMetaData blockMetaData = meta.getBlocks().get(blockIndex);
    List<ColumnChunkMetaData> columnsInOrder = blockMetaData.getColumns();
    Map<ColumnPath, ColumnDescriptor> descriptorsMap = schema.getColumns().stream().collect(
      Collectors.toMap(x -> ColumnPath.get(x.getPath()), x -> x));
    for (int i = 0; i < columnsInOrder.size(); i += 1) {
      ColumnChunkMetaData chunk = columnsInOrder.get(i);
      ColumnReadStoreImpl crstore = new ColumnReadStoreImpl(store, new DummyGroupConverter(), schema, createdBy);
      ColumnDescriptor columnDescriptor = descriptorsMap.get(chunk.getPath());
      writer.startColumn(columnDescriptor, crstore.getColumnReader(columnDescriptor).getTotalValueCount(), codecName);
      processChunk(reader, writer, chunk, createdBy, codecName);
      writer.endColumn();
    }
    writer.endBlock();
    store = reader.readNextRowGroup();
    blockIndex++;
  }
}
 

private static void add(ParquetMetadata footer) {
  for (BlockMetaData blockMetaData : footer.getBlocks()) {
    ++ blockCount;
    MessageType schema = footer.getFileMetaData().getSchema();
    recordCount += blockMetaData.getRowCount();
    List<ColumnChunkMetaData> columns = blockMetaData.getColumns();
    for (ColumnChunkMetaData columnMetaData : columns) {
      ColumnDescriptor desc = schema.getColumnDescription(columnMetaData.getPath().toArray());
      add(
          desc,
          columnMetaData.getValueCount(),
          columnMetaData.getTotalSize(),
          columnMetaData.getTotalUncompressedSize(),
          columnMetaData.getEncodings(),
          columnMetaData.getStatistics());
    }
  }
}
 

public Map<String, Long> getColumnSizeInBytes(Path inputFile) throws IOException {
  Map<String, Long> colSizes = new HashMap<>();
  ParquetMetadata pmd = ParquetFileReader.readFooter(new Configuration(), inputFile, ParquetMetadataConverter.NO_FILTER);

  for (BlockMetaData block : pmd.getBlocks()) {
    for (ColumnChunkMetaData column : block.getColumns()) {
      String colName = column.getPath().toDotString();
      colSizes.put(colName, column.getTotalSize() + colSizes.getOrDefault(colName, 0L));
    }
  }

  return colSizes;
}
 

public BloomFilterReader(ParquetFileReader fileReader, BlockMetaData block) {
  this.reader = fileReader;
  this.columns = new HashMap<>();
  for (ColumnChunkMetaData column : block.getColumns()) {
    columns.put(column.getPath(), column);
  }
}
 

private static long end(List<BlockMetaData> blocks, String requestedSchema) {
  MessageType requested = MessageTypeParser.parseMessageType(requestedSchema);
  long length = 0;

  for (BlockMetaData block : blocks) {
    List<ColumnChunkMetaData> columns = block.getColumns();
    for (ColumnChunkMetaData column : columns) {
      if (requested.containsPath(column.getPath().toArray())) {
        length += column.getTotalSize();
      }
    }
  }
  return length;
}
 

public ParquetInputSplit getParquetInputSplit(FileStatus fileStatus, String requestedSchema, Map<String, String> readSupportMetadata) throws IOException {
  MessageType requested = MessageTypeParser.parseMessageType(requestedSchema);
  long length = 0;

  for (BlockMetaData block : this.getRowGroups()) {
    List<ColumnChunkMetaData> columns = block.getColumns();
    for (ColumnChunkMetaData column : columns) {
      if (requested.containsPath(column.getPath().toArray())) {
        length += column.getTotalSize();
      }
    }
  }

  BlockMetaData lastRowGroup = this.getRowGroups().get(this.getRowGroupCount() - 1);
  long end = lastRowGroup.getStartingPos() + lastRowGroup.getTotalByteSize();

  long[] rowGroupOffsets = new long[this.getRowGroupCount()];
  for (int i = 0; i < rowGroupOffsets.length; i++) {
    rowGroupOffsets[i] = this.getRowGroups().get(i).getStartingPos();
  }

  return new ParquetInputSplit(
          fileStatus.getPath(),
          hdfsBlock.getOffset(),
          end,
          length,
          hdfsBlock.getHosts(),
          rowGroupOffsets
  );
}
 

private ColumnIndexStoreImpl(ParquetFileReader reader, BlockMetaData block, Set<ColumnPath> paths) {
  // TODO[GS]: Offset index for every paths will be required; pre-read the consecutive ones at once?
  // TODO[GS]: Pre-read column index based on filter?
  this.reader = reader;
  Map<ColumnPath, IndexStore> store = new HashMap<>();
  for (ColumnChunkMetaData column : block.getColumns()) {
    ColumnPath path = column.getPath();
    if (paths.contains(path)) {
      store.put(path, new IndexStoreImpl(column));
    }
  }
  this.store = store;
}
 

/**
 * Instantiate a new DictionaryPageReader.
 *
 * @param reader The target ParquetFileReader
 * @param block The target BlockMetaData
 *
 * @throws NullPointerException if {@code reader} or {@code block} is
 *           {@code null}
 */
DictionaryPageReader(ParquetFileReader reader, BlockMetaData block) {
  this.reader = Objects.requireNonNull(reader);
  this.columns = new HashMap<>();
  this.dictionaryPageCache = new ConcurrentHashMap<>();

  for (ColumnChunkMetaData column : block.getColumns()) {
    columns.put(column.getPath().toDotString(), column);
  }
}
 

private void printRowGroup(Logger console, int index, BlockMetaData rowGroup, MessageType schema) {
  long start = rowGroup.getStartingPos();
  long rowCount = rowGroup.getRowCount();
  long compressedSize = rowGroup.getCompressedSize();
  long uncompressedSize = rowGroup.getTotalByteSize();
  String filePath = rowGroup.getPath();

  console.info(String.format("\nRow group %d:  count: %d  %s records  start: %d  total: %s%s\n%s",
      index, rowCount,
      humanReadable(((float) compressedSize) / rowCount),
      start, humanReadable(compressedSize),
      filePath != null ? " path: " + filePath : "",
      new TextStringBuilder(80).appendPadding(80, '-')));

  int size = maxSize(Iterables.transform(rowGroup.getColumns(),
      new Function<ColumnChunkMetaData, String>() {
        @Override
        public String apply(@Nullable ColumnChunkMetaData input) {
          return input == null ? "" : input.getPath().toDotString();
        }
      }));

  console.info(String.format("%-" + size + "s  %-9s %-9s %-9s %-10s %-7s %s",
      "", "type", "encodings", "count", "avg size", "nulls", "min / max"));
  for (ColumnChunkMetaData column : rowGroup.getColumns()) {
    printColumnChunk(console, size, column, schema);
  }
}
 

/**
 * Return dictionary per row group for all binary columns in given parquet file.
 * @param fs filesystem object.
 * @param filePath parquet file to scan
 * @return pair of dictionaries found for binary fields and list of binary fields which are not dictionary encoded.
 * @throws IOException
 */
public static Pair<Map<ColumnDescriptor, Dictionary>, Set<ColumnDescriptor>> readDictionaries(FileSystem fs, Path filePath, CompressionCodecFactory codecFactory) throws IOException {
  // Passing the max footer length is not required in this case as the parquet reader would already have failed.
  final ParquetMetadata parquetMetadata = SingletonParquetFooterCache.readFooter(fs, filePath, ParquetMetadataConverter.NO_FILTER,
    ExecConstants.PARQUET_MAX_FOOTER_LEN_VALIDATOR.getDefault().getNumVal());
  if (parquetMetadata.getBlocks().size() > 1) {
    throw new IOException(
      format("Global dictionaries can only be built on a parquet file with a single row group, found %d row groups for file %s",
        parquetMetadata.getBlocks().size(), filePath));
  }
  final BlockMetaData rowGroupMetadata = parquetMetadata.getBlocks().get(0);
  final Map<ColumnPath, ColumnDescriptor> columnDescriptorMap = Maps.newHashMap();

  for (ColumnDescriptor columnDescriptor : parquetMetadata.getFileMetaData().getSchema().getColumns()) {
    columnDescriptorMap.put(ColumnPath.get(columnDescriptor.getPath()), columnDescriptor);
  }

  final Set<ColumnDescriptor> columnsToSkip = Sets.newHashSet(); // columns which are found in parquet file but are not dictionary encoded
  final Map<ColumnDescriptor, Dictionary> dictionaries = Maps.newHashMap();
  try(final FSInputStream in = fs.open(filePath)) {
    for (ColumnChunkMetaData columnChunkMetaData : rowGroupMetadata.getColumns()) {
      if (isBinaryType(columnChunkMetaData.getType())) {
        final ColumnDescriptor column = columnDescriptorMap.get(columnChunkMetaData.getPath());
        // if first page is dictionary encoded then load dictionary, otherwise skip this column.
        final PageHeaderWithOffset pageHeader = columnChunkMetaData.getPageHeaders().get(0);
        if (PageType.DICTIONARY_PAGE == pageHeader.getPageHeader().getType()) {
          dictionaries.put(column, readDictionary(in, column, pageHeader, codecFactory.getDecompressor(columnChunkMetaData.getCodec())));
        } else {
          columnsToSkip.add(column);
        }
      }
    }
  }
  return new ImmutablePair<>(dictionaries, columnsToSkip);
}
 

private ColumnChunkMetaData getColumnChunkMetaData(BlockMetaData blockMetaData, ColumnDescriptor columnDescriptor)
        throws IOException
{
    for (ColumnChunkMetaData metadata : blockMetaData.getColumns()) {
        if (metadata.getPath().equals(ColumnPath.get(columnDescriptor.getPath()))) {
            return metadata;
        }
    }
    throw new ParquetCorruptionException("Metadata is missing for column: %s", columnDescriptor);
}
 

@Test
public void test() throws Exception {
  Configuration conf = new Configuration();
  Path root = new Path("target/tests/TestParquetWriter/");
  enforceEmptyDir(conf, root);
  MessageType schema = parseMessageType(
      "message test { "
      + "required binary binary_field; "
      + "required int32 int32_field; "
      + "required int64 int64_field; "
      + "required boolean boolean_field; "
      + "required float float_field; "
      + "required double double_field; "
      + "required fixed_len_byte_array(3) flba_field; "
      + "required int96 int96_field; "
      + "} ");
  GroupWriteSupport.setSchema(schema, conf);
  SimpleGroupFactory f = new SimpleGroupFactory(schema);
  Map<String, Encoding> expected = new HashMap<String, Encoding>();
  expected.put("10-" + PARQUET_1_0, PLAIN_DICTIONARY);
  expected.put("1000-" + PARQUET_1_0, PLAIN);
  expected.put("10-" + PARQUET_2_0, RLE_DICTIONARY);
  expected.put("1000-" + PARQUET_2_0, DELTA_BYTE_ARRAY);
  for (int modulo : asList(10, 1000)) {
    for (WriterVersion version : WriterVersion.values()) {
      Path file = new Path(root, version.name() + "_" + modulo);
      ParquetWriter<Group> writer = new ParquetWriter<Group>(
          file,
          new GroupWriteSupport(),
          UNCOMPRESSED, 1024, 1024, 512, true, false, version, conf);
      for (int i = 0; i < 1000; i++) {
        writer.write(
            f.newGroup()
            .append("binary_field", "test" + (i % modulo))
            .append("int32_field", 32)
            .append("int64_field", 64l)
            .append("boolean_field", true)
            .append("float_field", 1.0f)
            .append("double_field", 2.0d)
            .append("flba_field", "foo")
            .append("int96_field", Binary.fromConstantByteArray(new byte[12])));
      }
      writer.close();
      ParquetReader<Group> reader = ParquetReader.builder(new GroupReadSupport(), file).withConf(conf).build();
      for (int i = 0; i < 1000; i++) {
        Group group = reader.read();
        assertEquals("test" + (i % modulo), group.getBinary("binary_field", 0).toStringUsingUTF8());
        assertEquals(32, group.getInteger("int32_field", 0));
        assertEquals(64l, group.getLong("int64_field", 0));
        assertEquals(true, group.getBoolean("boolean_field", 0));
        assertEquals(1.0f, group.getFloat("float_field", 0), 0.001);
        assertEquals(2.0d, group.getDouble("double_field", 0), 0.001);
        assertEquals("foo", group.getBinary("flba_field", 0).toStringUsingUTF8());
        assertEquals(Binary.fromConstantByteArray(new byte[12]),
            group.getInt96("int96_field",0));
      }
      reader.close();
      ParquetMetadata footer = readFooter(conf, file, NO_FILTER);
      for (BlockMetaData blockMetaData : footer.getBlocks()) {
        for (ColumnChunkMetaData column : blockMetaData.getColumns()) {
          if (column.getPath().toDotString().equals("binary_field")) {
            String key = modulo + "-" + version;
            Encoding expectedEncoding = expected.get(key);
            assertTrue(
                key + ":" + column.getEncodings() + " should contain " + expectedEncoding,
                column.getEncodings().contains(expectedEncoding));
          }
        }
      }
      assertEquals("Object model property should be example",
          "example", footer.getFileMetaData().getKeyValueMetaData()
              .get(ParquetWriter.OBJECT_MODEL_NAME_PROP));
    }
  }
}
 

/**
 * Reads all the columns requested from the row group at the current file position. It may skip specific pages based
 * on the column indexes according to the actual filter. As the rows are not aligned among the pages of the different
 * columns row synchronization might be required. See the documentation of the class SynchronizingColumnReader for
 * details.
 *
 * @return the PageReadStore which can provide PageReaders for each column
 * @throws IOException
 *           if any I/O error occurs while reading
 */
public PageReadStore readNextFilteredRowGroup() throws IOException {
  if (currentBlock == blocks.size()) {
    return null;
  }
  if (!options.useColumnIndexFilter()) {
    return readNextRowGroup();
  }
  BlockMetaData block = blocks.get(currentBlock);
  if (block.getRowCount() == 0) {
    throw new RuntimeException("Illegal row group of 0 rows");
  }
  ColumnIndexStore ciStore = getColumnIndexStore(currentBlock);
  RowRanges rowRanges = getRowRanges(currentBlock);
  long rowCount = rowRanges.rowCount();
  if (rowCount == 0) {
    // There are no matching rows -> skipping this row-group
    advanceToNextBlock();
    return readNextFilteredRowGroup();
  }
  if (rowCount == block.getRowCount()) {
    // All rows are matching -> fall back to the non-filtering path
    return readNextRowGroup();
  }

  this.currentRowGroup = new ColumnChunkPageReadStore(rowRanges);
  // prepare the list of consecutive parts to read them in one scan
  ChunkListBuilder builder = new ChunkListBuilder();
  List<ConsecutivePartList> allParts = new ArrayList<ConsecutivePartList>();
  ConsecutivePartList currentParts = null;
  for (ColumnChunkMetaData mc : block.getColumns()) {
    ColumnPath pathKey = mc.getPath();
    ColumnDescriptor columnDescriptor = paths.get(pathKey);
    if (columnDescriptor != null) {
      OffsetIndex offsetIndex = ciStore.getOffsetIndex(mc.getPath());

      OffsetIndex filteredOffsetIndex = filterOffsetIndex(offsetIndex, rowRanges,
          block.getRowCount());
      for (OffsetRange range : calculateOffsetRanges(filteredOffsetIndex, mc, offsetIndex.getOffset(0))) {
        BenchmarkCounter.incrementTotalBytes(range.getLength());
        long startingPos = range.getOffset();
        // first part or not consecutive => new list
        if (currentParts == null || currentParts.endPos() != startingPos) {
          currentParts = new ConsecutivePartList(startingPos);
          allParts.add(currentParts);
        }
        ChunkDescriptor chunkDescriptor = new ChunkDescriptor(columnDescriptor, mc, startingPos,
            (int) range.getLength());
        currentParts.addChunk(chunkDescriptor);
        builder.setOffsetIndex(chunkDescriptor, filteredOffsetIndex);
      }
    }
  }
  // actually read all the chunks
  for (ConsecutivePartList consecutiveChunks : allParts) {
    consecutiveChunks.readAll(f, builder);
  }
  for (Chunk chunk : builder.build()) {
    currentRowGroup.addColumn(chunk.descriptor.col, chunk.readAllPages());
  }

  // avoid re-reading bytes the dictionary reader is used after this call
  if (nextDictionaryReader != null) {
    nextDictionaryReader.setRowGroup(currentRowGroup);
  }

  advanceToNextBlock();

  return currentRowGroup;
}
 

private ParquetFileMetadata getParquetFileMetadata(FileAttributes file, AtomicInteger currentNumSplits, long maxSplits) throws IOException {
  final ParquetMetadata metadata =
    SingletonParquetFooterCache.readFooter(fs, file, ParquetMetadataConverter.NO_FILTER, maxFooterLength);
  final int numSplits = currentNumSplits.addAndGet(metadata.getBlocks().size());
  if (numSplits > maxSplits) {
    throw new TooManySplitsException(
      String.format("Too many splits encountered when processing parquet metadata at file %s, maximum is %d but encountered %d splits thus far.",
        file.getPath(), maxSplits, numSplits));
  }

  final MessageType schema = metadata.getFileMetaData().getSchema();

  Map<SchemaPath, OriginalType> originalTypeMap = Maps.newHashMap();
  schema.getPaths();
  for (String[] path : schema.getPaths()) {
    originalTypeMap.put(SchemaPath.getCompoundPath(path), getOriginalType(schema, path, 0));
  }

  List<RowGroupMetadata> rowGroupMetadataList = Lists.newArrayList();

  ArrayList<SchemaPath> ALL_COLS = new ArrayList<>();
  ALL_COLS.add(AbstractRecordReader.STAR_COLUMN);
  boolean autoCorrectCorruptDates = formatConfig.autoCorrectCorruptDates;
  ParquetReaderUtility.DateCorruptionStatus containsCorruptDates = ParquetReaderUtility.detectCorruptDates(metadata, ALL_COLS, autoCorrectCorruptDates);
  if(logger.isDebugEnabled()){
    logger.debug(containsCorruptDates.toString());
  }
  final Map<ColumnTypeMetadata.Key, ColumnTypeMetadata> columnTypeInfo = Maps.newHashMap();
  int rowGroupIdx = 0;
  for (BlockMetaData rowGroup : metadata.getBlocks()) {
    List<ColumnMetadata> columnMetadataList = Lists.newArrayList();
    long length = 0;
    for (ColumnChunkMetaData col : rowGroup.getColumns()) {
      ColumnMetadata columnMetadata;

      // statistics might just have the non-null counts with no min/max they might be
      // initialized to zero instead of null.
      // check statistics actually have non null values (or) column has all nulls.
      boolean statsAvailable = (col.getStatistics() != null && !col.getStatistics().isEmpty()
        && (col.getStatistics().hasNonNullValue()) || col.getStatistics().getNumNulls() ==
        rowGroup.getRowCount());

      Statistics<?> stats = col.getStatistics();
      String[] columnName = col.getPath().toArray();
      SchemaPath columnSchemaName = SchemaPath.getCompoundPath(columnName);
      ColumnTypeMetadata columnTypeMetadata =
          new ColumnTypeMetadata(columnName, col.getType(), originalTypeMap.get(columnSchemaName));

      columnTypeInfo.put(new ColumnTypeMetadata.Key(columnTypeMetadata.name), columnTypeMetadata);
      if (statsAvailable) {
        // Write stats only if minVal==maxVal. Also, we then store only maxVal
        Object mxValue = null;
        if (stats.genericGetMax() != null && stats.genericGetMin() != null &&
            stats.genericGetMax().equals(stats.genericGetMin())) {
          mxValue = stats.genericGetMax();
          if (containsCorruptDates == ParquetReaderUtility.DateCorruptionStatus.META_SHOWS_CORRUPTION
              && columnTypeMetadata.originalType == OriginalType.DATE) {
            mxValue = ParquetReaderUtility.autoCorrectCorruptedDate((Integer) mxValue);
          }
        }
        columnMetadata =
            new ColumnMetadata(columnTypeMetadata.name, mxValue, stats.getNumNulls());
      } else {
        // log it under trace to avoid lot of log entries.
        logger.trace("Stats are not available for column {}, rowGroupIdx {}, file {}",
            columnSchemaName, rowGroupIdx, file.getPath());
        columnMetadata = new ColumnMetadata(columnTypeMetadata.name,null, null);
      }
      columnMetadataList.add(columnMetadata);
      length += col.getTotalSize();
    }

    RowGroupMetadata rowGroupMeta =
        new RowGroupMetadata(rowGroup.getStartingPos(), length, rowGroup.getRowCount(),
            getHostAffinity(fs, file, rowGroup.getStartingPos(), length), columnMetadataList);

    rowGroupMetadataList.add(rowGroupMeta);
    rowGroupIdx++;
  }

  return new ParquetFileMetadata(file, file.size(), rowGroupMetadataList, columnTypeInfo);
}
 

@Test
public void test() throws Exception {
  Configuration conf = new Configuration();
  Path root = new Path("target/tests/TestParquetWriter/");
  FileSystem fs = root.getFileSystem(conf);
  if (fs.exists(root)) {
    fs.delete(root, true);
  }
  fs.mkdirs(root);
  MessageType schema = parseMessageType(
      "message test { "
      + "required binary binary_field; "
      + "required int32 int32_field; "
      + "required int64 int64_field; "
      + "required boolean boolean_field; "
      + "required float float_field; "
      + "required double double_field; "
      + "required fixed_len_byte_array(3) flba_field; "
      + "required int96 int96_field; "
      + "optional binary null_field; "
      + "} ");
  GroupWriteSupport.setSchema(schema, conf);
  SimpleGroupFactory f = new SimpleGroupFactory(schema);
  Map<String, Encoding> expected = new HashMap<String, Encoding>();
  expected.put("10-" + PARQUET_1_0, PLAIN_DICTIONARY);
  expected.put("1000-" + PARQUET_1_0, PLAIN);
  expected.put("10-" + PARQUET_2_0, RLE_DICTIONARY);
  expected.put("1000-" + PARQUET_2_0, DELTA_BYTE_ARRAY);
  for (int modulo : asList(10, 1000)) {
    for (WriterVersion version : WriterVersion.values()) {
      Path file = new Path(root, version.name() + "_" + modulo);
      ParquetWriter<Group> writer = new ParquetWriter<Group>(
          file,
          new GroupWriteSupport(),
          UNCOMPRESSED, 1024, 1024, 512, true, false, version, conf);
      for (int i = 0; i < 1000; i++) {
        writer.write(
            f.newGroup()
            .append("binary_field", "test" + (i % modulo))
            .append("int32_field", 32)
            .append("int64_field", 64l)
            .append("boolean_field", true)
            .append("float_field", 1.0f)
            .append("double_field", 2.0d)
            .append("flba_field", "foo")
            .append("int96_field", Binary.fromConstantByteArray(new byte[12])));
      }
      writer.close();

      ParquetReader<Group> reader = ParquetReader.builder(new GroupReadSupport(), file).withConf(conf).build();
      for (int i = 0; i < 1000; i++) {
        Group group = reader.read();
        assertEquals("test" + (i % modulo), group.getBinary("binary_field", 0).toStringUsingUTF8());
        assertEquals(32, group.getInteger("int32_field", 0));
        assertEquals(64l, group.getLong("int64_field", 0));
        assertEquals(true, group.getBoolean("boolean_field", 0));
        assertEquals(1.0f, group.getFloat("float_field", 0), 0.001);
        assertEquals(2.0d, group.getDouble("double_field", 0), 0.001);
        assertEquals("foo", group.getBinary("flba_field", 0).toStringUsingUTF8());
        assertEquals(Binary.fromConstantByteArray(new byte[12]), group.getInt96("int96_field",
            0));
        assertEquals(0, group.getFieldRepetitionCount("null_field"));
      }
      reader.close();
      ParquetMetadata footer = readFooter(conf, file, NO_FILTER);
      for (BlockMetaData blockMetaData : footer.getBlocks()) {
        for (ColumnChunkMetaData column : blockMetaData.getColumns()) {
          if (column.getPath().toDotString().equals("binary_field")) {
            String key = modulo + "-" + version;
            Encoding expectedEncoding = expected.get(key);
            assertTrue(
                key + ":" + column.getEncodings() + " should contain " + expectedEncoding,
                column.getEncodings().contains(expectedEncoding));
          }
        }
      }
    }
  }
}
 

private void addRowGroup(ParquetMetadata parquetMetadata, List<RowGroup> rowGroups, BlockMetaData block) {
    //rowGroup.total_byte_size = ;
    List<ColumnChunkMetaData> columns = block.getColumns();
    List<ColumnChunk> parquetColumns = new ArrayList<ColumnChunk>();
    for (ColumnChunkMetaData columnMetaData : columns) {
      ColumnChunk columnChunk = new ColumnChunk(columnMetaData.getFirstDataPageOffset()); // verify this is the right offset
      columnChunk.file_path = block.getPath(); // they are in the same file for now
      columnChunk.meta_data = new ColumnMetaData(
          getType(columnMetaData.getType()),
          toFormatEncodings(columnMetaData.getEncodings()),
          Arrays.asList(columnMetaData.getPath().toArray()),
          toFormatCodec(columnMetaData.getCodec()),
          columnMetaData.getValueCount(),
          columnMetaData.getTotalUncompressedSize(),
          columnMetaData.getTotalSize(),
          columnMetaData.getFirstDataPageOffset());
      if (columnMetaData.getEncodingStats() != null && columnMetaData.getEncodingStats().hasDictionaryPages()) {
        columnChunk.meta_data.setDictionary_page_offset(columnMetaData.getDictionaryPageOffset());
      }
      columnChunk.meta_data.setBloom_filter_offset(columnMetaData.getBloomFilterOffset());
      if (!columnMetaData.getStatistics().isEmpty()) {
        columnChunk.meta_data.setStatistics(toParquetStatistics(columnMetaData.getStatistics(), this.statisticsTruncateLength));
      }
      if (columnMetaData.getEncodingStats() != null) {
        columnChunk.meta_data.setEncoding_stats(convertEncodingStats(columnMetaData.getEncodingStats()));
      }
//      columnChunk.meta_data.index_page_offset = ;
//      columnChunk.meta_data.key_value_metadata = ; // nothing yet

      IndexReference columnIndexRef = columnMetaData.getColumnIndexReference();
      if (columnIndexRef != null) {
        columnChunk.setColumn_index_offset(columnIndexRef.getOffset());
        columnChunk.setColumn_index_length(columnIndexRef.getLength());
      }
      IndexReference offsetIndexRef = columnMetaData.getOffsetIndexReference();
      if (offsetIndexRef != null) {
        columnChunk.setOffset_index_offset(offsetIndexRef.getOffset());
        columnChunk.setOffset_index_length(offsetIndexRef.getLength());
      }

      parquetColumns.add(columnChunk);
    }
    RowGroup rowGroup = new RowGroup(parquetColumns, block.getTotalByteSize(), block.getRowCount());
    rowGroups.add(rowGroup);
  }