Examples with ParquetDataSource com.facebook.presto.parquet.ParquetDataSource used on opensource projects

Example 1 with ParquetDataSource

use of com.facebook.presto.parquet.ParquetDataSource in project presto by prestodb.

the class DeltaPageSourceProvider method createParquetPageSource.

private static ConnectorPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, long fileSize, List<DeltaColumnHandle> columns, SchemaTableName tableName, DataSize maxReadBlockSize, boolean batchReaderEnabled, boolean verificationEnabled, TypeManager typeManager, TupleDomain<DeltaColumnHandle> effectivePredicate, FileFormatDataSourceStats stats, boolean columnIndexFilterEnabled) {
    AggregatedMemoryContext systemMemoryContext = newSimpleAggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        FSDataInputStream inputStream = hdfsEnvironment.getFileSystem(user, path, configuration).open(path);
        dataSource = buildHdfsParquetDataSource(inputStream, path, stats);
        ParquetMetadata parquetMetadata = MetadataReader.readFooter(dataSource, fileSize).getParquetMetadata();
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        Optional<MessageType> message = columns.stream().filter(column -> column.getColumnType() == REGULAR || isPushedDownSubfield(column)).map(column -> getColumnType(typeManager.getType(column.getDataType()), fileSchema, column, tableName, path)).filter(Optional::isPresent).map(Optional::get).map(type -> new MessageType(fileSchema.getName(), type)).reduce(MessageType::union);
        MessageType requestedSchema = message.orElse(new MessageType(fileSchema.getName(), ImmutableList.of()));
        ImmutableList.Builder<BlockMetaData> footerBlocks = ImmutableList.builder();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (firstDataPage >= start && firstDataPage < start + length) {
                footerBlocks.add(block);
            }
        }
        Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, requestedSchema);
        TupleDomain<ColumnDescriptor> parquetTupleDomain = getParquetTupleDomain(descriptorsByPath, effectivePredicate);
        Predicate parquetPredicate = buildPredicate(requestedSchema, parquetTupleDomain, descriptorsByPath);
        final ParquetDataSource finalDataSource = dataSource;
        ImmutableList.Builder<BlockMetaData> blocks = ImmutableList.builder();
        List<ColumnIndexStore> blockIndexStores = new ArrayList<>();
        for (BlockMetaData block : footerBlocks.build()) {
            Optional<ColumnIndexStore> columnIndexStore = ColumnIndexFilterUtils.getColumnIndexStore(parquetPredicate, finalDataSource, block, descriptorsByPath, columnIndexFilterEnabled);
            if (predicateMatches(parquetPredicate, block, finalDataSource, descriptorsByPath, parquetTupleDomain, columnIndexStore, columnIndexFilterEnabled)) {
                blocks.add(block);
                blockIndexStores.add(columnIndexStore.orElse(null));
            }
        }
        MessageColumnIO messageColumnIO = getColumnIO(fileSchema, requestedSchema);
        ParquetReader parquetReader = new ParquetReader(messageColumnIO, blocks.build(), dataSource, systemMemoryContext, maxReadBlockSize, batchReaderEnabled, verificationEnabled, parquetPredicate, blockIndexStores, columnIndexFilterEnabled);
        ImmutableList.Builder<String> namesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Type> typesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Optional<Field>> fieldsBuilder = ImmutableList.builder();
        for (DeltaColumnHandle column : columns) {
            checkArgument(column.getColumnType() == REGULAR || column.getColumnType() == SUBFIELD, "column type must be regular or subfield column");
            String name = column.getName();
            Type type = typeManager.getType(column.getDataType());
            namesBuilder.add(name);
            typesBuilder.add(type);
            if (isPushedDownSubfield(column)) {
                Subfield pushedDownSubfield = getPushedDownSubfield(column);
                List<String> nestedColumnPath = nestedColumnPath(pushedDownSubfield);
                Optional<ColumnIO> columnIO = findNestedColumnIO(lookupColumnByName(messageColumnIO, pushedDownSubfield.getRootName()), nestedColumnPath);
                if (columnIO.isPresent()) {
                    fieldsBuilder.add(constructField(type, columnIO.get()));
                } else {
                    fieldsBuilder.add(Optional.empty());
                }
            } else if (getParquetType(type, fileSchema, column, tableName, path).isPresent()) {
                fieldsBuilder.add(constructField(type, lookupColumnByName(messageColumnIO, name)));
            } else {
                fieldsBuilder.add(Optional.empty());
            }
        }
        return new ParquetPageSource(parquetReader, typesBuilder.build(), fieldsBuilder.build(), namesBuilder.build(), new RuntimeStats());
    } catch (Exception exception) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (exception instanceof PrestoException) {
            throw (PrestoException) exception;
        }
        if (exception instanceof ParquetCorruptionException) {
            throw new PrestoException(DELTA_BAD_DATA, exception);
        }
        if (exception instanceof AccessControlException) {
            throw new PrestoException(PERMISSION_DENIED, exception.getMessage(), exception);
        }
        if (nullToEmpty(exception.getMessage()).trim().equals("Filesystem closed") || exception instanceof FileNotFoundException) {
            throw new PrestoException(DELTA_CANNOT_OPEN_SPLIT, exception);
        }
        String message = format("Error opening Hive split %s (offset=%s, length=%s): %s", path, start, length, exception.getMessage());
        if (exception.getClass().getSimpleName().equals("BlockMissingException")) {
            throw new PrestoException(DELTA_MISSING_DATA, message, exception);
        }
        throw new PrestoException(DELTA_CANNOT_OPEN_SPLIT, message, exception);
    }
}

Example 2 with ParquetDataSource

use of com.facebook.presto.parquet.ParquetDataSource in project presto by prestodb.

the class IcebergPageSourceProvider method createParquetPageSource.

private static ConnectorPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, SchemaTableName tableName, List<IcebergColumnHandle> regularColumns, boolean useParquetColumnNames, DataSize maxReadBlockSize, boolean batchReaderEnabled, boolean verificationEnabled, TupleDomain<IcebergColumnHandle> effectivePredicate, FileFormatDataSourceStats fileFormatDataSourceStats, boolean columnIndexFilterEnabled) {
    AggregatedMemoryContext systemMemoryContext = newSimpleAggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        ExtendedFileSystem fileSystem = hdfsEnvironment.getFileSystem(user, path, configuration);
        FileStatus fileStatus = fileSystem.getFileStatus(path);
        long fileSize = fileStatus.getLen();
        long modificationTime = fileStatus.getModificationTime();
        HiveFileContext hiveFileContext = new HiveFileContext(true, NO_CACHE_CONSTRAINTS, Optional.empty(), Optional.of(fileSize), modificationTime, false);
        FSDataInputStream inputStream = fileSystem.openFile(path, hiveFileContext);
        dataSource = buildHdfsParquetDataSource(inputStream, path, fileFormatDataSourceStats);
        ParquetMetadata parquetMetadata = MetadataReader.readFooter(dataSource, fileSize).getParquetMetadata();
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        // Mapping from Iceberg field ID to Parquet fields.
        Map<Integer, org.apache.parquet.schema.Type> parquetIdToField = fileSchema.getFields().stream().filter(field -> field.getId() != null).collect(toImmutableMap(field -> field.getId().intValue(), Function.identity()));
        List<org.apache.parquet.schema.Type> parquetFields = regularColumns.stream().map(column -> {
            if (parquetIdToField.isEmpty()) {
                // This is a migrated table
                return getParquetTypeByName(column.getName(), fileSchema);
            }
            return parquetIdToField.get(column.getId());
        }).collect(toList());
        // TODO: support subfield pushdown
        MessageType requestedSchema = new MessageType(fileSchema.getName(), parquetFields.stream().filter(Objects::nonNull).collect(toImmutableList()));
        Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, requestedSchema);
        TupleDomain<ColumnDescriptor> parquetTupleDomain = getParquetTupleDomain(descriptorsByPath, effectivePredicate);
        Predicate parquetPredicate = buildPredicate(requestedSchema, parquetTupleDomain, descriptorsByPath);
        final ParquetDataSource finalDataSource = dataSource;
        List<BlockMetaData> blocks = new ArrayList<>();
        List<ColumnIndexStore> blockIndexStores = new ArrayList<>();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            Optional<ColumnIndexStore> columnIndexStore = ColumnIndexFilterUtils.getColumnIndexStore(parquetPredicate, finalDataSource, block, descriptorsByPath, columnIndexFilterEnabled);
            if ((firstDataPage >= start) && (firstDataPage < (start + length)) && predicateMatches(parquetPredicate, block, dataSource, descriptorsByPath, parquetTupleDomain, columnIndexStore, columnIndexFilterEnabled)) {
                blocks.add(block);
                blockIndexStores.add(columnIndexStore.orElse(null));
            }
        }
        MessageColumnIO messageColumnIO = getColumnIO(fileSchema, requestedSchema);
        ParquetReader parquetReader = new ParquetReader(messageColumnIO, blocks, dataSource, systemMemoryContext, maxReadBlockSize, batchReaderEnabled, verificationEnabled, parquetPredicate, blockIndexStores, columnIndexFilterEnabled);
        ImmutableList.Builder<String> namesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Type> prestoTypes = ImmutableList.builder();
        ImmutableList.Builder<Optional<Field>> internalFields = ImmutableList.builder();
        for (int columnIndex = 0; columnIndex < regularColumns.size(); columnIndex++) {
            IcebergColumnHandle column = regularColumns.get(columnIndex);
            namesBuilder.add(column.getName());
            org.apache.parquet.schema.Type parquetField = parquetFields.get(columnIndex);
            Type prestoType = column.getType();
            prestoTypes.add(prestoType);
            if (parquetField == null) {
                internalFields.add(Optional.empty());
            } else {
                internalFields.add(constructField(column.getType(), messageColumnIO.getChild(parquetField.getName())));
            }
        }
        return new ParquetPageSource(parquetReader, prestoTypes.build(), internalFields.build(), namesBuilder.build(), new RuntimeStats());
    } catch (Exception e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) e;
        }
        String message = format("Error opening Iceberg split %s (offset=%s, length=%s): %s", path, start, length, e.getMessage());
        if (e instanceof ParquetCorruptionException) {
            throw new PrestoException(ICEBERG_BAD_DATA, message, e);
        }
        if (e instanceof BlockMissingException) {
            throw new PrestoException(ICEBERG_MISSING_DATA, message, e);
        }
        throw new PrestoException(ICEBERG_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 3 with ParquetDataSource

use of com.facebook.presto.parquet.ParquetDataSource in project presto by prestodb.

the class ParquetPageSourceFactory method createParquetPageSource.

public static ConnectorPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, long fileSize, List<HiveColumnHandle> columns, SchemaTableName tableName, boolean useParquetColumnNames, DataSize maxReadBlockSize, boolean batchReaderEnabled, boolean verificationEnabled, TypeManager typeManager, StandardFunctionResolution functionResolution, TupleDomain<HiveColumnHandle> effectivePredicate, FileFormatDataSourceStats stats, HiveFileContext hiveFileContext, ParquetMetadataSource parquetMetadataSource, boolean columnIndexFilterEnabled) {
    AggregatedMemoryContext systemMemoryContext = newSimpleAggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        FSDataInputStream inputStream = hdfsEnvironment.getFileSystem(user, path, configuration).openFile(path, hiveFileContext);
        dataSource = buildHdfsParquetDataSource(inputStream, path, stats);
        ParquetMetadata parquetMetadata = parquetMetadataSource.getParquetMetadata(dataSource, fileSize, hiveFileContext.isCacheable()).getParquetMetadata();
        if (!columns.isEmpty() && columns.stream().allMatch(hiveColumnHandle -> hiveColumnHandle.getColumnType() == AGGREGATED)) {
            return new AggregatedParquetPageSource(columns, parquetMetadata, typeManager, functionResolution);
        }
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        Optional<MessageType> message = columns.stream().filter(column -> column.getColumnType() == REGULAR || isPushedDownSubfield(column)).map(column -> getColumnType(typeManager.getType(column.getTypeSignature()), fileSchema, useParquetColumnNames, column, tableName, path)).filter(Optional::isPresent).map(Optional::get).map(type -> new MessageType(fileSchema.getName(), type)).reduce(MessageType::union);
        MessageType requestedSchema = message.orElse(new MessageType(fileSchema.getName(), ImmutableList.of()));
        ImmutableList.Builder<BlockMetaData> footerBlocks = ImmutableList.builder();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (firstDataPage >= start && firstDataPage < start + length) {
                footerBlocks.add(block);
            }
        }
        Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, requestedSchema);
        TupleDomain<ColumnDescriptor> parquetTupleDomain = getParquetTupleDomain(descriptorsByPath, effectivePredicate);
        Predicate parquetPredicate = buildPredicate(requestedSchema, parquetTupleDomain, descriptorsByPath);
        final ParquetDataSource finalDataSource = dataSource;
        ImmutableList.Builder<BlockMetaData> blocks = ImmutableList.builder();
        List<ColumnIndexStore> blockIndexStores = new ArrayList<>();
        for (BlockMetaData block : footerBlocks.build()) {
            Optional<ColumnIndexStore> columnIndexStore = ColumnIndexFilterUtils.getColumnIndexStore(parquetPredicate, finalDataSource, block, descriptorsByPath, columnIndexFilterEnabled);
            if (predicateMatches(parquetPredicate, block, finalDataSource, descriptorsByPath, parquetTupleDomain, columnIndexStore, columnIndexFilterEnabled)) {
                blocks.add(block);
                blockIndexStores.add(columnIndexStore.orElse(null));
                hiveFileContext.incrementCounter("parquet.blocksRead", 1);
                hiveFileContext.incrementCounter("parquet.rowsRead", block.getRowCount());
                hiveFileContext.incrementCounter("parquet.totalBytesRead", block.getTotalByteSize());
            } else {
                hiveFileContext.incrementCounter("parquet.blocksSkipped", 1);
                hiveFileContext.incrementCounter("parquet.rowsSkipped", block.getRowCount());
                hiveFileContext.incrementCounter("parquet.totalBytesSkipped", block.getTotalByteSize());
            }
        }
        MessageColumnIO messageColumnIO = getColumnIO(fileSchema, requestedSchema);
        ParquetReader parquetReader = new ParquetReader(messageColumnIO, blocks.build(), dataSource, systemMemoryContext, maxReadBlockSize, batchReaderEnabled, verificationEnabled, parquetPredicate, blockIndexStores, columnIndexFilterEnabled);
        ImmutableList.Builder<String> namesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Type> typesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Optional<Field>> fieldsBuilder = ImmutableList.builder();
        for (HiveColumnHandle column : columns) {
            checkArgument(column.getColumnType() == REGULAR || column.getColumnType() == SYNTHESIZED, "column type must be regular or synthesized column");
            String name = column.getName();
            Type type = typeManager.getType(column.getTypeSignature());
            namesBuilder.add(name);
            typesBuilder.add(type);
            if (column.getColumnType() == SYNTHESIZED) {
                Subfield pushedDownSubfield = getPushedDownSubfield(column);
                List<String> nestedColumnPath = nestedColumnPath(pushedDownSubfield);
                Optional<ColumnIO> columnIO = findNestedColumnIO(lookupColumnByName(messageColumnIO, pushedDownSubfield.getRootName()), nestedColumnPath);
                if (columnIO.isPresent()) {
                    fieldsBuilder.add(constructField(type, columnIO.get()));
                } else {
                    fieldsBuilder.add(Optional.empty());
                }
            } else if (getParquetType(type, fileSchema, useParquetColumnNames, column, tableName, path).isPresent()) {
                String columnName = useParquetColumnNames ? name : fileSchema.getFields().get(column.getHiveColumnIndex()).getName();
                fieldsBuilder.add(constructField(type, lookupColumnByName(messageColumnIO, columnName)));
            } else {
                fieldsBuilder.add(Optional.empty());
            }
        }
        return new ParquetPageSource(parquetReader, typesBuilder.build(), fieldsBuilder.build(), namesBuilder.build(), hiveFileContext.getStats());
    } catch (Exception e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) e;
        }
        if (e instanceof ParquetCorruptionException) {
            throw new PrestoException(HIVE_BAD_DATA, e);
        }
        if (e instanceof AccessControlException) {
            throw new PrestoException(PERMISSION_DENIED, e.getMessage(), e);
        }
        if (nullToEmpty(e.getMessage()).trim().equals("Filesystem closed") || e instanceof FileNotFoundException) {
            throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, e);
        }
        String message = format("Error opening Hive split %s (offset=%s, length=%s): %s", path, start, length, e.getMessage());
        if (e.getClass().getSimpleName().equals("BlockMissingException")) {
            throw new PrestoException(HIVE_MISSING_DATA, message, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 4 with ParquetDataSource

use of com.facebook.presto.parquet.ParquetDataSource in project presto by prestodb.

the class MetadataReader method readFooter.

public static ParquetFileMetadata readFooter(ParquetDataSource parquetDataSource, long fileSize) throws IOException {
    // Parquet File Layout:
    // 
    // MAGIC
    // variable: Data
    // variable: Metadata
    // 4 bytes: MetadataLength
    // MAGIC
    validateParquet(fileSize >= MAGIC.length() + POST_SCRIPT_SIZE, "%s is not a valid Parquet File", parquetDataSource.getId());
    // EXPECTED_FOOTER_SIZE is an int, so this will never fail
    byte[] buffer = new byte[toIntExact(min(fileSize, EXPECTED_FOOTER_SIZE))];
    parquetDataSource.readFully(fileSize - buffer.length, buffer);
    Slice tailSlice = wrappedBuffer(buffer);
    Slice magic = tailSlice.slice(tailSlice.length() - MAGIC.length(), MAGIC.length());
    if (!MAGIC.equals(magic)) {
        throw new ParquetCorruptionException(format("Not valid Parquet file: %s expected magic number: %s got: %s", parquetDataSource.getId(), Arrays.toString(MAGIC.getBytes()), Arrays.toString(magic.getBytes())));
    }
    int metadataLength = tailSlice.getInt(tailSlice.length() - POST_SCRIPT_SIZE);
    int completeFooterSize = metadataLength + POST_SCRIPT_SIZE;
    long metadataFileOffset = fileSize - completeFooterSize;
    validateParquet(metadataFileOffset >= MAGIC.length() && metadataFileOffset + POST_SCRIPT_SIZE < fileSize, "Corrupted Parquet file: %s metadata index: %s out of range", parquetDataSource.getId(), metadataFileOffset);
    // Ensure the slice covers the entire metadata range
    if (tailSlice.length() < completeFooterSize) {
        byte[] footerBuffer = new byte[completeFooterSize];
        parquetDataSource.readFully(metadataFileOffset, footerBuffer, 0, footerBuffer.length - tailSlice.length());
        // Copy the previous slice contents into the new buffer
        tailSlice.getBytes(0, footerBuffer, footerBuffer.length - tailSlice.length(), tailSlice.length());
        tailSlice = wrappedBuffer(footerBuffer, 0, footerBuffer.length);
    }
    FileMetaData fileMetaData = readFileMetaData(tailSlice.slice(tailSlice.length() - completeFooterSize, metadataLength).getInput());
    List<SchemaElement> schema = fileMetaData.getSchema();
    validateParquet(!schema.isEmpty(), "Empty Parquet schema in file: %s", parquetDataSource.getId());
    MessageType messageType = readParquetSchema(schema);
    List<BlockMetaData> blocks = new ArrayList<>();
    List<RowGroup> rowGroups = fileMetaData.getRow_groups();
    if (rowGroups != null) {
        for (RowGroup rowGroup : rowGroups) {
            BlockMetaData blockMetaData = new BlockMetaData();
            blockMetaData.setRowCount(rowGroup.getNum_rows());
            blockMetaData.setTotalByteSize(rowGroup.getTotal_byte_size());
            List<ColumnChunk> columns = rowGroup.getColumns();
            validateParquet(!columns.isEmpty(), "No columns in row group: %s", rowGroup);
            String filePath = columns.get(0).getFile_path();
            for (ColumnChunk columnChunk : columns) {
                validateParquet((filePath == null && columnChunk.getFile_path() == null) || (filePath != null && filePath.equals(columnChunk.getFile_path())), "all column chunks of the same row group must be in the same file");
                ColumnMetaData metaData = columnChunk.meta_data;
                String[] path = metaData.path_in_schema.stream().map(value -> value.toLowerCase(Locale.ENGLISH)).toArray(String[]::new);
                ColumnPath columnPath = ColumnPath.get(path);
                PrimitiveType primitiveType = messageType.getType(columnPath.toArray()).asPrimitiveType();
                PrimitiveTypeName primitiveTypeName = primitiveType.getPrimitiveTypeName();
                ColumnChunkMetaData column = ColumnChunkMetaData.get(columnPath, primitiveType, CompressionCodecName.fromParquet(metaData.codec), PARQUET_METADATA_CONVERTER.convertEncodingStats(metaData.encoding_stats), readEncodings(metaData.encodings), readStats(metaData.statistics, primitiveTypeName), metaData.data_page_offset, metaData.dictionary_page_offset, metaData.num_values, metaData.total_compressed_size, metaData.total_uncompressed_size);
                column.setColumnIndexReference(toColumnIndexReference(columnChunk));
                column.setOffsetIndexReference(toOffsetIndexReference(columnChunk));
                blockMetaData.addColumn(column);
            }
            blockMetaData.setPath(filePath);
            blocks.add(blockMetaData);
        }
    }
    Map<String, String> keyValueMetaData = new HashMap<>();
    List<KeyValue> keyValueList = fileMetaData.getKey_value_metadata();
    if (keyValueList != null) {
        for (KeyValue keyValue : keyValueList) {
            keyValueMetaData.put(keyValue.key, keyValue.value);
        }
    }
    ParquetMetadata parquetMetadata = new ParquetMetadata(new org.apache.parquet.hadoop.metadata.FileMetaData(messageType, keyValueMetaData, fileMetaData.getCreated_by()), blocks);
    return new ParquetFileMetadata(parquetMetadata, toIntExact(metadataLength));
}