Examples with FileMetaData parquet.hadoop.metadata.FileMetaData used on opensource projects

Example 1 with FileMetaData

use of parquet.hadoop.metadata.FileMetaData in project presto by prestodb.

the class ParquetTester method assertFileContents.

private static void assertFileContents(JobConf jobConf, TempFile tempFile, Iterable<?> expectedValues, Type type) throws IOException, InterruptedException {
    Path path = new Path(tempFile.getFile().toURI());
    FileSystem fileSystem = path.getFileSystem(jobConf);
    ParquetMetadata parquetMetadata = ParquetMetadataReader.readFooter(fileSystem, path);
    FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
    MessageType fileSchema = fileMetaData.getSchema();
    long size = fileSystem.getFileStatus(path).getLen();
    FSDataInputStream inputStream = fileSystem.open(path);
    ParquetDataSource dataSource = new HdfsParquetDataSource(path, size, inputStream);
    TypeManager typeManager = new TypeRegistry();
    ParquetReader parquetReader = new ParquetReader(fileSchema, fileSchema, parquetMetadata.getBlocks(), dataSource, typeManager, new AggregatedMemoryContext());
    assertEquals(parquetReader.getPosition(), 0);
    int rowsProcessed = 0;
    Iterator<?> iterator = expectedValues.iterator();
    for (int batchSize = parquetReader.nextBatch(); batchSize >= 0; batchSize = parquetReader.nextBatch()) {
        ColumnDescriptor columnDescriptor = fileSchema.getColumns().get(0);
        Block block = parquetReader.readPrimitive(columnDescriptor, type);
        for (int i = 0; i < batchSize; i++) {
            assertTrue(iterator.hasNext());
            Object expected = iterator.next();
            Object actual = decodeObject(type, block, i);
            assertEquals(actual, expected);
        }
        rowsProcessed += batchSize;
        assertEquals(parquetReader.getPosition(), rowsProcessed);
    }
    assertFalse(iterator.hasNext());
    assertEquals(parquetReader.getPosition(), rowsProcessed);
    parquetReader.close();
}

Example 2 with FileMetaData

use of parquet.hadoop.metadata.FileMetaData in project presto by prestodb.

the class ParquetHiveRecordCursor method createParquetRecordReader.

private ParquetRecordReader<FakeParquetRecord> createParquetRecordReader(HdfsEnvironment hdfsEnvironment, String sessionUser, Configuration configuration, Path path, long start, long length, List<HiveColumnHandle> columns, boolean useParquetColumnNames, TypeManager typeManager, boolean predicatePushdownEnabled, TupleDomain<HiveColumnHandle> effectivePredicate) {
    ParquetDataSource dataSource = null;
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(sessionUser, path, configuration);
        dataSource = buildHdfsParquetDataSource(fileSystem, path, start, length);
        ParquetMetadata parquetMetadata = hdfsEnvironment.doAs(sessionUser, () -> ParquetFileReader.readFooter(configuration, path, NO_FILTER));
        List<BlockMetaData> blocks = parquetMetadata.getBlocks();
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        PrestoReadSupport readSupport = new PrestoReadSupport(useParquetColumnNames, columns, fileSchema);
        List<parquet.schema.Type> fields = columns.stream().filter(column -> column.getColumnType() == REGULAR).map(column -> getParquetType(column, fileSchema, useParquetColumnNames)).filter(Objects::nonNull).collect(toList());
        MessageType requestedSchema = new MessageType(fileSchema.getName(), fields);
        LongArrayList offsets = new LongArrayList(blocks.size());
        for (BlockMetaData block : blocks) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (firstDataPage >= start && firstDataPage < start + length) {
                if (predicatePushdownEnabled) {
                    ParquetPredicate parquetPredicate = buildParquetPredicate(columns, effectivePredicate, fileMetaData.getSchema(), typeManager);
                    if (predicateMatches(parquetPredicate, block, dataSource, requestedSchema, effectivePredicate)) {
                        offsets.add(block.getStartingPos());
                    }
                } else {
                    offsets.add(block.getStartingPos());
                }
            }
        }
        ParquetInputSplit split = new ParquetInputSplit(path, start, start + length, length, null, offsets.toLongArray());
        TaskAttemptContext taskContext = ContextUtil.newTaskAttemptContext(configuration, new TaskAttemptID());
        return hdfsEnvironment.doAs(sessionUser, () -> {
            ParquetRecordReader<FakeParquetRecord> realReader = new PrestoParquetRecordReader(readSupport);
            realReader.initialize(split, taskContext);
            return realReader;
        });
    } catch (Exception e) {
        Throwables.propagateIfInstanceOf(e, PrestoException.class);
        if (e instanceof InterruptedException) {
            Thread.currentThread().interrupt();
            throw Throwables.propagate(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);
    } finally {
        if (dataSource != null) {
            try {
                dataSource.close();
            } catch (IOException ignored) {
            }
        }
    }
}

Example 3 with FileMetaData

use of parquet.hadoop.metadata.FileMetaData in project presto by prestodb.

the class ParquetPageSourceFactory method createParquetPageSource.

public static ParquetPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, Properties schema, List<HiveColumnHandle> columns, boolean useParquetColumnNames, TypeManager typeManager, boolean predicatePushdownEnabled, TupleDomain<HiveColumnHandle> effectivePredicate) {
    AggregatedMemoryContext systemMemoryContext = new AggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(user, path, configuration);
        dataSource = buildHdfsParquetDataSource(fileSystem, path, start, length);
        ParquetMetadata parquetMetadata = ParquetMetadataReader.readFooter(fileSystem, path);
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        List<parquet.schema.Type> fields = columns.stream().filter(column -> column.getColumnType() == REGULAR).map(column -> getParquetType(column, fileSchema, useParquetColumnNames)).filter(Objects::nonNull).collect(toList());
        MessageType requestedSchema = new MessageType(fileSchema.getName(), fields);
        List<BlockMetaData> blocks = new ArrayList<>();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (firstDataPage >= start && firstDataPage < start + length) {
                blocks.add(block);
            }
        }
        if (predicatePushdownEnabled) {
            ParquetPredicate parquetPredicate = buildParquetPredicate(columns, effectivePredicate, fileMetaData.getSchema(), typeManager);
            final ParquetDataSource finalDataSource = dataSource;
            blocks = blocks.stream().filter(block -> predicateMatches(parquetPredicate, block, finalDataSource, requestedSchema, effectivePredicate)).collect(toList());
        }
        ParquetReader parquetReader = new ParquetReader(fileSchema, requestedSchema, blocks, dataSource, typeManager, systemMemoryContext);
        return new ParquetPageSource(parquetReader, dataSource, fileSchema, requestedSchema, length, schema, columns, effectivePredicate, typeManager, useParquetColumnNames, systemMemoryContext);
    } catch (Exception e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) 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);
    }
}