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

Example 1 with RichColumnDescriptor

use of com.facebook.presto.parquet.RichColumnDescriptor 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 RichColumnDescriptor

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

the class TestParquetPredicateUtils method testParquetTupleDomainPrimitive.

@Test
public void testParquetTupleDomainPrimitive() {
    HiveColumnHandle columnHandle = new HiveColumnHandle("my_primitive", HiveType.valueOf("bigint"), parseTypeSignature(StandardTypes.BIGINT), 0, REGULAR, Optional.empty(), Optional.empty());
    Domain singleValueDomain = Domain.singleValue(BIGINT, 123L);
    TupleDomain<HiveColumnHandle> domain = withColumnDomains(ImmutableMap.of(columnHandle, singleValueDomain));
    MessageType fileSchema = new MessageType("hive_schema", new PrimitiveType(OPTIONAL, INT64, "my_primitive"));
    Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, fileSchema);
    TupleDomain<ColumnDescriptor> tupleDomain = getParquetTupleDomain(descriptorsByPath, domain);
    assertEquals(tupleDomain.getDomains().get().size(), 1);
    ColumnDescriptor descriptor = tupleDomain.getDomains().get().keySet().iterator().next();
    assertEquals(descriptor.getPath().length, 1);
    assertEquals(descriptor.getPath()[0], "my_primitive");
    Domain predicateDomain = Iterables.getOnlyElement(tupleDomain.getDomains().get().values());
    assertEquals(predicateDomain, singleValueDomain);
}

Example 3 with RichColumnDescriptor

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

the class TestParquetPredicateUtils method testParquetTupleDomainStructArray.

@Test
public void testParquetTupleDomainStructArray() {
    HiveColumnHandle columnHandle = new HiveColumnHandle("my_array_struct", HiveType.valueOf("array<struct<a:int>>"), parseTypeSignature(StandardTypes.ARRAY), 0, REGULAR, Optional.empty(), Optional.empty());
    RowType.Field rowField = new RowType.Field(Optional.of("a"), INTEGER);
    RowType rowType = RowType.from(ImmutableList.of(rowField));
    TupleDomain<HiveColumnHandle> domain = withColumnDomains(ImmutableMap.of(columnHandle, Domain.notNull(new ArrayType(rowType))));
    MessageType fileSchema = new MessageType("hive_schema", new GroupType(OPTIONAL, "my_array_struct", new GroupType(REPEATED, "bag", new GroupType(OPTIONAL, "array_element", new PrimitiveType(OPTIONAL, INT32, "a")))));
    Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, fileSchema);
    TupleDomain<ColumnDescriptor> tupleDomain = getParquetTupleDomain(descriptorsByPath, domain);
    assertTrue(tupleDomain.getDomains().get().isEmpty());
}

Example 4 with RichColumnDescriptor

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

the class TupleDomainParquetPredicate method getParquetUserDefinedPredicate.

public FilterPredicate getParquetUserDefinedPredicate() {
    FilterPredicate filter = null;
    // we assume the relation cross domains are 'or'
    for (RichColumnDescriptor column : columns) {
        Domain domain = effectivePredicate.getDomains().get().get(column);
        if (domain == null || domain.isNone()) {
            continue;
        }
        if (domain.isAll()) {
            continue;
        }
        FilterPredicate columnFilter = FilterApi.userDefined(FilterApi.intColumn(ColumnPath.get(column.getPath()).toDotString()), new ParquetUserDefinedPredicateTupleDomain(domain));
        if (filter == null) {
            filter = columnFilter;
        } else {
            filter = FilterApi.or(filter, columnFilter);
        }
    }
    return filter;
}

Example 5 with RichColumnDescriptor

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

the class TupleDomainParquetPredicate method matches.

@Override
public boolean matches(long numberOfRows, Map<ColumnDescriptor, Statistics<?>> statistics, ParquetDataSourceId id) throws ParquetCorruptionException {
    if (numberOfRows == 0) {
        return false;
    }
    if (effectivePredicate.isNone()) {
        return false;
    }
    Map<ColumnDescriptor, Domain> effectivePredicateDomains = effectivePredicate.getDomains().orElseThrow(() -> new IllegalStateException("Effective predicate other than none should have domains"));
    for (RichColumnDescriptor column : columns) {
        Domain effectivePredicateDomain = effectivePredicateDomains.get(column);
        if (effectivePredicateDomain == null) {
            continue;
        }
        Statistics<?> columnStatistics = statistics.get(column);
        if (columnStatistics == null || columnStatistics.isEmpty()) {
        // no stats for column
        } else {
            Domain domain = getDomain(column, effectivePredicateDomain.getType(), numberOfRows, columnStatistics, id);
            if (effectivePredicateDomain.intersect(domain).isNone()) {
                return false;
            }
        }
    }
    return true;
}