Examples with ReaderPageSource io.trino.plugin.hive.ReaderPageSource used on opensource projects

Example 6 with ReaderPageSource

use of io.trino.plugin.hive.ReaderPageSource in project trino by trinodb.

the class IcebergPageSourceProvider method createParquetPageSource.

private static ReaderPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, ConnectorIdentity identity, Configuration configuration, Path path, long start, long length, long fileSize, List<IcebergColumnHandle> regularColumns, ParquetReaderOptions options, TupleDomain<IcebergColumnHandle> effectivePredicate, FileFormatDataSourceStats fileFormatDataSourceStats, Optional<NameMapping> nameMapping) {
    AggregatedMemoryContext memoryContext = newSimpleAggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(identity, path, configuration);
        FSDataInputStream inputStream = hdfsEnvironment.doAs(identity, () -> fileSystem.open(path));
        dataSource = new HdfsParquetDataSource(new ParquetDataSourceId(path.toString()), fileSize, inputStream, fileFormatDataSourceStats, options);
        // extra variable required for lambda below
        ParquetDataSource theDataSource = dataSource;
        ParquetMetadata parquetMetadata = hdfsEnvironment.doAs(identity, () -> MetadataReader.readFooter(theDataSource));
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        if (nameMapping.isPresent() && !ParquetSchemaUtil.hasIds(fileSchema)) {
            // NameMapping conversion is necessary because MetadataReader converts all column names to lowercase and NameMapping is case sensitive
            fileSchema = ParquetSchemaUtil.applyNameMapping(fileSchema, convertToLowercase(nameMapping.get()));
        }
        // 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()));
        Optional<ReaderColumns> columnProjections = projectColumns(regularColumns);
        List<IcebergColumnHandle> readColumns = columnProjections.map(readerColumns -> (List<IcebergColumnHandle>) readerColumns.get().stream().map(IcebergColumnHandle.class::cast).collect(toImmutableList())).orElse(regularColumns);
        List<org.apache.parquet.schema.Type> parquetFields = readColumns.stream().map(column -> parquetIdToField.get(column.getId())).collect(toList());
        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, UTC);
        List<BlockMetaData> blocks = new ArrayList<>();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (start <= firstDataPage && firstDataPage < start + length && predicateMatches(parquetPredicate, block, dataSource, descriptorsByPath, parquetTupleDomain)) {
                blocks.add(block);
            }
        }
        MessageColumnIO messageColumnIO = getColumnIO(fileSchema, requestedSchema);
        ParquetReader parquetReader = new ParquetReader(Optional.ofNullable(fileMetaData.getCreatedBy()), messageColumnIO, blocks, Optional.empty(), dataSource, UTC, memoryContext, options);
        ImmutableList.Builder<Type> trinoTypes = ImmutableList.builder();
        ImmutableList.Builder<Optional<Field>> internalFields = ImmutableList.builder();
        for (int columnIndex = 0; columnIndex < readColumns.size(); columnIndex++) {
            IcebergColumnHandle column = readColumns.get(columnIndex);
            org.apache.parquet.schema.Type parquetField = parquetFields.get(columnIndex);
            Type trinoType = column.getBaseType();
            trinoTypes.add(trinoType);
            if (parquetField == null) {
                internalFields.add(Optional.empty());
            } else {
                // The top level columns are already mapped by name/id appropriately.
                ColumnIO columnIO = messageColumnIO.getChild(parquetField.getName());
                internalFields.add(IcebergParquetColumnIOConverter.constructField(new FieldContext(trinoType, column.getColumnIdentity()), columnIO));
            }
        }
        return new ReaderPageSource(new ParquetPageSource(parquetReader, trinoTypes.build(), internalFields.build()), columnProjections);
    } catch (IOException | RuntimeException e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof TrinoException) {
            throw (TrinoException) 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 TrinoException(ICEBERG_BAD_DATA, message, e);
        }
        if (e instanceof BlockMissingException) {
            throw new TrinoException(ICEBERG_MISSING_DATA, message, e);
        }
        throw new TrinoException(ICEBERG_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 7 with ReaderPageSource

use of io.trino.plugin.hive.ReaderPageSource in project trino by trinodb.

the class IcebergPageSourceProvider method createPageSource.

@Override
public ConnectorPageSource createPageSource(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorSplit connectorSplit, ConnectorTableHandle connectorTable, List<ColumnHandle> columns, DynamicFilter dynamicFilter) {
    IcebergSplit split = (IcebergSplit) connectorSplit;
    IcebergTableHandle table = (IcebergTableHandle) connectorTable;
    List<IcebergColumnHandle> icebergColumns = columns.stream().map(IcebergColumnHandle.class::cast).collect(toImmutableList());
    Map<Integer, Optional<String>> partitionKeys = split.getPartitionKeys();
    List<IcebergColumnHandle> regularColumns = columns.stream().map(IcebergColumnHandle.class::cast).filter(column -> !partitionKeys.containsKey(column.getId())).collect(toImmutableList());
    TupleDomain<IcebergColumnHandle> effectivePredicate = table.getUnenforcedPredicate().intersect(dynamicFilter.getCurrentPredicate().transformKeys(IcebergColumnHandle.class::cast)).simplify(ICEBERG_DOMAIN_COMPACTION_THRESHOLD);
    HdfsContext hdfsContext = new HdfsContext(session);
    ReaderPageSource dataPageSource = createDataPageSource(session, hdfsContext, new Path(split.getPath()), split.getStart(), split.getLength(), split.getFileSize(), split.getFileFormat(), regularColumns, effectivePredicate, table.getNameMappingJson().map(NameMappingParser::fromJson));
    Optional<ReaderProjectionsAdapter> projectionsAdapter = dataPageSource.getReaderColumns().map(readerColumns -> new ReaderProjectionsAdapter(regularColumns, readerColumns, column -> ((IcebergColumnHandle) column).getType(), IcebergPageSourceProvider::applyProjection));
    return new IcebergPageSource(icebergColumns, partitionKeys, dataPageSource.get(), projectionsAdapter);
}

Example 8 with ReaderPageSource

use of io.trino.plugin.hive.ReaderPageSource in project trino by trinodb.

the class OrcPageSourceFactory method createPageSource.

@Override
public Optional<ReaderPageSource> createPageSource(Configuration configuration, ConnectorSession session, Path path, long start, long length, long estimatedFileSize, Properties schema, List<HiveColumnHandle> columns, TupleDomain<HiveColumnHandle> effectivePredicate, Optional<AcidInfo> acidInfo, OptionalInt bucketNumber, boolean originalFile, AcidTransaction transaction) {
    if (!isDeserializerClass(schema, OrcSerde.class)) {
        return Optional.empty();
    }
    List<HiveColumnHandle> readerColumnHandles = columns;
    Optional<ReaderColumns> readerColumns = projectBaseColumns(columns);
    if (readerColumns.isPresent()) {
        readerColumnHandles = readerColumns.get().get().stream().map(HiveColumnHandle.class::cast).collect(toUnmodifiableList());
    }
    ConnectorPageSource orcPageSource = createOrcPageSource(hdfsEnvironment, session.getIdentity(), configuration, path, start, length, estimatedFileSize, readerColumnHandles, columns, isUseOrcColumnNames(session), isFullAcidTable(Maps.fromProperties(schema)), effectivePredicate, legacyTimeZone, orcReaderOptions.withMaxMergeDistance(getOrcMaxMergeDistance(session)).withMaxBufferSize(getOrcMaxBufferSize(session)).withStreamBufferSize(getOrcStreamBufferSize(session)).withTinyStripeThreshold(getOrcTinyStripeThreshold(session)).withMaxReadBlockSize(getOrcMaxReadBlockSize(session)).withLazyReadSmallRanges(getOrcLazyReadSmallRanges(session)).withNestedLazy(isOrcNestedLazy(session)).withBloomFiltersEnabled(isOrcBloomFiltersEnabled(session)), acidInfo, bucketNumber, originalFile, transaction, stats);
    return Optional.of(new ReaderPageSource(orcPageSource, readerColumns));
}

Example 9 with ReaderPageSource

use of io.trino.plugin.hive.ReaderPageSource in project trino by trinodb.

the class IcebergPageSourceProvider method createOrcPageSource.

private static ReaderPageSource createOrcPageSource(HdfsEnvironment hdfsEnvironment, ConnectorIdentity identity, Configuration configuration, Path path, long start, long length, long fileSize, List<IcebergColumnHandle> columns, TupleDomain<IcebergColumnHandle> effectivePredicate, OrcReaderOptions options, FileFormatDataSourceStats stats, TypeManager typeManager, Optional<NameMapping> nameMapping) {
    OrcDataSource orcDataSource = null;
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(identity, path, configuration);
        FSDataInputStream inputStream = hdfsEnvironment.doAs(identity, () -> fileSystem.open(path));
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), fileSize, options, inputStream, stats);
        OrcReader reader = OrcReader.createOrcReader(orcDataSource, options).orElseThrow(() -> new TrinoException(ICEBERG_BAD_DATA, "ORC file is zero length"));
        List<OrcColumn> fileColumns = reader.getRootColumn().getNestedColumns();
        if (nameMapping.isPresent() && !hasIds(reader.getRootColumn())) {
            fileColumns = fileColumns.stream().map(orcColumn -> setMissingFieldIds(orcColumn, nameMapping.get(), ImmutableList.of(orcColumn.getColumnName()))).collect(toImmutableList());
        }
        Map<Integer, OrcColumn> fileColumnsByIcebergId = mapIdsToOrcFileColumns(fileColumns);
        TupleDomainOrcPredicateBuilder predicateBuilder = TupleDomainOrcPredicate.builder().setBloomFiltersEnabled(options.isBloomFiltersEnabled());
        Map<IcebergColumnHandle, Domain> effectivePredicateDomains = effectivePredicate.getDomains().orElseThrow(() -> new IllegalArgumentException("Effective predicate is none"));
        Optional<ReaderColumns> columnProjections = projectColumns(columns);
        Map<Integer, List<List<Integer>>> projectionsByFieldId = columns.stream().collect(groupingBy(column -> column.getBaseColumnIdentity().getId(), mapping(IcebergColumnHandle::getPath, toUnmodifiableList())));
        List<IcebergColumnHandle> readColumns = columnProjections.map(readerColumns -> (List<IcebergColumnHandle>) readerColumns.get().stream().map(IcebergColumnHandle.class::cast).collect(toImmutableList())).orElse(columns);
        List<OrcColumn> fileReadColumns = new ArrayList<>(readColumns.size());
        List<Type> fileReadTypes = new ArrayList<>(readColumns.size());
        List<ProjectedLayout> projectedLayouts = new ArrayList<>(readColumns.size());
        List<ColumnAdaptation> columnAdaptations = new ArrayList<>(readColumns.size());
        for (IcebergColumnHandle column : readColumns) {
            verify(column.isBaseColumn(), "Column projections must be based from a root column");
            OrcColumn orcColumn = fileColumnsByIcebergId.get(column.getId());
            if (orcColumn != null) {
                Type readType = getOrcReadType(column.getType(), typeManager);
                if (column.getType() == UUID && !"UUID".equals(orcColumn.getAttributes().get(ICEBERG_BINARY_TYPE))) {
                    throw new TrinoException(ICEBERG_BAD_DATA, format("Expected ORC column for UUID data to be annotated with %s=UUID: %s", ICEBERG_BINARY_TYPE, orcColumn));
                }
                List<List<Integer>> fieldIdProjections = projectionsByFieldId.get(column.getId());
                ProjectedLayout projectedLayout = IcebergOrcProjectedLayout.createProjectedLayout(orcColumn, fieldIdProjections);
                int sourceIndex = fileReadColumns.size();
                columnAdaptations.add(ColumnAdaptation.sourceColumn(sourceIndex));
                fileReadColumns.add(orcColumn);
                fileReadTypes.add(readType);
                projectedLayouts.add(projectedLayout);
                for (Map.Entry<IcebergColumnHandle, Domain> domainEntry : effectivePredicateDomains.entrySet()) {
                    IcebergColumnHandle predicateColumn = domainEntry.getKey();
                    OrcColumn predicateOrcColumn = fileColumnsByIcebergId.get(predicateColumn.getId());
                    if (predicateOrcColumn != null && column.getColumnIdentity().equals(predicateColumn.getBaseColumnIdentity())) {
                        predicateBuilder.addColumn(predicateOrcColumn.getColumnId(), domainEntry.getValue());
                    }
                }
            } else {
                columnAdaptations.add(ColumnAdaptation.nullColumn(column.getType()));
            }
        }
        AggregatedMemoryContext memoryUsage = newSimpleAggregatedMemoryContext();
        OrcDataSourceId orcDataSourceId = orcDataSource.getId();
        OrcRecordReader recordReader = reader.createRecordReader(fileReadColumns, fileReadTypes, projectedLayouts, predicateBuilder.build(), start, length, UTC, memoryUsage, INITIAL_BATCH_SIZE, exception -> handleException(orcDataSourceId, exception), new IdBasedFieldMapperFactory(readColumns));
        return new ReaderPageSource(new OrcPageSource(recordReader, columnAdaptations, orcDataSource, Optional.empty(), Optional.empty(), memoryUsage, stats), columnProjections);
    } catch (Exception e) {
        if (orcDataSource != null) {
            try {
                orcDataSource.close();
            } catch (IOException ignored) {
            }
        }
        if (e instanceof TrinoException) {
            throw (TrinoException) e;
        }
        String message = format("Error opening Iceberg split %s (offset=%s, length=%s): %s", path, start, length, e.getMessage());
        if (e instanceof BlockMissingException) {
            throw new TrinoException(ICEBERG_MISSING_DATA, message, e);
        }
        throw new TrinoException(ICEBERG_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 10 with ReaderPageSource

use of io.trino.plugin.hive.ReaderPageSource in project trino by trinodb.

the class TestTimestampMicros method createPageSource.

private ConnectorPageSource createPageSource(ConnectorSession session, File parquetFile, String columnName, HiveType columnHiveType, Type columnType) {
    // TODO after https://github.com/trinodb/trino/pull/5283, replace the method with
    // return FileFormat.PRESTO_PARQUET.createFileFormatReader(session, HDFS_ENVIRONMENT, parquetFile, columnNames, columnTypes);
    HivePageSourceFactory pageSourceFactory = StandardFileFormats.TRINO_PARQUET.getHivePageSourceFactory(HDFS_ENVIRONMENT).orElseThrow();
    Properties schema = new Properties();
    schema.setProperty(SERIALIZATION_LIB, HiveStorageFormat.PARQUET.getSerde());
    ReaderPageSource pageSourceWithProjections = pageSourceFactory.createPageSource(new Configuration(false), session, new Path(parquetFile.toURI()), 0, parquetFile.length(), parquetFile.length(), schema, List.of(createBaseColumn(columnName, 0, columnHiveType, columnType, REGULAR, Optional.empty())), TupleDomain.all(), Optional.empty(), OptionalInt.empty(), false, AcidTransaction.NO_ACID_TRANSACTION).orElseThrow();
    pageSourceWithProjections.getReaderColumns().ifPresent(projections -> {
        throw new IllegalStateException("Unexpected projections: " + projections);
    });
    return pageSourceWithProjections.get();
}